@svgrid/mcp 3.0.44 → 3.0.46

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/data.js CHANGED
@@ -4604,7 +4604,7 @@ export const docs = [
4604
4604
  "path": "docs/changelog.md",
4605
4605
  "title": "Changelog",
4606
4606
  "section": "Overview",
4607
- "markdown": "# Changelog\n\nThe user-facing log of what shipped. Format follows\n[Keep a Changelog 1.1.0](https://keepachangelog.com/en/1.1.0/);\nversioning follows [Semantic Versioning](https://semver.org/).\n\nEvery release has up to six sections:\n- **Added** - new features\n- **Changed** - existing functionality, behaviour changes\n- **Deprecated** - soon-to-be-removed features\n- **Removed** - features removed in this release\n- **Fixed** - bug fixes\n- **Security** - vulnerability patches (with CVE ids when applicable)\n\nPre-release entries live under `## [Unreleased]`. They graduate to a\ndated heading on publish via `pnpm changeset version` (see\n[.changeset/README.md](../.changeset/README.md)).\n\nFor machine-readable releases, fetch\n[`/changelog.json`](/changelog.json) - same content, parseable shape.\n\n## [Unreleased]\n\n### @svgrid/grid\n\n#### Added\n\n- **`trustedHtml(html)`.** `sanitizeHtml`, ready for an HTML sink on a page that requires Trusted\n Types: there it is a `TrustedHTML` of the `svgrid` policy, elsewhere the same string.\n\n- **`toggleOnSecondClick` on a checkbox column.** The first click selects the cell, a click on the selected cell toggles it, and a finger toggles on the tap. Off by default.\n\n- **`parseClipboardText`, and more for the fill and paste hooks.** `parseClipboardText(text)` splits a\n paste into rows of cells, for a consumer whose copy quotes a cell. `processCellForFill` also gets\n `sources` and `index`, and `processCellFromClipboard` gets `cell`, the place in the pasted block.\n Left out, a grid fills and pastes as it did.\n\n- **`keepFocus`.** When the focused control of a region is removed or disabled, focus\n moves to the control now in its place, not to the body. `createFocusTrap` and\n `menuFocus` use it, so dialogs, drawers and menus have it. `use:keepFocus` gives it\n to any other region.\n- **`describedBy` on `SvModal`.** The id of what a dialog says, read with its title.\n\n- **`holdRowsOnWrite`.** A paste, a clear or a fill changes rows where they stand, as a\n typed edit does, for an app that orders the view itself when the person moves on.\n- **`onPasteOverflow`.** The lines of a paste past the last row, split into cells, inside\n the paste's undo step. Without it they are dropped, as before.\n- **`gridLabel`.** The accessible name of the grid, on its table element. SvGrid Base\n passes the table's name.\n- **Very long grids scroll true at both ends.** Past the height a browser paints\n cleanly the grid now scales its scroll, with a pixel staying a pixel near each end,\n so the rows past about 560,000 of a million draw as they should and Ctrl+End\n reaches the last row.\n- **`onColumnVisibilityChange` and `onColumnPinningChange`.** Hiding,\n showing, pinning or unpinning a column, from the column menu, the api\n setters or `api.setState`, now calls back with the hidden ids or the\n `{ left, right }` pinning, the same shapes `api.getState()` reports. Both\n are deduped on the value and do not fire on mount. An app saving the\n layout no longer has to poll `getState()` to notice either. The\n `<sv-grid>` element fires `columnvisibilitychange` and\n `columnpinningchange`.\n\n- **`SvGridProps` is exported.** The full prop type, for a wrapper that\n forwards a typed options bag to `<SvGrid>`. Without it a wrapper had to\n hand-write a `declare module \"@svgrid/grid\"` stub, which shadows the\n package's own types.\n\n- **`chartSpecToTable`.** The rows and columns a spec amounts to, in the\n shape `SvGrid` takes: one row per category with a column per series (a\n grouped axis adds a Group column, a histogram From and To, candles their\n four prices, a box plot its five numbers), per point for a scatter, per\n link for a sankey or chord, per leaf for a tree map or sunburst with a\n column per level, per day for a calendar, one row for a gauge. Number\n columns carry the chart's format (a right-axis series its axis's) and a\n date axis is a date column. The gallery demos' Chart | Grid switch is this\n behind a sortable grid.\n\n- **The Chart panel charts the pivot in pivot mode.** With `pivot` and\n `charting` on one grid, the pivot bar carries the Chart toggle and the\n panel draws the pivot on screen: row groups as categories (nested rows a\n grouped axis), column groups as series, the measures' format on the axis,\n totals left out. The data pickers step aside with a note and the Type\n select keeps the shapes a pivot can take; Stacked, 100%, Horizontal, the\n Format tab, the builder, saved charts, export and Describe still apply. A\n click on a bar filters the innermost row dimension when the grid has a\n column for it, which re-runs the pivot. It used to chart the source rows by\n the source columns whatever was on screen. The mapping is\n `pivotResultToChartSpec` (with `pivotChartType`, `pivotFilterColumn`)\n in the lazy chart engine; enterprise's `pivotToChartSpec` is the same\n function. `bucketsToChartSpec` (server aggregates as a spec) moved there\n from the controller, out of the base bundle.\n\n- **Scatter points select.** A click or Enter on a scatter or bubble point\n selects it the way a bar selects: the point's `label` (or its series) is\n the category, its y the value, and `selectable`, `bind:selected`,\n `onSelectionChange` and `onSelect` all apply. The dots had no click\n handler, so on a scatter chart those props did nothing.\n- **`ChartPointRef.index`.** A selected bar carries its category index\n beside its label, and two refs with an index match only on the same one.\n A grouped axis repeats its leaf labels (Q1 to Q4 under each year), so a\n click on the second Q2 used to select the first Q2 with it. A ref written\n without an index still matches by label.\n- **Data labels on a 100% chart read the share.** Under `stacked100` the\n axis is already a share whatever `valueFormat` says; the labels on the\n bars now say \"18%\" as well, rather than \"$18\", or \"1800%\" when the format\n was percent. A `dataLabels.formatter` still receives the raw value and the\n tooltip keeps it.\n- **A series on the right axis reads in that axis's format.** The tooltip,\n the screen-reader table, the live region and the data labels format a\n series with `axis: 'right'` through `y2Axis.format` or `formatter`, so a\n margin of 0.29 beside a revenue in dollars reads \"29%\" and not \"$0.29\".\n Without a format of its own the right axis is plain numbers, as its ticks\n are.\n- **The breadcrumb survives `toolbar={false}`.** A drilled sunburst, tree\n map or pie with the toolbar off still shows its breadcrumb in a strip of\n its own; it used to have no way back up.\n- **Chart axis model.** `xAxis`, `yAxis` and `y2Axis` on a spec take\n `min`, `max`, `nice`, `tickCount`, `tickInterval`, `format` or a\n `formatter`, `title`, `gridLines`, `labelRotation`, `reversed` and, for\n the x axis, `type: 'number'` for a numeric axis that spreads categories by\n value. `referenceBands` shade a range on either axis, `referenceLines` can\n be vertical, `title` / `subtitle` / `caption` frame the chart. Series\n gained `marker` shapes and per-point `markers` / `colors`, `strokeWidth`,\n `dash`, `opacity`, `gradient`, `step` lines, `connectNulls` and a\n `nullAs` policy; `dataLabels` takes a placement, a formatter and overlap\n hiding; the component takes a `tooltip` snippet, `tooltipFormat`,\n `tooltipMode: 'single'`, a `legend` side and a `legendItem` snippet, and\n `autosize`. `rowsToChartSpec` reduces with `min`, `max`, `median`,\n `first`, `last`, `countDistinct` and `pN` and buckets a date dimension by\n `day` / `week` / `month` / `quarter` / `year`.\n- **Decimation for long series.** `decimate: 'auto'` (the default) thins a\n line or area with more points than pixels through LTTB, which keeps the\n shape; `{ method: 'minmax' }` keeps every spike. Zoom is applied first, so\n zooming into 300 of 100,000 points shows all 300. `lttb`, `minMaxIndices`,\n `decimateSpec` and `pickCategories` are exported.\n- **Thirteen more chart types**: `histogram` (with `binValues` and\n `rowsToHistogramSpec`), `range-bar`, `range-area`, `lollipop`,\n `dumbbell`, `pareto` (`paretoSpec`), `stream` (`stackOffset: 'wiggle'` or\n `'silhouette'`), `sunburst`, `radial-bar`, `radial-column`,\n `nightingale`, `chord` and `bullet`; plus hollow and Heikin-Ashi candles\n (`candleStyle`) and pyramid / cone funnels (`funnelShape`). All of them are\n in the grid panel's picker, gated on what the current columns can feed.\n- **Zoom, pan and presets.** `zoomable` takes an object: `wheel` (`true` or\n `'modifier'` for Ctrl + wheel), `pinch`, `pan` (`'shift'`, `'mode'` or\n both) and `axis` (`'x'`, `'y'`, `'xy'`). The window is a bindable\n `zoom` prop of category indices with `zoomTo` / `resetZoom` on the\n component and an `onZoom` callback; `rangePresets` puts 1W / 1M / 3M / 6M /\n YTD / 1Y / All in the toolbar on a time axis.\n- **Synchronized charts.** Charts with the same `syncGroup` share their\n crosshair and zoom window, matched by label and then by date. The grid\n panel's `syncTabs` keeps one window across its chart tabs.\n- **Context menu, animation, drilldown, selection, keyboard.** `contextMenu`\n (built-in export / zoom / series items, your own `MenuItem`s, or a function\n of the clicked point; Shift + F10 opens it), `animate` (a data-update tween\n and `'fade'` / `'grow'` / `'wipe'` enter effects, off under reduced\n motion), `drillable` + `drillPath` for tree maps and sunbursts with a\n breadcrumb, `selectable` + `selected` + `onSelectionChange` for point\n selection with dimming, Up / Down stepping through series with a polite live\n region (`announce`).\n- **Financial toolkit.** `bollingerBands`, `rsi`, `macd`, `vwap`, `atr`,\n `stochastic`, `wma` and `obv`, NaN-padded like the moving averages;\n `'wma:N'`, `'bb:N:K'` (a shaded band) and `'vwap'` overlays;\n `indicatorPane` and `SvChartPanes`, which stacks indicator panes under a\n price chart with one x axis, one gutter and one crosshair; `rowsToOhlcSpec`\n and `resampleOhlc` (daily bars to weeks / months); `lastPriceLine`;\n annotation `shape` (`'flag'`, `'pin'`, `'square'`) with a tooltip\n `text`; `volumes` on a series.\n- **Drawing tools.** `drawable` puts trend line, horizontal ray, Fibonacci\n retracement, rectangle, arrow and note tools in the toolbar; the drawings\n are data points (`spec.drawings`, `onDrawingsChange`) so they survive a\n zoom, a resize or new data. A selected drawing has draggable handles and\n Delete removes it.\n- **PDF and print** with no PDF library: `chartToPdfBlob`, `downloadChartPdf`\n (a one-page A4 or Letter PDF with the chart as a JPEG and the titles as\n text) and `printChart`, in the toolbar, the context menu and the panel's\n export menu.\n- **Chart builder in the grid panel.** The Build button opens a modal with a\n type gallery (a live thumbnail per type), the data pickers and a Format tab\n (titles, legend and label placement, axis title / min / max / format / grid\n lines, per-series colour / type / axis / marker / width). The format is\n `ChartFormatState` on the chart tab, applied by `applyChartFormat` to every\n derived spec, saved with the view and reachable through\n `configureChart({ format })`. A link toggle unlinks a chart from the grid\n (`configureChart({ frozen: true })`) so it keeps its data while you filter.\n `charting.onChartCreated` / `onChartChanged` report the panel's lifecycle;\n `charting` also takes `zoom`, `rangePresets`, `syncGroup`, `syncTabs`,\n `contextMenu` and `animate`.\n- **Candlesticks from the grid panel.** Pick Candlestick or OHLC bars and the\n panel finds the Date, Open, High, Low, Close and Volume columns by name;\n Indicators chips add panes and overlays; `configureChart({ ohlc,\n indicators })` and the AI chart plan reach the same thing.\n- **A scatter series' `overlay` is a regression of y on x.** Until now the\n scatter layout ignored `overlay` (the fits ran on the index, which a\n scatter has no use for). The regressions (`linearFit`, `polynomialFit`,\n `exponentialFit`, `logarithmicFit`, `powerFit`) take an optional `xs`\n and return `predict(x)`; `regressionFit(values, overlay, xs)` names one by\n its overlay string; a scatter series with `overlay` gets the curve drawn\n across the plot (sampled evenly in log space on a log axis) and its\n equation and R-squared in the point tooltip.\n- **Chart gallery: fourteen full-size demos, one per family, and a docs page\n for them.** Bars, lines, areas, pie / donut / sunburst, scatter and bubble,\n combination, histogram and box plot, heat map and calendar, radar and\n radial, tree map and sunburst, sankey and chord, waterfall / funnel /\n pareto, gauges and bullets, and an annotated chart (demos 438 to 451),\n each with its own data and the switches that family needs, embedded on\n `help/charts/gallery` and the type pages.\n- **Grid chart panel: a deeper Format tab, one set of pickers, a\n localized panel and saved charts.** The builder's Format tab adds series\n labels, the crosshair's axis pills, a compact rule below a width\n (`compactBelow`, through `CHART_RESPONSIVE_PRESETS.compact`), a stack\n group per bar or area series and the chart's font, size, background and\n text colour (`ChartFormatState.seriesLabels` / `crosshairLabels` /\n `compactBelow` / `series[].stack` / `style`). The panel's pickers and the\n builder's Data tab are one component (`SvGridChartPickers`), so the two\n cannot disagree about what a type needs. Every string the panel and the\n builder show is a `chart*` key on `localization.text`\n (`GridChartPanelMessages`, loaded with the panel; `GridMessages` is now\n the chrome map and the panel map together). `api.saveChart(name)`,\n `applySavedChart`, `removeSavedChart`, `getSavedCharts` and\n `configureChart({ saved })` keep a tab's whole configuration under a name;\n the panel has a popover for them and `getState()` carries `savedCharts`\n when there are any.\n- **Chart diagnostics, a plain-language description, live feeds and a JSON\n schema.** `validateChartSpec` returns what is off about a spec (unknown\n keys with the nearest name, an unknown type, a series short of the\n categories, a log axis at zero, an overlay the engine does not read, an\n anchor naming no category, and more); `SvChart` logs the findings once per\n spec in development, `aiChart` attaches the errors to its plan, and the\n MCP server's `svgrid_check_code` applies the rules to a static spec\n literal with a rename for a misspelt key. `chartSummary` reads a chart in\n a sentence (trend, extremes, shares, correlation, last close) and\n `SvChart` hands it to assistive technology as `aria-description` and the\n table caption (`describe`), the context menu and the panel copy it, and\n `aiExplainChart` grounds an Explain button on it. `appendPoints` builds\n the next spec of a feed with every per-category array kept in step and a\n rolling window, and `live` on the chart skips the tween and enter effect.\n `docs/schemas/chart-spec.json` is generated from the types, served at\n svgrid.com/schemas and by `svgrid_get`, and a test keeps it equal to the\n validator's lists.\n- **Series and interaction depth.** Areas join `stack` groups like bars\n (`stacked100` and `stackOffset` run per pile); a numeric x axis takes\n `scale: 'log'` (even decades, 2 / 5 minors, non-positive categories\n dropped, decimation in log positions); regression overlays `poly:N`,\n `exp`, `log` and `power` next to `linear`, each carrying `r2` and\n `equation` on the overlay line and read in the tooltip, with\n `linearFit` / `polynomialFit` / `exponentialFit` / `logarithmicFit` /\n `powerFit` / `rSquared` / `computeOverlayFit` / `overlayName` exported\n and `charting.trend` accepting the new names; hovering a mark dims the\n other series (`hoverHighlight`, on by default, snapping within 18px) and\n `onHover` reports the point once per change; `tooltipPosition` parks the\n tooltip in a plot corner and `tooltipSticky` pins it on a click; data\n labels take `rotation` and `connector` (pushed clear of a collision with\n a leader line); `spec.style` sets a chart's font, size, background and\n text / grid colours over the tokens (`chartStyleVars`), honoured by the\n exports.\n- **Accessibility pass on the chart.** Every non-cartesian family (pie,\n sunburst and the radial charts, heat map, calendar, tree map, funnel,\n sankey nodes and links, radar, scatter) is one Tab stop with a roving\n focus: arrows walk the marks (a heat map by row and column, a calendar by\n day and week), Home / End / PageUp / PageDown jump, the focused mark\n raises its tooltip, Enter / Space select or drill, and every mark is named\n and has a focus ring. Zoom from the keyboard: `+` / `-` / `0` on a\n focused category and Shift + arrows to pan, and the brush is a `slider`\n (arrows pan and resize, Home / End, `0`). The toolbar is a `toolbar`, the\n tooltip a `tooltip`, decorative icons are hidden from the tree. Forced\n colors (Windows High Contrast) keep the data colours and repaint the\n chrome with the system palette. `localeText` on `SvChart` (and\n `charting.localeText` for the grid panel) translates every string the\n chart renders itself, with `ChartMessages`, `defaultChartMessages`,\n `resolveChartMessages` and `chartMessage` exported. A new\n `tests/e2e/chart-a11y.spec.ts` runs axe-core against the rendered chart\n on three demos and drives the keyboard model in a real browser.\n- **Stack groups, series labels, responsive rules, pie callouts.** `stack`\n on a bar series names the pile it joins, so plan and actual stack side by\n side in each category and the axis fits the tallest stack; `seriesLabels`\n names each line at its last point in a reserved gutter, pushed apart where\n lines end together; `responsive` patches the spec by the rendered size\n (axes merge one level deep, a rule's `legend` moves or hides the\n component's legend; `resolveResponsive` and `matchResponsiveRules` are\n exported); `dataLabels: { placement: 'outside' }` on a pie draws callout\n labels with leaders instead of percentages, and a spec can carry its own\n `dataLabels`; `visible: false` starts a series hidden with its legend chip\n dimmed; the crosshair reads off the axes with a category pill and a value\n pill (`crosshairLabels={false}` turns them off).\n- **`moveCells` - drag a selected range to move or copy it.** Grab a selected\n range by its border and drop it somewhere else. Move by default, copy with\n Ctrl / Cmd, and the modifier is read at DROP time so it can be pressed or\n released mid-drag. On whenever cell selection is on; `moveCells={false}`\n opts out. A drop is refused outright - nothing changes - when it would land\n off-grid or when any source or destination cell is read-only, rather than\n moving the part that fits and silently dropping the rest.\n- **`selectionBar` - a floating bar over the grid while rows are selected**,\n showing the count and the actions that apply to the whole selection. `true`\n gives the count and a Clear button; an array is shorthand for `{ actions }`;\n the object form also sets `position` (`'bottom'` default, or `'top'`),\n `maxVisible` and `hideClear`. The prop and its types are free; the bar\n itself is an Enterprise renderer, and without it the grid shows a short upsell\n in its place, the same way the scheduler and board views do.\n- **Edge auto-scroll for every drag that extends a rectangle** - the fill\n handle, the range move, and plain drag-select. All three were previously\n capped at whatever was already on screen when the drag began, which on a grid\n built for 100k rows meant the gesture mostly did not work. One shared rAF loop\n scrolls the container when the pointer is inside a 40px edge band, on a linear\n ramp to 20px/frame, in both axes. It keeps scrolling with the pointer parked\n still, which is the whole point of holding at an edge.\n- **Row and column resizing, both opt-in** (`rowResize`, `columnResize`).\n `columnResize` gives every header a drag handle: drag the edge, double-click\n it to size the column to its content, or focus it and use Left/Right (Shift\n for a 1px step). `rowResize` does the same vertically and brings the\n row-header column with it, since that is where the grip belongs; the grid\n stores the heights itself, so neither needs a function-valued `rowHeight`.\n Both load their code on demand, so a grid that leaves them off downloads\n neither.\n- **`ColumnDef.resizable`** - per-column opt-out, consulted only when the grid\n has `columnResize` on. `resizable: false` removes that column's handle\n entirely, so the drag, the arrow keys and double-click-to-autosize all go with\n it, and the column menu drops its Autosize item. For the columns whose width\n is part of the layout: a row-number gutter, a checkbox column, a fixed icon\n column. Programmatic sizing (`api.setColumnWidth`, `api.autosizeColumn`,\n `fitColumns`) still applies.\n- **`SvGridViewState` is exported.** It is the return type of the public\n `api.getState()`, so writing a typed \"named views\" feature previously meant\n either re-declaring the shape or reaching into an internal path.\n- **Form layer depth in `SvForm` / `createForm`.** Schema-driven forms gained:\n conditional / dependent fields (`visible` and `disabled` accept a\n `(values) => boolean` predicate; hidden fields drop out of validation and the\n submit payload), grouped **sections** and a multi-step **wizard** (`stepper`,\n with per-step validation gating), repeatable **field arrays**\n (`type: 'array'` with `itemFields`, plus `addItem` / `removeItem` / `moveItem`),\n **async validation** (debounced, stale-response-guarded, per-field\n `validating` state), **dirty tracking + `reset()`**, and **submit states**\n (`isSubmitting`, an async `onSubmit`, and `setErrors()` for server-side field\n errors). New demos: dynamic form, wizard, and field array.\n- **Public headless cores for the menu / select family** (`createMenu`,\n `createPopoverSelect`, shared `list-nav`), so the popup pickers can be driven\n headlessly and composed into custom UI.\n- **`color` on `ListOption`** - render a small color swatch before the label in\n `SvListBox` and `SvDropDownList`.\n- **`block` on `SvDateTimePicker`** - stretch the field to fill its container\n (100% width), so it fills a grid cell when used as an editor.\n- **`onClosed` on `SvDrawer`** - fires after the exit animation ends and the\n drawer leaves the DOM.\n- **`loading` on `SvField`**, and **`locked`** on `SvDockLayout` /\n `SvDockManager`.\n- **`summary` on `<SvGrid>`** - shortcut alias for `enableRowSummaries`,\n which turns on the footer row that aggregates every filtered row. It wins\n when both are set, the same precedence `selectable` has over\n `enableCellSelection`.\n- **`summary` on a column** - choose that column's footer aggregate\n (`'sum'`, `'avg'`, `'min'`, `'max'`, `'count'`, `'countDistinct'`,\n `'extent'`, `'first'`, or a custom function), or `false` to leave the cell\n blank. Without it a column keeps the old default: sum a numeric column,\n `Count: N` otherwise.\n- **Development-time configuration checks.** The grid used to fail silently on\n the most common mistakes - a misspelled `field` rendered a column of blank\n cells and printed nothing. It now warns once per problem, in dev builds only:\n\n - a column `field`, `groupBy` entry, `treeData.parentField` / `idField` or\n `treeData.column` that does not exist (with a \"did you mean\" for near misses),\n - two columns resolving to the same id,\n - `pageSize` set when pagination was never turned on,\n - a column marked `sortable` while sorting is not enabled,\n - `initialColumnPinning` while column virtualization is on, which silently\n prevents the columns from sticking,\n - `externalPagination` without `rowCount`, `externalSort` without\n `onSortingChange`, and `externalFilter` without `onFiltersChange`.\n\n The checks live in a lazy chunk behind a dev-only branch, so a production\n build never loads them and the base bundle is unchanged.\n\n#### Changed\n\n- **`showColumnFilters` no longer puts an input in the header.** Set to\n `true`, it also drew a plain \"Filter\" text box in every header cell,\n which with `showFilterRow` made two filter inputs per column. It now\n only turns on the column menu's filter section, as documented. For\n always-visible inputs use the filter row: `showFilterRow` or\n `filterMode=\"row\"`.\n- **The find bar loads on the first Ctrl+F.** It is its own chunk now, so a\n grid nobody searches does not carry it; the first open shows the bar once\n the chunk has loaded.\n- **`alignedGridGroup` loads its module on demand.** The shared group that\n mirrors horizontal scroll and column widths between grids is now a lazy\n chunk, fetched by the first grid that sets the prop, the way managed row\n drag is. A grid outside a group no longer carries it, which kept the base\n bundle under its 100 KB budget after this release's scroll and mount\n changes. Mirroring starts once the chunk has loaded, a frame on a warm\n cache.\n- **BREAKING: column resizing is now off by default.** Every column used to have\n a drag handle unconditionally - there was no prop, no per-column option and no\n way to turn it off, so a `width` you set was only ever a starting value a user\n could drag away. Set `columnResize` to get the handles back; the 241 demos in\n this repo that want them now say so. Sorting and filtering have always been\n opt-in behind `sortable` / `filterable`, and resizing was the odd one out.\n- **BREAKING: the footer summary row is now off by default.**\n `enableRowSummaries` used to default to `true`, so a plain\n `<SvGrid {data} {columns} />` grew an aggregate footer nobody asked for and\n every caller that did not want one had to opt out - 375 call sites in this\n repo passed `enableRowSummaries={false}` against 6 that opted in, and the\n API reference already documented the default as `false`. If you were\n relying on the old behaviour, add `summary` (or `enableRowSummaries`).\n- **Ember is now the default theme preset** (`defaultThemePreset` in\n `@svgrid/grid/themes`). It is what the demo gallery and svgrid.com open on, so\n a scaffolded or Studio-generated app now looks like the demos unless you pick\n a preset. Previously the default was Tailwind.\n- **Popup editors keep themselves on screen.** `SvAutoComplete`, `SvComboBox`,\n `SvDropDownList` and the date pickers now measure the space to the viewport\n edge: the panel flips up when there is not enough room below and clamps its\n `max-height` to the available height, so a picker opened near the bottom of the\n window no longer spills off screen.\n- **Dock manager pop-out to a new window is now off by default**\n (`allowPopout={false}`); opt in explicitly. Added a dedicated pop-out demo.\n\n#### Fixed\n\n- **`fitColumns` with `responsive` fills the row when a `hideBelow` column is hidden.** The fit still\n counted the hidden column's width, so the shown columns stopped short and left a blank strip that\n wide at the right edge.\n- **Cut clears only what reached the clipboard.** Ctrl+X, the Cut menu item and the cut command\n cleared the cells at once, before the clipboard write had finished, so a write the browser refused\n (a page without focus, a denied permission) lost the values. The clear now waits for the write, and\n runs only if the text landed and the selection has not moved meanwhile.\n `copySelectionToClipboard()` returns a promise of whether the text landed.\n- **The pinned-column divider in a contrast theme.** Forced colors keep a fill's alpha, so the\n `pinSplitter` divider stayed a faint line and vanished on `Canvas`. It is drawn in `CanvasText`,\n and in `Highlight` under the pointer, with focus and while dragged.\n- **Chart panel: an Export item that is off looks off.** The PDF item that cannot draw the chart's\n text said why but looked like the items that work; it is dimmed as an off CSV item is.\n- **Scheduler model: a day the clocks change is drawn by the clock.** `layoutDayEvents` placed an\n event by the real minutes since midnight, so on a 25-hour day every event sat an hour below the\n row its time is labelled with, and on a 23-hour day an hour above; an event in the last hour of a\n band fell out of it. It now counts wall-clock minutes (`wallMinutes`, `atWallMinute`, both new\n exports). On a week that starts on a 23-hour Sunday, `monthWeekSegments` put Monday's bars under\n Sunday. `timelineAxis` for `timelineDay` ends the band at its hour by the clock and names each\n real hour by its clock (a 25-hour day shows 1 AM twice), where it left a gap or an overlap.\n- **A menu item that is off keeps the arrow keys.** `menuFocus`, `createMenu` (`SvMenu`,\n `SvContextMenu`, `SvMenuList`) and the column and context menus mark it `aria-disabled` and keep\n it in the arrow-key and first-letter order, as the ARIA menu pattern does, so a screen reader\n hears it and why it is off. Its click does nothing. Before, the arrows skipped it.\n- **Forced colors: the header checkbox of a part-selected grid keeps its dash.** The dash was a\n fill, which a contrast theme removes, so the box read as empty.\n- **A chart's PDF refusal is seen at the toolbar and in the grid's chart panel.** For a title the\n PDF's font cannot draw, the toolbar's PDF button did nothing on screen: the reason went to the\n screen reader only. It is now marked off (`aria-disabled`, the reason as its tooltip) once it is\n pointed at or focused, and a press shows the reason in the button. The chart panel's Export menu\n wrote no file and said nothing; its PDF item now reads the reason and is off.\n- **The chart panel's Export menu takes the keyboard.** Arrow keys, Home, End and first letters move\n in it, focus goes to its first item when it opens and back to Export when it closes (it fell to the\n page). Its CSV item, when the chart has no table of values, stays focusable and says why.\n- **Rich text renders under Trusted Types.** `SvRichCell`, `SvRichText` and the sanitizer passed plain\n strings to `DOMParser`, `{@html}` and `innerHTML`, so a page with `require-trusted-types-for 'script'`\n threw and drew nothing. They now go through one policy, `svgrid`, made only where the browser has\n `trustedTypes`. A CSP with a policy list needs `trusted-types svgrid svelte-trusted-html`\n ([SECURITY.md](https://github.com/sv-grid/sv-grid/blob/main/SECURITY.md#trusted-types)).\n- **Chart Print runs under Trusted Types.** `printChart` built its window with `document.write`; it\n now builds it node by node and draws the chart as a `data:` image.\n- **A chart PDF never writes `?` for a letter.** Its text is in the PDF's built-in Helvetica, which\n has Western European letters only. The euro sign, curly quotes and dashes now come out right. A\n title, subtitle or caption with other letters (Cyrillic, Greek, CJK, Hebrew, emoji) is refused: the\n menu's PDF item says which text and points to PNG or SVG, and `chartToPdfBlob` and `buildChartPdf`\n throw a `ChartPdfTextError` naming it. `pdfUndrawable()` tells beforehand.\n\n- **A block cache that refreshes a block still being read frees that read's slot.** `refresh()`,\n `applyRows` and an insert or remove gave up a read in flight without counting it as done, so\n `loading` stayed true for good, and with `maxConcurrentRequests` reached no block loaded again.\n\n- **The \"Could not load these rows.\" line of a failed run sits on the first row wholly on screen.**\n It sat on the first rendered row, which overscan puts above the screen, so a person scrolled\n into a failed block saw a tinted band and no words.\n\n- **An edit deep in a server row model no longer walks the rows.** `find`, `findIndex`, `some` and `indexOf` on windowed\n data read only what is loaded, a paste, clear or fill over rows held in place copies only those, and `cellEdited`\n hands over `rowId`.\n\n- **A filter's count is said once.** Opening or closing a group under a filter no longer repeats\n it, and a closed group counts the records the filter keeps: it read \"28 of 30\" with 24 shown.\n- **`createFocusTrap` and `menuFocus` never drop focus to the body.** An opener that left the page,\n or turned disabled or hidden, hands focus to what held it. A dialog opened from a menu item goes\n back to what the menu went back to. A tooltip closes with Escape.\n\n- **CSV: a cell that starts with a line feed is guarded.** Tab and carriage return were,\n and a spreadsheet reads `\\n=1+1` as a formula too.\n- **A commit keeps the keyboard in the grid.** A pick in a select or dropdown cell, or\n Escape on a multi select, left focus on the body and arrows dead. Tab in a dropdown\n cell left the grid; it now commits and moves on, as the docs say.\n- **Group rows follow their state.** `aria-expanded`, the chevron and the button's\n label went stale until the row was drawn again.\n- **A grouped grid behaves as a treegrid.** Left and Right close and open a group, Enter\n and Space toggle it, and rows of records carry `aria-level`. Enter on a group's row\n used to move down.\n- **The filter count counts records.** \"28 of 34 rows match\" included group headers.\n- **A menu scrolls to the item the keyboard moved to.**\n- **A first click on the cell seeded on mount selects it.** It used to open the editor.\n- **Moving a column by hand is an undo step**, as a resize is. The api's setter records none.\n- **\"Add condition\" in the filter menu keeps focus**, and closing that menu goes back\n to the funnel.\n\n- **Keyboard and focus.** The Menu key and Shift+F10 open the context menu of the\n active cell. The column and context menus take focus, walk with the arrows, Home,\n End and first letters, and give focus back. `menuFocus` gives a menu of your own\n the same. The find bar takes focus when opened from a focused control.\n- **Focus stays in view.** Moving the active cell, or tabbing to a header control,\n stops clear of pinned columns on either side and above a sticky footer.\n- **Right to left.** The left and right arrows go where they point, horizontal\n scrolling to the active cell works, and the pin splitter follows the pinned columns.\n- **A grouped grid is a `treegrid`**, and a group row's cell is the active descendant.\n- **Menus, filter panel and find bar translate** through `localization.text`, as do\n the row number, select row, column menu, filter and resize names. `setModalCloseLabel`\n names the x on every `SvModal`.\n- **Ember**'s muted text reaches 4.5:1 on a pinned header. The column menu button is\n 24 px.\n- **IME.** A key of a composition no longer moves the cell or commits the editor.\n- **Smaller base bundle.** What a fill or a range move writes when the drag ends loads\n with the drag (`range-commit.ts`).\n- **`SvRichText` sanitizes its value.** It wrote `value` to `innerHTML`, so a stored string ran\n its handlers. `value` now goes through `sanitizeHtml` on the way in and out, and `sanitizeHtml`\n keeps a text alignment, rebuilt from scratch, so the alignment buttons survive it.\n- **CSV injection through a list.** `escapeFormulas` guarded strings only, so a multi-select value\n `=1+1` written with `rawValues: true` reached the file as a formula. It now guards everything\n but numbers.\n- **Open groups stay open.** In a grid with no `expanded` prop, a change to any other\n prop, such as new `data`, folded every group the person had opened.\n- **A row model that groups on the server is not grouped again in the page.**\n With `rowModel` from `createServerRowModel` and a grouping in the grid's\n state (a view, `groupBy`), the grid bucketed the rows it had loaded under\n its own group rows. A model with `group` accessors now hands over its rows\n as they are; the grouping state stays for whatever shows it.\n\n- **A `rowModel` set after mount still sorts and filters on the server.**\n The grid read `externalSort` and `externalFilter` once, at mount. A model\n handed over later (one loaded on demand) was sorted and filtered again in\n the page, placeholders included: with a million rows behind it, a view\n filter held the page for seconds. The grid now follows the model's\n `setSort` and `setFilter` whenever it arrives.\n\n- **`SvRichText` lists had no bullets or numbers in apps with a CSS reset.**\n Tailwind's preflight sets `list-style: none` on every list, so a bulleted\n list in the editor read as plain lines. The editor now sets `disc` and\n `decimal` itself.\n\n- **A custom column header did not sort when its label was clicked, and did\n not see the sort change.** The header wrapper is itself `role=\"button\"`, so\n the check meant to skip clicks on buttons inside a custom header matched the\n wrapper and skipped every click on the header's own text; only a click on\n empty header background sorted. The `header:` callback also ran once, so\n `ctx.column.getIsSorted()` kept its first value. Both are fixed: a click on\n the label sorts, a button inside the header still gets its own click, and\n the callback runs again when the column's sort changes.\n- **Grouping five million rows ran out of memory.** Every group row carried\n ten closures and a Map of its ancestors' values, and a million rows grouped\n five deep make about a million group rows. Group rows now share their\n methods and keep their state in one object. Grouping 1,000,000 rows that\n way adds about 500 MB instead of 785 MB, and 5,000,000 rows group in 56 s\n on a laptop within Chrome's default heap.\n- **Every selection change serialized the whole selection.** Deciding\n whether to call `onRowSelectionChange` ran `JSON.stringify` over the\n selection state; selecting half of a 5,000,000-row grid built a string of\n tens of megabytes per change. The two states are now compared key by key.\n- **A mount built rows nobody could see, and laid the grid out five\n times.** The first paint now draws the rows on screen and at most two\n overscan rows, and the rest of `overscan` follows a frame after that paint:\n a 520 px grid built 26 rows to show 14. Reading the scroll position of the\n container the grid had just created, and each scrollbar measuring its\n track on connect and on every attribute change, forced four layouts of the\n half-built grid before it painted. The position is now confirmed on the\n next frame and the track is measured by the scrollbar's resize observer;\n the one forced layout left is the container size the windows need.\n- **The scrollbar thumb trailed the rows by a frame.** The custom scrollbar\n read the scroll position on the frame after the scroll flush, so the thumb\n moved one frame late and each scroll step paid a second style pass and\n paint for it. The flush now hands the scrollbar the position it already\n has, and the scrollbar keeps its measured track instead of measuring it\n inside the flush.\n- **A vertical scroll frame repainted every rendered row.** Body rows were\n not positioned, so on any frame that moved or rewrote a row Chrome painted\n every rendered row again cell by cell. Flex-layout body rows are now\n `position: relative` (no z-index, no clip), which lets the browser reuse an\n unchanged row's paint whole. In six interleaved rounds of the 100,000-row\n browser bench, a 60 px scroll step went from 4.05 to 3.68 ms of\n main-thread work (median). Sticky frozen and group rows keep\n `position: sticky`.\n- **Expanding one group re-grouped every row.** A toggle re-ran the row\n pipeline, and the grouping stage bucketed all rows again although only the\n expansion had changed. It now keeps its groups while its input rows,\n grouping and columns are the same. On a 100,000-row grid grouped by five\n columns, one expand went from 521 to 98 ms, and expanding or collapsing\n every group from 473 to 82 ms.\n- **A calendar date with `timeZone: 'UTC'` showed the day before.** A\n bare date string such as `'2026-10-07'` is read as local midnight, so a\n date `format` whose `options` carried `timeZone: 'UTC'` (the usual fix\n from when such strings were read as UTC) showed October 6th anywhere east\n of UTC. A calendar date has no zone: `timeZone` is now ignored for one,\n on screen and in exports.\n- **A cell note's corner marker sat at the grid's corner.** The marker is\n positioned against its cell, but a cell with a note was not positioned, so\n it anchored to the grid. A cell with a note is now a positioning context,\n like the active cell.\n- **A sort or filter kept every column's value on every row.** Reading one\n column built and cached an array of all of a row's values, so sorting a\n 1,000,000-row grid by one column left 1,000,000 of them behind. A plain\n `field` column is now read from the data; a computed (`fieldFn`) column\n still builds the array. The heap a sorted and filtered 1,000,000 x 9 grid\n adds went from 196 MB to 128 MB in the browser harness.\n- **`applyTransaction({ remove })` scanned the data once per row object.**\n Removing 10,000 row objects from 1,000,000 rows took about 1.8 s to find\n them; past eight objects the transaction now builds one identity map.\n- **Group footer and grand-total rows drew checkboxes.** With\n `showRowSelection`, each totals row had a row-selection checkbox, and\n ticking it selected the synthetic row. A checkbox column with no `aggregate`\n showed a checkbox too, holding a leaf's value carried over from the group\n row. Totals rows keep the selection cell for alignment but leave it empty,\n and a boolean column without an aggregate renders nothing there.\n- **Toggling a checkbox cell did not fire `onCellValueChange`.** A click or\n Space on a checkbox cell wrote the row and stopped, so an app that saves\n edits from `onCellValueChange` lost every checkbox change, and Ctrl+Z could\n not take one back. The toggle now takes the inline editor's commit path:\n one undo step, then `onCellValueChange` with the old and new value and the\n row's position in `data`.\n\n- **Arrow keys lagged whenever the active cell had to scroll.** On the\n 10,000 x 53 large-dataset demo one ArrowDown at the bottom of the viewport\n cost about 37 ms of main-thread work and painted over two frames; it is now\n about 8 ms and paints once. Five things added up. Every rendered row was\n measured with `getBoundingClientRect` on each scroll tick, on a grid that\n had no `autoRowHeight` to use the number for, which forced a table layout\n per row (about 30 per key press). The scroll geometry the custom scrollbar\n reads was pulled inside a derived at the start of the flush, forcing a\n second full table layout that the row writes then invalidated. The\n virtualizer keyed rows by their position in the window, so a one-row\n scroll rewrote all ~30 rows x every rendered column; rows are now keyed\n `index mod windowLength`, so the rows that stay keep their `<tr>` untouched\n and the one that entered is moved with a single `insertBefore`. The row\n window only updated after the browser's `scroll` event and a\n `requestAnimationFrame`, a frame after the key press; keyboard navigation\n now syncs the virtualizer inline, since it knows where it scrolled. And\n each cell carried three selection-dependent deriveds plus the fill handle's\n `{#if}`, re-marked on every move; they are one `{@const}` now, `null` for\n a plain cell. Also closed on the way: `key in editedCellValues` on a deep\n `$state` proxy created a signal per rendered cell that never went away, and\n the header checkbox's tri-state walked every row on each arrow key.\n\n- **A heat map drew a series legend in colours its cells never use.** The\n ramp on the right is its key; the row of palette chips under it (and a\n calendar's) is gone.\n\n- **A time axis put its ticks on nothing in particular.** `xType: 'time'`\n stepped a fixed 30 or 365 days from the Unix epoch, so a tick labelled\n \"2025\" stood weeks from New Year, a \"Jun 25\" tick sat mid-June, and a\n thirty-month axis carried two labels on a plot with room for eight. Ticks\n now land on calendar boundaries in UTC (days, Mondays, the first of a\n month, quarter or year, or every few years) at about one label per 90px,\n and each label is sized to the gap between ticks rather than the axis span,\n so a quarterly axis no longer reads \"2024, 2024, 2024, 2025\". The labels\n are read in UTC too: a local-time read of a 'YYYY-MM-DD' category in the\n Americas labelled every tick with the day before.\n- **The tooltip echoed the ISO date.** On a time or ordinal-time axis the\n tooltip title, the crosshair pill and the live region wrote \"2025-06-01\";\n they write \"Jun 2025\" when every category is the first of a month, \"2025\"\n when every one is New Year, and \"Jun 1, 2025\" otherwise, through\n `xAxis.formatter` when one is set. A calendar cell names its day the same\n way. The category in `onHover`, `onSelect` and `selected` is unchanged.\n- **The screen-reader table and the CSV read \"NaN\".** A NaN gap (the way a\n forecast series marks the months before it starts) wrote the word \"NaN\"\n into the hidden data table and the CSV export, and a `null` fell back to\n 0, which told a screen reader the outage month had no sign-ups. Both are\n empty cells now.\n- **\"Double-click again to clear\" left a legend chip isolated and hidden.**\n A double-click arrives as click, click, dblclick: with a chip isolated the\n first click cleared the isolation, the second hid the chip and the dblclick\n isolated it again. A click that clears an isolation is remembered for a\n moment, and the double-click that follows it on the same chip clears; on\n another chip it moves the isolation there.\n- **A radar ignored `yAxis.min` / `max`.** The rim was always the data\n maximum, so scores out of 100 filled the dial to whatever the best score\n was and two radars could not be compared. The rim honours a pinned axis,\n a value past it sits on it, and a pinned minimum is the centre.\n- **A pie sized its gutters by the widest label on either side.** Small\n slices with long names on the left kept the same room on the right for a\n \"Chrome 62%\", and the pie drew at half the radius the pane had. The\n gutters are per side, the pie sits centred in what they leave, the floor\n radius is a quarter of the short side, and each side truncates to its own\n room.\n- **A waterfall tooltip read \"$0\" on a total.** A total bar holds 0 in the\n data and its running sum in the geometry; the tooltip and the\n screen-reader table now read the drawn value. Its x labels also tilt by\n the same fit rule as a bar chart's, with the angle the renderer needs:\n they were flagged as rotated with no angle and drew truncated instead.\n- **Histogram edges read \"1.3k, 1.3k\".** Bin edges are written at the\n precision their width needs (whole numbers with separators for a 44ms\n bin, two decimals for a 0.2 bin) rather than in the compact tick form that\n gave two neighbouring edges one label.\n- **A heat map ignored `dataLabels` in the spec.** The cell label read the\n component prop alone, so `dataLabels: { show: true }` in the spec drew\n nothing. Colour legends on heat maps and calendars also round their steps\n to the precision the range earns (\"20, 94, 167\", not \"20, 93.5, 167\").\n- **The range area read as three lines.** Its band filled at the wash an\n area under a line gets, with both edges at full stroke. The band is the\n mark: it fills at 0.35 and its edges are thin.\n- **Annotate mode pinned a tooltip instead of a note** when `tooltipSticky`\n was on too; the annotate click now owns the click.\n- **Drawings on a time axis warned on every point.** The drawing tools write\n a timestamp for every pointer position, and the spec validator matched\n those against the category list. It skips the check on a time or number\n axis, as it already did for reference lines and annotations.\n- **The panel's export menu and saved-charts dialog ignored Escape.** Both\n close on Escape and hand focus back to the button that opened them.\n- **A gauge wrote \"99.2 %\".** A symbol unit (%, degrees) hugs the number; a\n word unit (\"ms\") keeps its space.\n- **The right axis title was clipped.** Its glyphs point outward under the\n rotation and ran past the edge of the SVG.\n- **Pie callout labels ran off the chart in a narrow pane.** The pie kept a\n fixed 34px margin for its outside labels whatever their length, so a\n \"Samsung Internet 3%\" in a 380px pane was clipped to \"%\" and \"Cl\". The\n radius now makes room for the widest label on each side, keeps at least a\n fifth of the short side, and hands the renderer a per-side character\n budget to truncate to when even that is not enough.\n- **The brush strip inherited the chart's titles, axis labels and marks.**\n A title and subtitle took 45 of the strip's 88 pixels and the x labels\n another 30, leaving the mini-map a five-pixel smear under an empty band.\n The brush spec now mutes titles, axis titles and labels, reference lines\n and bands, annotations, drawings, series labels and data labels.\n- **A category chart tilted its x labels by count, not by fit.** More than\n eight categories, or any label over nine characters, tilted every label\n to -40 degrees even on a thousand-pixel chart with room to spare. `auto`\n now tilts only when the widest label does not fit its slot.\n- **A heat map wrote every column label whatever the cell width**, so\n twenty-four hours at 33px a cell read \"00:0001:0002:00\". The column\n labels thin out the way a bar chart's do.\n- **A calendar labelled the padding days of the next year.** The grid pads\n the range to a full week, and a December ending on a Thursday picked up a\n \"Jan\" tick for the two padding days. Month labels stay inside the range.\n- **A waterfall could not open on a value.** A total bar always read the\n running sum, so \"Revenue 4300\" marked as a total drew as zero and every\n bar after it hung below the axis. A total with a value now sets the sum;\n a total with 0 still reads it.\n- **`rowsToChartSpec` with `topN` duplicated a category called \"Other\".**\n A real \"Other\" in the data came out beside the folded one, two categories\n with one name, which the renderer threw on as a duplicate key. The tail\n now folds into that category.\n- **A pin annotation showed \"O…\" and nothing else.** The head held\n `truncate(label, 2)`, an initial plus an ellipsis, and no label was written\n beside it. The head carries the initial and the label sits beside the pin\n like a dot's does.\n- **A chart's hidden data table took up to 17,000px of scrollable space.**\n The screen-reader table (one row per category, capped at 1,000) carried the\n visually-hidden recipe itself, but on a `<table>` `height: 1px` and\n `width: 1px` are minimums and `overflow: hidden` does not clip, so the\n table was invisible and as tall as its rows, and every scroll container\n around a chart (the demo pages first of all) scrolled on through that much\n empty space. The clip now sits on a wrapping div; the table keeps its id,\n which the svg still points at.\n- **Two ARIA faults the axe audit found.** The chart's SVG was\n `role=\"img\"`, whose children are presentational to assistive technology,\n so the focusable marks inside it (category zones, slices, a drawing) were\n a nested-interactive violation and could be skipped by a screen reader; an\n interactive chart is a labelled `group` now (a thumbnail with\n `interactive={false}` stays an image). Sankey links carried an\n `aria-label` with no role, which ARIA prohibits; they are named images in\n the focus order.\n- **A tree map ignored `spec.tree`.** The field is documented as an alias of\n `treemap` and the sunburst honoured it; the tree map read only `treemap`.\n- **An autosized chart with a side legend grew without end in a plain block.**\n `legend=\"left\"` / `\"right\"` lays the chart out as a three-row grid, and a\n grid puts a gap between every track whether or not the toolbar and brush\n rows have anything in them. The autosize maths counted a gap only for the\n chrome that existed, so in a parent that does not pin the height, the host\n measured 8 or 16px taller than the plot it was told to fit, the plot grew\n into it, and the next measurement grew again - a few hundred pixels a\n second until the page ran out. Hiding the legend at runtime had the\n opposite fault: a dimension binding keeps its last value once its element\n is gone, so the vanished legend went on being subtracted and the plot\n ratcheted down to its 120px floor. Both fixed; a fixed-height parent hid\n them, which is why the docs' sizing example never showed either.\n- **`block` did nothing on eight editors.** The prop is declared once in\n `SvEditorProps`, and its own doc comment describes the symptom: \"in a form\n grid a row of inputs each stopping at a different width reads as broken\".\n `SvDropDownList`, `SvComboBox`, `SvMultiSelect`, `SvTreeSelect`,\n `SvGridSelect`, `SvAutoComplete`, `SvTextArea` and `SvDateRangeInput`\n accepted it by type and ignored it, so they stopped at their own 200-280px\n default while the text inputs beside them filled the row. `<SvForm>` was\n also not passing it to seven of the controls it mounts - both halves had to be\n fixed for a form row to line up.\n- **`cellDataType: 'dateString'` stored a timestamp after an edit.**\n Committing ran through the shared date coercion (`new Date(v).toISOString()`),\n so picking Christmas stored `2026-12-25T01:00:00.000Z` and the cell showed a\n timestamp beside neighbours showing plain dates - on a calendar day that\n depended on the user's timezone, since the conversion went through UTC. The\n picked date is now kept as the picked date.\n- **The filter row mounted a native date input** while the cell editor mounted\n the grid's own picker, so one column looked like two different products\n depending on where you touched it. Both use the grid's picker now.\n- **`editorType: 'number'` dropped the character you were typing.** The\n editor mounted `<input type=\"number\">`, whose value sanitization reports\n `\"\"` for anything not already a valid float - so `12.` read as empty, and\n typing `12.5` produced `125`. Same for a lone `-` and `1e`. It is a\n text input with `inputmode=\"decimal\"` and a digit filter now; the trade is\n the native spinner.\n- **Cell context menus and the comment editor floated at stale coordinates\n after a scroll.** Both are positioned at raw cursor coordinates and rendered\n `position: fixed`, but neither was in the scroll effect's gate, so they hung\n over unrelated cells. The comment editor SAVES on scroll rather than\n discarding - backdrop-click and Escape are the gestures that mean \"throw this\n away\"; a scroll is not, and must not destroy typed text.\n- **The footer summary row server-rendered as empty cells.** The totals were\n assigned from an `$effect`, and effects never run during SSR - so the server\n emitted a summary row that took up space and drew its border but held no\n numbers. They appeared only after hydration (a visible pop-in), and never at\n all for a reader with JavaScript off. A grid under the aggregation cell\n limit now derives its totals synchronously, so they ship with the HTML; a\n very large grid keeps the rAF-deferred effect so it still paints before it\n totals. `pnpm ssr:check` asserts the real aggregate now.\n- **The grid painted one unmeasured frame on every mount.** `hasMeasured`\n gates the custom scrollbars, the 16px scrollbar gutter on a trailing\n right-aligned column and the top pager (which carries a `border-top`), but\n it only flipped inside the `ResizeObserver` callback - and that callback is\n deliberately deferred by one `requestAnimationFrame`, so the browser was\n guaranteed to paint a frame with those missing and correct it on the next.\n On a first load that was the \"flashing scrollbar\"; in an app that recreates\n the grid per navigation (a `{#key}` around it, or a route that remounts) it\n showed up as the grid flashing on every sort / filter / page change, with a\n stray border line. The container is now measured synchronously in the mount\n effect - before the first paint - and the observer still handles later\n resizes. Covered by an e2e regression test, since jsdom has no layout to\n miss a paint with.\n- **Pager page-size trigger rendered as a stock button until first click.** The\n closed trigger is a placeholder until the lazy dropdown chunk loads, but its\n box styling only existed in that chunk's scoped CSS, so server-rendered and\n freshly-mounted pagers showed a UA-bordered, left-hugging `<button>` inside the\n bordered box. The placeholder now gets the same rules from `SvGrid.css`.\n- **Virtualized lists no longer flash blank on a fast scrollbar-thumb drag.**\n A fast drag can move the viewport into the windowed list's off-screen padding\n faster than JS can re-render the rows there; the padding now paints faint row\n skeletons, so `SvListBox` / `SvDropDownList` stay visually filled at any scroll\n speed. Also removed an app-wide `flushSync` from the scroll handler that could\n stall the very first scroll.\n- **Dock manager auto-hide now shows every tab of a multi-tab leaf.** Sending a\n leaf with two or more tabs to auto-hide created a single edge tab; it now\n creates one edge tab per pane.\n- **Ticking one row in a large grid took over a second.** Once any row was\n selected, every change walked all rows through the lazy row list to work out\n the header checkbox. At 1,000,000 rows (jsdom) one tick went from 1.3-1.8 s\n to 4-36 ms, and select-all from 4.2 s to 0.4 s.\n- **Tab never left the grid.** Every Tab was consumed and the cursor clamped\n at the last cell, so no key moved focus out (WCAG 2.1.2). Tab on the last\n cell and Shift+Tab on the first now leave the grid.\n- **No focus indicator when the grid is reached by Tab, or under forced\n colors.** The active-cell ring stayed hidden until a click, and Windows High\n Contrast removed it (it is a box-shadow). It now shows on keyboard focus, and\n forced colors get a `Highlight` outline.\n- **The cell editor had no accessible name.** The text, number and multi-line\n editors are now labelled with their column's header.\n\n#### Security\n\n- **CSV / formula injection in exports.** `exportCsv`, `exportTsv` and the\n chart's CSV download wrote text such as `=HYPERLINK(...)` as is, which a\n spreadsheet runs as a formula. Text starting with `=`, `+`, `-`, `@`, a tab\n or a carriage return is now written with a leading `'`; numbers and signed\n numbers in text form (`-12.50`) are unchanged. `escapeFormulas: false` turns\n it off; `copyToClipboard` takes `escapeFormulas: true` to turn it on.\n\n### @svgrid/enterprise\n\n#### Added\n\n- **SvGrid Base: Delete forever can be one transaction.** `transaction` next to `execute` on\n `createSqlDataSource`, `createBaseEndpoint` and `baseFromDatabase`, and `linkedFrom` on\n `createKitHandlers`, make it one server step. See Undo, trash and import for each driver.\n\n- **SvGrid Base: Edit fields says what it sets off.** Past 10,000 records its question gives the\n number of automation runs the change can start, and \"Also run automations\" can be unticked.\n\n- **SvGrid Base: a comment not sent stays with its record.** Stepping to another record no longer\n carries it into that record's box, and closing the record keeps it and says so.\n\n- **SvGrid Base: `baseFromEndpoint` takes a live feed.** `realtime: (table, apply) => stop` is called per\n table and per junction table.\n\n- **Gantt: `pickUpToMove` and `text`.** Keyboard moves in the scheduler's way, and every\n word of the view for a page in another language. Base uses both.\n\n- **SvGrid Base: soft delete.** `softDelete: 'deleted_at'` on `createSqlDataSource` and\n `createSupabaseDataSource` marks a record instead of removing it. Reads, counts, groups,\n totals, lookups and rollups skip it, a restore clears the mark and rewrites nothing, and\n emptying the trash is the real delete. `createKitHandlers` takes `restore` and `purge`,\n inside the tenant's `scope`.\n- **SvGrid Base: a view always gets its answer.** When a data source cannot run a\n filter, sort, group or total on a computed field, or fails on one, the page reads\n the table a block at a time, works the field out with the cells' code and serves\n the view by id, with the count in the tab bar. A record that changes is updated\n where it stands, and taking the sort or filter away stops the read. `computes` on\n `baseSource` says what the source can work out itself; a `queryRecords` backend\n needs `computes: () => true` to keep those requests in the database.\n- **SvGrid Base: \"+\" opens a draft.** On a big table a new record is stored at its\n first value and not at the click, so one closed empty leaves nothing behind. It\n starts with the values of the view's filter, what the database fills in comes back\n into it, and one the database refuses keeps its values and tries again. \"+\" is in\n every view of records, and Ctrl+Z takes a stored record out again.\n- **SvGrid Base: add a field from the header.** `columnAdder` is on by default for\n whoever may change the schema, and the \"+\" stays at the right edge of a wide\n table. `createSqlDataSource` with `schemaChanges` adds the column on Postgres,\n MySQL and SQLite, `addField` on `baseSource` does it on your own source, and a\n source that cannot shows the `ALTER TABLE` to run. A new field never takes over a\n column already in the table.\n- **SvGrid Base: Undo keeps what it cannot write.** A record goes back with every column\n read, and a column the database generates is left to the database, on any engine.\n- **SvGrid Base, in use.** A record's position is the view's own (\"1 of 1,000,000\"),\n and Next and Previous read past what is loaded. A click or Enter opens a Board card.\n Ctrl K works from anywhere on the page. A group of mixed ticks shows as mixed, group\n headers show what the cell shows, and a field deleted from a view's filter leaves it\n (undo brings the condition back). A record outside the view's filter says so.\n- **SvGrid Base: select all means all.** On a big table the header checkbox selects\n every record of the view as a rule, and the count says so. The selection bar's\n Delete reads what the rule covers, asks with the exact number, and deletes in\n blocks with a Stop button. Up to 10,000 can be undone, in one write.\n- **Block writes.** `ServerDataSource` gains optional `createRows` and `deleteRows`;\n `createSqlDataSource` and `createSupabaseDataSource` implement both.\n- **`filterModel.searchIn`.** The fields a global search may look in. A base sets it\n so that a field hidden by a share or a permission cannot be probed by searching.\n- **Server row model:** `pin` and `unpin` show a row above the list that no block\n holds, outside the count, and `leafTotal` gives a grouped view its true size.\n- **Computed SQL:** dates, `MAX` and `MIN` on MySQL and SQLite, `FIND` on MySQL, and\n `ROUND` and `INT` on SQLite.\n- **SvGrid Base: a table has no size limit.** A table used to stop at `limit`\n records, 10,000 by default, and a table on `serverRows` kept that many in the\n page for everything but its grid views. Now no read stops short. A table that\n outgrows the page (`holdInPage`, the new name of `limit`) is read from the\n database a part at a time, by itself, and the page holds none of its records.\n Links, lookups and rollups read records by id, and a rollup over a link stored\n on the big table is added up by the server: in demo 723, a table of 200\n customers shows each one's orders and revenue over a million. The link picker,\n search across tables and the grid's search box search the whole table.\n Automations run on any record, and a scheduled one on every match. Share\n links show the whole table. A new record's autonumber is the table's next.\n New in the data source contract: `groupRawKeys`, for a link grouped by id.\n\n- **SvGrid Base: `createSqlDataSource` runs a view's filter.** It used to leave\n `filterModel.expression` out, so a filtered view on a SQL table showed every\n record. It also filters, sorts, groups and adds up on formulas, lookups and\n rollups now: `computedSql` (`@svgrid/enterprise/base/compute`) turns them into\n SQL, or says it cannot rather than answer differently from the cell.\n\n- **SvGrid Base: formulas in SQL say what a cell says.** More functions translate\n (`SWITCH`, text slices, `MOD`, `POWER`, `SQRT`, days and weeks on dates), and\n each is checked against a cell on real Postgres, MySQL and SQLite over blanks,\n negatives and quotes. `generatedColumnSql` writes a formula as a generated\n column for Supabase, and `generated` makes a filter read it.\n\n- **SvGrid Base: `schemaChanges` on `createSqlDataSource`.** On Postgres a field's\n type change is one `ALTER` and a table's copy one `CREATE`. Off by default.\n Without it, a big table's values are rewritten sixteen writes at a time.\n\n- **SvGrid Base: Supabase with aggregates off.** How a project starts. Kanban\n lanes and link counts now read their size with plain counts, and a total that\n needs aggregates says how to turn them on (`AggregatesOffError`).\n\n- **SvGrid Base: a link edits from either side of a big table.** A customer's\n orders tick in the picker, and each tick writes that order. Names pasted into\n a link cell are found on the server.\n\n- **SvGrid Base: changing a field's type on a big table converts its values.**\n It asks first, then uses `convertField` or rewrites them a block at a time.\n\n- **SvGrid Base: Download as SQL writes every record of a big table,** a block\n at a time. Duplicate table copies records that fit in the page, or in the\n database with the `duplicate` option.\n\n- **SvGrid Base: a Kanban lane keeps its order.** Sorted by a number field, a\n card dropped elsewhere in its lane writes that number.\n\n- **SvGrid Base: every view reads a whole table of millions.** Views other than\n grids used to show the first records of a big table.\n Now a Kanban lane shows its true size and reads its cards as they scroll by;\n a calendar reads the dates on screen, with each busy day's size and a \"+N\n more\" that lists all of that day; a Gantt reads its rows by start date; a\n gallery, list and inbox draw and read what is in sight; a chart and a\n dashboard have the server add up. It is the same `getRows` request a grid view\n sends, so a source that answers a grid view answers these. `queryRecords` now\n answers `aggregations` and `needsGrandTotal`. In demo 723 (a dev build, one\n laptop), each view of the million opens in under a second.\n\n- **SvGrid Base: a Rich text field type (`fieldType: 'richText'`).** It edits\n in a panel over the cell with a formatting toolbar: bold, italic, underline,\n strikethrough, bullet and numbered lists, links and clear formatting.\n `fieldProps: { toolbar: 'full' }` adds headings, quotes, code blocks, undo\n and redo. Ctrl+B / I / U format the selection and Ctrl+K adds a link;\n Ctrl+Enter, Tab or a click outside saves and Escape cancels. The value is\n HTML, sanitized when it is stored and again before it is drawn. A grid row\n shows the formatting on one line, a taller row height as lines, and the\n expanded record and the form show all of it. Search, the filter, sorting,\n copy, lookups and formulas read the words without the markup. Converting\n from long text renders its Markdown; converting back keeps a line per\n paragraph and list item. Demo 722 (release notes) shows both toolbars, a\n filter view and a formula over a note; demo 511 (content calendar) uses it\n for the brief. The long text option that renders Markdown is now labelled \"Format\n with Markdown\", to tell the two apart.\n\n- **SvGrid Base in your language.** Every word a base shows (the tab bar,\n the view bar and its panels, the record view, forms, the field editor and\n the field types, menus, notices, color and icon names) comes from one table\n of about 1,100 messages. `setBaseMessages(words)`, `<SvBase messages>` or\n `<sv-base>`'s `messages` property hand over the words you have; the rest\n stay English. Counts take `.one` / `.other` forms, and a `plural` rule\n for languages with more. `BASE_MESSAGES` lists every key with its English.\n A trash item for a table with one record now reads \"1 record\", not\n \"1 records\".\n\n- **SvGrid Base: Vote and Sparkline field types.** A vote field keeps who\n voted (user ids): a click on the cell, Enter or Space adds the signed-in\n person's vote or takes it back, through the grid's edit path, so undo and\n protected fields hold. It sorts by the count and filters as people (\"has any\n of me\"). A sparkline field draws a list of numbers as a line, area or column\n chart, typed and pasted as `3, 7, 4, 9`, sorted by the last value.\n\n- **SvGrid Base: a trash kept in your backend.** `<SvBase trashStore={{ list, add,\n remove }}>` reads the trash when the base opens and tells the store of every\n item deleted, restored, undone or emptied, so it survives a reload and is\n the same for everyone.\n\n- **SvGrid Base on a phone: boards and calendars.** A Kanban lane takes the\n table's width and snaps into place as you swipe; a calendar opens in its\n agenda. Gallery cards and forms were one column already.\n\n- **SvGrid Base: tables of millions of records.** `baseSource(ds, { serverRows: true })`\n has a table's grid views read it a block at a time through the server row\n model, with the view's filter (`filterModel.expression`), sort, groups and\n search sent to the server, and its edits, new records and deletes written by\n id. Kanban, calendar, gallery and the other views show the first `limit`\n records, and say so. Grouped views get one row per key with its count from\n the server, and the view bar counts the server's records. `queryRecords`\n answers such a request over records a Node server holds,\n `fieldFilterToSql` turns a view's filter into a `WHERE` clause for\n Postgres, MySQL or SQLite, and `fieldFilterToPostgrest` /\n `supabaseFieldFilter` into a PostgREST filter: `createSupabaseDataSource`\n has a new `expression` hook, and `baseFromSupabase` a `serverRows`\n option. A link edited there changes the field that links back for that\n record only (`linkChanged`), not for the records the page never loaded.\n Demo 723 pages through 1,000,000 orders.\n\n- **SvGrid Base: formulas on a server (`@svgrid/enterprise/base/compute`).**\n `computeRecords(tables, tableId)` works out every formula, lookup, rollup\n and count with the grid's own engine, and `evaluateFilter` /\n `filterRecords` test records against a view's filter, with no DOM and no\n Svelte, so an API, an AI tool or a report sees the values people see. A\n test works out a base both ways and compares them record by record.\n\n- **SvGrid Base: permissions.** `<SvBase permissions>` (and a table's own\n `permissions`) take `canEdit`, `canCreate`, `canDelete`,\n `canEditSchema`, `canEditViews`, `canComment`, `fields` (`'hidden'` or\n `'readOnly'` per field) and `viewsOnly`. The screen offers only what the\n person may do: no \"+\" row or Insert record without `canCreate`, no field\n editor or table menu without `canEditSchema`, view changes kept in the page\n without `canEditViews`. A hidden field takes the formulas, lookups and\n rollups that read it along, and the stored records and columns are never\n changed by what a person is not shown. `<sv-base>` takes the same as its\n `permissions` attribute.\n\n- **SvGrid Base: realtime changes update only what changed.** The\n `realtime` callback's `apply(changes)` takes inserts, updates (the whole\n record or just the changed columns) and deletes, or Supabase Realtime events\n as they come, and updates those records in place, a burst at once, with the\n same highlight and credit as before. `baseFromSupabase` uses it. `apply()`\n with nothing still reads the table again.\n\n- **SvGrid Base: a record history that lasts.** `setRecordHistory({ list, add })`\n reads a record's history from your audit log each time it is opened (a\n promise is fine) and hands it each change made in the page.\n\n- **Change review in the record view.** An expanded record lists its pending\n proposals under their fields, with Accept and Reject per field and for the\n record. `review.acceptRecord`, `rejectRecord`, `pendingIn`, `sources`,\n `pendingFrom` and `revertSource` (everything one agent did, across its\n sets) are new, and the panel has a From picker when more than one source\n has proposals.\n\n- **SvGrid Base: undo for tables, form footers and link allowlists.** Adding,\n renaming, duplicating, deleting and importing a table are undo steps.\n `formPrefill` takes a list of the fields a link may fill\n (`prefillFields` on `SvBaseForm`), and a form view takes `formFooter`\n text or `<SvBase formFooter>` as a snippet. `<SvBase onMention>` hears\n mentions. `<sv-base>` fires `cellchange`, `recordadd`, `recordsdelete`,\n `trashchange` and `mention`, and takes a `read-only` attribute.\n\n- **A grouped grid prints its groups.** `print()` printed a grouped grid's\n leaf rows only. It now prints a header row above each group, \"Region: EU\n (12)\", indented by level; `groupRows: false` leaves them out.\n\n- **`print({ open: false })` returns the printable document.** `print()` and\n `printGrid()` now resolve to the HTML they built, and `open: false` skips\n the print window and dialog, for a preview pane or anything that times the\n printable view. It is a plain table, not the grid.\n\n- **PDF: `pdf.fit`, `pdf.repeatColumns` and `pdf.font`.** `fit` chooses what\n happens when the columns do not fit across the page (shrink the text, split\n the columns over pages, or both, the default); `repeatColumns` says how many\n leading columns repeat on each page of a split. `font` prints with a font\n you supply, base64 or a URL, for scripts Roboto lacks (Chinese, Japanese,\n Korean, Arabic, Hebrew, Thai, Indic, emoji); without one the export now\n warns when it sees them.\n- **PDF export in a worker: `pdf.worker` and `@svgrid/enterprise/pdf-worker`.**\n pdfmake renders in one synchronous run, about a second per thousand rows on\n a desktop machine, and on the page that froze the tab. With\n `pdf.worker: () => new Worker(...)` and a worker script that calls\n `servePdfWorker()`, the document goes to the worker as data and the page\n stays responsive: a 5,000-row export took about two seconds while the page\n kept drawing at 60 fps, with progress through `onProgress` and an abort that\n terminates the worker. A PDF of more than 2,000 rows rendered on the page now\n logs a warning that points here. `buildPdfDocData` and `hydratePdfDoc` split\n the document builder for this; `buildPdfDocDefinition` works as before.\n- **Grouped PDFs total like the grid.** Subtotals use each column's\n `aggregate` (avg, min, max, count, countDistinct, a function) instead of\n always summing, a total row closes the table (`pdf.grandTotal: false` to\n leave it out), and group names are formatted as on screen. `getColumns()` in\n `@svgrid/grid` now reports each column's `aggregate` for this.\n- **PDF: `pdf.labels`, `pdf.watermark`, `pdf.author`.** Replace the footer's\n page and date text and the subtotal and total labels; print a faint\n watermark across every page; and fill the document properties, so a viewer\n shows the report's title rather than the file name.\n- **PDF column widths follow the text.** Each column gets at least its\n longest word and the rest of the page goes to the columns with the most to\n say, instead of equal widths: a 60-row, 8-column order list went from three\n pages to two.\n- **The pivot designer's Chart view grew up.** It offers every shape a pivot\n can take (bar, line, area, lollipop, pareto, the three radial forms, pie,\n funnel, waterfall, radar, heat map, stream) in the panel's groups, with\n Stacked, 100% and Horizontal where the type uses them, the chart's own\n toolbar (PNG, SVG, PDF, print, copy), its spoken summary and a legend\n under the plot. It had six types and a Stacked box.\n\n- **`pivotToChartSpec` carries the measure's format and name.** Pass the\n value chip's `format` and the chart reads \"$470k\" on the axis and\n \"$469,662\" in the tooltip, in the measure's currency and locale; percentage\n points get a formatter that appends the sign. With one measure the value\n axis is named after it (`yAxisTitle` overrides, `null` for none). The\n designer's Chart view passes the format when its measures agree, and\n formats plainly when a layout mixes a currency and a count.\n\n- **PDF export with charts and a KPI strip.** `pdf.charts` prints charts with\n the table: a rendered chart's element is rasterised on export through the\n grid chart's PNG path, or pass an image data URL; each takes a title, a\n caption and a width, and goes above the table or below it\n (`chartsPosition`). `pdf.kpis` prints a strip of headline numbers, each a\n label over a big value with an optional coloured delta line. Roadmap\n discussion 36.\n- **The selection bar renderer, plus bulk edit.** `enableSelectionBar()`\n (or `installEnterprise`) registers the renderer behind the free\n `selectionBar` prop. The bar carries a count chip, your actions, an overflow\n menu past `maxVisible`, and Clear; at phone width it collapses to icons and\n keeps an accessible name on each. **Edit fields** opens a drawer that edits\n one or many rows at once: a field the selection disagrees on opens blank and\n stays per-row unless you touch it, so applying to 40 rows does not flatten the\n values you did not mean to change.\n- **Scheduler booking rules and conflict detection.** New model helpers\n (`overlapCount` for per-resource overlaps, `overlapsBands` for working-hours\n bands) back resource double-booking checks and business-hours enforcement.\n New demos: booking rules, financial trading hours, and HR shift coverage.\n\n#### Fixed\n\n- **SvGrid Base: one Undo or Redo at a time.** Redo pressed while a range Undo on a paged table was\n still writing read the records before the Undo's later blocks landed: the data ended undone or half\n pasted while the page said \"Redone: edit 6,000 records\". Now a second Undo or Redo while one runs is\n refused with \"Undo is still running.\" and the buttons show it busy. An Undo or Redo of more than one\n block runs in the paste's dialog with its count and Stop, leaving the page asks first, and what it did\n is said when it ends, with the records or cells it changed (not those a teammate kept, a deleted\n record or a refused write). A small paste's count is said once its writes have answered.\n The dialog covers every Undo and Redo of more than a block: a delete (its Undo says the records\n it put back, not those a teammate restored first), the records a paste added, and more than\n 1,000 cells of a table the page holds. A cell undo of a record the view no longer shows holds its\n turn until that record's write has answered.\n- **SvGrid Base: Undo of a new record before its insert answers.** The step was a cell edit until the\n database answered, so Undo wrote an empty name and the record stayed. The \"add record\" step now\n exists from the moment the row does, what is typed into the record folds into it, and Undo waits for\n the insert, then takes the record out. A refused new record's Activity no longer shows what was typed.\n- **SvGrid Base: Try again from a record that fell to the end.** It wrote the first line into that\n record and offered to add the rest as new records. Now it writes nothing and says how many rows stand\n below the record against how many the first try covered; the dialog of a Try again names the rows it\n writes, which it covers on screen.\n- **SvGrid Base: a range over 1,000 records on Supabase or PostgREST is written.** The ids check read the\n whole range in one request, and a server that answers 1,000 rows at most (`db-max-rows`, the Supabase\n default) gave fewer back: every paste, clear, fill, one value or copy of more than 1,000 records said\n \"Records moved\" and wrote nothing. A short answer now reads on, a page of the server's own size at a\n time, and only a different id sequence counts as moved; a source with no cap still answers each check in\n one request. These checks select the key alone on Supabase and PostgREST. A download, a duplicated table\n and a view's group reads from a source that sends no count also read on past a short answer.\n- **SvGrid Base: a Cut the browser refused says so.** It did nothing in silence; now \"The browser would\n not put the cells on the clipboard, so nothing was cut.\"\n- **SvGrid Base: a notice that stays no longer covers \"+ Add record\" or a record's Delete and Close.**\n Notices in the bottom-right corner start above the button, or an open record's footer, where they\n meet.\n- **SvGrid Base: a table named by its key keeps its records.** A table with no text column shows\n its key as its first field. A record added there was deleted and created again under a new key at\n the next change of any record, a tab switch or a sort, on Postgres, MySQL and SQL Server: a foreign\n key with `ON DELETE CASCADE` took its children with it, \"record created\" automations ran each time\n and a table that marks deletes filled its trash. Edit field on the key could change its type, which\n wrote every record again under a new key, and offered \"Who can edit its values\". With a feed, a new\n record's key stayed blank and the page counted its own insert as a teammate's change. The key field\n now holds the key the database gave the record, in the key's own type, from a read, a write's answer\n or a feed; no edit, paste, fill, undo or field change writes it. Edit field on it changes its name\n and description only, and Duplicate field makes an ordinary field. A key column named `id` is the\n field `id_key` in the table's description, which `rename` maps back to `id`.\n- **SvGrid Base: a table with `customer` and `customer_id` opens.** The link from `customer_id`\n took the name `customer`, so two fields shared one id: the tab threw, showed the previous\n table's records, and the link and its field back were lost. Two keys that read as one name\n (`owner` and `owner_id`) did the same. Now a link whose name is taken keeps its column's name as\n its id and is headed \"Customer (link)\", or by its column when both are links (\"Owner id\"); both\n fields show, and each link keeps its field back.\n- **SvGrid Base: a client's filter on the tenant column is combined with the scope.** The scope\n replaced it, so `tenant_id = 2` asked by tenant 1 answered all of tenant 1's rows, and an\n `updateWhere` with that filter changed them. The two are combined now: another tenant's value\n finds no rows and changes none.\n- **SvGrid Base: a refused write names the record as stored.** A refused edit of the name field\n named the record by the value that was refused (\"Could not save Label of boom\"). A refusal by a\n foreign key of three columns named no field when the edit was to one of them: it names the key's\n fields now. The top bar's \"Not saved\" names them too.\n- **SvGrid Base: the status line opened in full shows no transport text.** It showed\n `svgrid kit: 409 Conflict - {json}` after the reason. It shows the reason, and the server's own\n words when it sent some that the reason does not say. A line with nothing more still opens, to read a\n reason the bar cuts short. Notices that quoted an error (a delete, a copy, a bulk edit, the\n trash) say the reason in words in place of the endpoint's or the REST API's status and body.\n- **SvGrid Base: a new record of a table whose database makes the key is read by no one before it\n has its key.** On a paged table, a draft (or a record whose insert was on its way) was asked for\n by its page id, and so were the records linking to it: Postgres refused the read on an integer\n key (\"invalid input syntax for type integer\"), and that text showed as a notice. Such a record is\n left out of those reads until the database has given it its key. A read that fails says so in\n words; the driver's text goes to the browser console.\n- **SvGrid Base: Duplicate field on a database table the base cannot add a column to copies\n nothing.** It copied the values into a column that was not there, so their writes could not land.\n It now says the copy needs a column and gives the statement that adds it.\n- **SvGrid Base on a REST API: the API's reason is shown, and a refusal of the values offers no Try\n again.** `createRestDataSource` passes `{ error }` or `{ message }` on as the reason, so a `422`\n says \"The server said: Amount must be 0 or more.\" where it said that no reason came. A server's\n words that do not end a sentence get a full stop, so \"You entered: -500\" after them reads apart. A `400`,\n `409` or `422` from any source offers no Try again, which would send the same values.\n- **SvGrid Base on Postgres: a foreign key to a table of another schema on the search path is a\n link.** `public.notes.task_id -> app.tasks` stayed plain text with both schemas on the path and\n both tables in the base. A key now links when its table is the one its bare name reaches. A key\n to a table that an earlier schema's table of that name hides stays plain, in its own schema too.\n- **SvGrid Base: a grouped paged view shows your own changes in its open groups.** Round 7 made a\n teammate's change read every open group again; every other refresh still read only the group\n rows. With no feed, a record you deleted stayed listed in its open group (and typing into it said\n it no longer exists), a record you restored from the trash or brought back with Undo did not show,\n and a new record, an edit that moved a record to another group, a renamed option, a converted field\n and the read after \"Records moved\" left the open groups as they were, while the counts changed.\n The server row model's `refresh()` with no route now reads the group rows and the loaded records of\n every open group on a grouped view, as `refresh({ all: true })` does, so every caller gets it.\n `refresh({ route: [] })` still reads the top level alone. On a grouped view or tree data an app's\n own `refresh()` now also reads the loaded blocks of each open group or node.\n- **SvGrid Base: a teammate's change to a sorted field takes its place in a paged view.** A feed that\n sends the row patched it where it stood, so a record whose sorted field changed stayed out of order\n until something else read the view. A change to a field the view sorts or groups by now reads the\n view again (every open group, when grouped); a change to any other field still patches the row and\n reads nothing. A view sorted or grouped by a formula or lookup reads again on any change, since a\n change to a field it reads moves it too.\n- **SvGrid Base: a face is never another person's while it loads.** User, created-by and\n last-modified-by cells reuse their faces by position on a scroll, and the browser kept drawing the\n last person's picture beside the new name until the new picture loaded. Another picture now gets\n a new `<img>`, empty until it loads, as attachment thumbnails do; the same picture keeps its\n element. The attachment preview's Next and Previous do the same.\n- **SvGrid Base: a checkbox cell is named by its field.** It read \"Checked\" or \"Not checked\" in\n English whatever `setBaseMessages` set, and a screen reader heard the state twice. Its name is the\n field's name (`fields.typeCheckbox` when the field has none), and `aria-checked` says the state.\n- **SvGrid Base: an image in rich text stays on the cell's line.** Under a reset that makes `img` a\n block (Tailwind's preflight), an image in a rich text cell took a line of its own: the text after it\n was cut off, or ran into the next row.\n- **Scheduler: a move keeps the event's length by the clock.** A drag, a keyboard carry, a timeline\n drag, the other selected events of a move, a series edit, Duplicate and a new event at a slot\n added real minutes to the start. On the night the clocks go back a one-hour event dropped on 1 am\n read and was stored as 1am - 1am, and a 45-minute one dropped on 1:30 ended at 1:15. The end is\n now the start plus the event's length by the clock, the length it had when picked up. An end the\n clocks skip lands after the skip, as a start does: on the night they go forward a one-hour event\n dropped on 1:30 reads 1:30am - 3:30am, one hour in real time. A timeline month drag over a\n 25-hour day ended an hour past midnight; a new all-day event on a timeline month was 24 hours,\n not one day. Both now keep whole days.\n- **Scheduler: the Agenda and the tooltip say when an event over several days ends.** A timed event\n from Monday to Friday read \"12am - 12am\"; it reads \"12am - Oct 30, 12am\". An all-day one over several\n days shows its first and last day instead of \"all day\".\n- **Scheduler: a keyboard carry lands at the time carried.** Each arrow stepped from where the last\n one landed, so an event picked up on Saturday at 2:30 am showed 3:30 am on the Sunday the clocks\n skip 2:30, and stayed at 3:30 am on Monday and after. Every step now counts from where the event\n was picked up (days by the calendar, slots by the clock), and a slot step always goes the way\n pressed. With 15-minute slots a step down from 1:45 am on that night landed at 3:45 am; it lands\n at 3:00 am.\n- **Scheduler: the event in hand stays in view.** A carry by the arrow keys never scrolled the\n calendar: the event went under the sticky header and out of sight. The day, week and timeline\n views now scroll with it, and a Gantt chart scrolls with a bar moved by the arrows.\n- **Scheduler: the top edge of an event that ends at midnight resizes it.** Dragged from 11 pm to\n 10 pm, an 11pm - 12am event became 12am - 12am, a whole day, and was written so. Its end is now\n read as 24:00 of its day.\n- **Scheduler: Duplicate can copy the event.** It called `onEventAdd` with the times alone, so the\n copy was a blank \"New event\". `onEventDuplicate(source, start, end, resourceId, allDay)` gets the\n event it copies. In a base Duplicate on a calendar event copies the record, as a card's Duplicate\n does. A whole-day event with no end field copied onto its own day; it copies onto the day after.\n- **SvGrid Base: keyboard focus shows on menu items under forced colours.** The \"All tables\" menu,\n the table menu and a summary's menu drew focus as a tint, which a contrast theme removes. They draw\n a `Highlight` outline there.\n- **SvGrid Base: a checkbox change is said in words.** A screen reader heard \"Done changed to true\";\n it hears \"Done changed to checked\" (`viewbar.valueChecked` and `viewbar.valueUnchecked`), and the\n record's Activity and the Inbox read the same words. A checkbox cell is `aria-readonly` only when it\n cannot be changed.\n- **SvGrid Base: the tenant column never names a record.** A table under `tenantColumn` with no\n `name` column took its first text column, the tenant, as every record's name: link chips all read\n \"A\" and errors said \"Could not save Site id of A\". The primary field is now the first text column\n that is not the tenant, not a key column and not a foreign key column (one that links nothing, a key\n of three columns or into another schema, included). With none, the record's key names it and shows\n as the first column, read-only, pinned and never hidden, moved or deleted, as any primary field is;\n links, errors and the record view read that name. An integer `tenant_id` placed first is passed over\n the same way, and the tenant is never the first column. `EntitySchema.foreignKeyColumns` lists every\n foreign key column a database read found.\n- **SvGrid Base: an empty database opens as \"No tables yet\".** `createBaseEndpoint` with `allTables`\n answered a database with no tables with a `403`, which the scaffolded page read as \"set\n `authorize`\", in development too. It now answers `200` with no tables and reads the database again\n on the next request, so tables made by `db:setup` show on a reload. On Postgres `allTables` listed\n only `public` while every query reached tables through the search path: it now lists the tables of\n `current_schemas(false)`, the first of a name winning as `to_regclass` resolves it, and the svgrid\n comments, the key default and the column types of a table come from that same table.\n- **SvGrid Base: a refused write says what is known.** A foreign key refusing a write (a value naming\n a record that is not there) read \"Other records depend on it.\" with Try again: it now reads \"The\n record Site id points to is not there.\" with no Try again, and the endpoint sends `409` with\n `code: 'link-missing'` and the key's `fields` when the database says which. A delete keeps\n `linked`. A write the server answered with an error (`request-failed`, a trigger's `RAISE\n EXCEPTION`) read \"The server gave no reason, so it may not have been stored.\"; it now reads \"Not\n saved: The server answered with an error. Whoever runs it can see why in its log.\" Only a write\n with no answer says \"may not have been stored\". New messages `shell.why.linkMissing`,\n `shell.why.linkMissingField`, `shell.why.linkMissingFields` and `shell.why.serverError`;\n `shell.why.unknown` has new words.\n- **SvGrid Base: a new record the database refused takes no second create.** Add row opens the new\n record's first cell; when its insert failed (an id column too narrow, say) while that cell was open,\n the row waited for the editor, and what was typed into it sent a create again (in a browser on\n MySQL, two more), each failing with its own notice. What is typed now goes with the failed record,\n for its Try again where there is one, the notice says \"Could not add\" with the name typed and the\n reason, and the row goes when the editor closes. A refused new record that stays listed (while another\n editor is open, or a draft on a table read a part at a time whose refusal no retry can change, such\n as an id column too narrow) takes no edit: a click, a double\n click, a typed key or a paste on it shows its notice instead of an editor, so nothing typed into it\n is dropped. An editor open on it when the refusal comes is closed with what it holds, which the\n notice names. A draft refused for a reason a retry may get past (no answer, a server error) keeps\n taking edits, and its next edit sends it again, as before.\n- **SvGrid Base: a query with no `pageSize` is a `400`.** It was a `500` (\"planToSql: limit must be a\n non-negative integer\"). `pageSize`, and `startRow` and `endRow` when given, must be whole numbers of\n 0 or more, or their digits; the answer says so.\n- **SvGrid Base: two foreign keys to one table give fields back named for their keys.** Customers got\n \"Orders\" and \"Orders 2\" for `customer_id` and `rev_customer_id`; they are now \"Orders (Customer)\"\n and \"Orders (Rev customer)\". One key alone keeps the plain name.\n- **`npx @svgrid/create`: the base template asks for current packages and has a favicon.** Its\n `package.json` asked for `@svgrid/grid ^3.0.11` and `@svgrid/enterprise ^3.1.16`, and it had no\n favicon, so the browser's request for one was a 404 on every page. It now has the SvGrid icon\n (`favicon.ico`, 16 and 32 pixel PNGs, and an Apple touch icon). The pivot dashboard and\n SvelteKit templates' floors moved up too.\n- **SvGrid Base: three records that move at once during a range read stop it.** On a sorted paged\n view, one teammate's edit that moved a record from the part read to past the end, one from the part\n not read into the part read, and one from past the end into the part not read, left every block's\n seam, the far end and the count as they were and repeated no record. A 6,000-line paste then wrote\n about 500 lines to the record beside the right one and said \"Changed 6,000 of 6,000 records\". The\n whole id sequence is checked now, in requests that ask for ids alone: a range longer than a block\n reads its ids first, in place of the record at the far end, and each block must bring the same ids\n in the same places; once the last block is read, the ids of the whole range are read again and must\n be the very ids read, in the same order. Any difference writes nothing. This holds for a range of\n one block too, and for copy, clear, one value and fill. A teammate's move inside the part already\n read now stops the job too, where before the paste went ahead. `createSqlDataSource` reads ids\n alone for these requests (`idsOnly`); a source that does not know the hint sends whole rows.\n- **SvGrid Base: Try again after \"Records moved\" goes from the first record selected.** In the same\n view it pasted by position: after a teammate's edit moved a record above the selection, every line\n landed one record down, a selected record got nothing and one outside the selection got a line. Try\n again now finds the first record selected by its id, goes from there over as many rows below it as\n before, as the list stands at the click, and selects those rows first so the screen shows what it\n writes. The notice says so: \"Records moved while this was read. Nothing was changed. Try again\n pastes from \"Acme\" down over the rows as they are now.\" If that record is gone or no longer\n among the rows shown, Try again writes nothing and says so. Opening or closing a group since counts\n as a change of the view. Paste, one value, clear, fill and copy alike. Messages `shell.rangeMoved`\n and `shell.copyMoved` are replaced by `shell.rangeMovedPaste`, `shell.rangeMovedValue`,\n `shell.rangeMovedClear`, `shell.rangeMovedFill` and `shell.copyMovedFrom` (with `{name}`): a\n `setBaseMessages` override of the old keys needs the new ones.\n- **SvGrid Base: Undo and Redo of a range say how many records kept a teammate's value.** Undo of\n a 2,500-record paste where a teammate had changed one record said \"Someone has changed this cell\n since\" before its blocks were written. It now says \"Someone has changed 1 of these records since,\n so it keeps their value.\" once the blocks are done. A small range undone cell by cell says \"1 of\n these cells\" in the same case, after its checks.\n- **SvGrid Base: a grouped paged view shows a teammate's change in its open groups.** A change from\n the feed re-read only the group rows: an open group kept showing old values with no sign they were\n old, a record moved to another group stayed listed under the old one, and with the changed fields in\n the feed the counts did not follow either. The group rows and the loaded records of every open group\n are now read again, so a value shows in place, a record whose group changed moves to its new group,\n and the counts follow. A feed that sends the changed fields patches the row at once and reads the\n groups again only when a grouped field changed. A group closed at the time reads again when it is\n opened. A read that fails keeps every group's rows and says so in the top bar, as an ungrouped view\n does. The server row model's `refresh({ all: true })` is what does it.\n- **SvGrid Base: Retry in a failed block takes the pointer under the pinned column divider.** Over\n the button the divider showed its resize cursor and its line, and the click relied on being passed\n on. The divider now lets the pointer through to the button while it is over it, and is the divider\n again once the pointer leaves. Its line is no longer drawn across a failed block's band, so the\n words and Retry are clear of it; everywhere else the line is as it was, under forced colours too.\n- **SvGrid Base: the top bar's full reason can be read on any device.** A narrow window cut the line\n short with an ellipsis, and the tooltip held only the server's own words, which a touch screen\n cannot show. A line with a reason is now a button: a tap, a click or Enter opens it in full,\n wrapped, with the server's words under it, and the bar grows only while it is open. Another press,\n or the notice going, closes it. The tooltip holds the whole line, then the server's words. A screen\n reader hears the line and the server's words once, from the button, which carries `aria-expanded`.\n- **SvGrid Base: each Restore in the trash is named by its record.** A screen reader heard a list of\n \"Restore\" buttons; it hears \"Restore Task 4\", as Delete forever already said \"Delete Task 4\n forever\". New message `shell.restoreItem`.\n- **SvGrid Base: \"All tables\" says which table is shown.** Its tables are radio items now, the shown\n one checked, so the tick is announced as well as drawn.\n- **Calendar: Day and Week on a day the clocks change.** Events were drawn an hour off their rows on\n a 25- or 23-hour day, in the page and in Base, and a drop or a slot wrote the time of the wrong\n row. They are drawn, dropped and written by the clock now. A carry by the arrow keys steps by the\n clock too: the night the clocks go back, 1:30 and one step is 2:00, not the second 1:00, and the\n event keeps its length. A Pro timeline's working hours on such a day, and the days after it, start\n at their hour; they were an hour early, and on a 25-hour day one day was counted twice.\n- **Calendar: an event over several days says its last day.** Its name gave the first day only, so a\n bar from Monday to Thursday was heard as a Monday event.\n- **SvGrid Base, WebKit with a finger: a tap beside a link selects, as everywhere else.** WebKit sends\n the press of a tap to the cell and its click to a link, chip or star a few pixels away. That\n click followed the link, opened the linked record or set the rating on the first tap. A click\n now acts only when the press began on the same thing; on a cell already selected a tap beside a\n link does what a mouse click there does, which opens the cell's editor, and does not follow it.\n- **SvGrid Base, forced colors: a ticked checkbox shows its tick.** The white tick sat on a fill that\n a contrast theme removes, so every ticked box read as empty. The tick, a given rating icon and\n your vote are drawn in the theme's text color.\n- **SvGrid Base: a menu item that is off is heard with why.** Download as Excel without jszip, and\n Delete table on the last table, are in the arrow-key order of their menu (`aria-disabled`).\n- **SvGrid Base: deleting the last record of a group in a list keeps focus among the records.**\n Focus went up to the group's header. It goes to the record after it, the first of the next group,\n or to the record before it when it was the last of the list.\n- **SvGrid Base: two records that trade places during a range read stop it.** A paste, copy, clear,\n one value or fill on a sorted paged view checked the ends of each block it read. When a teammate's\n edit moved one record into the part read and another out of it, the ends stayed put: one record was\n read twice, the other never, and the paste wrote a line to the wrong record and said \"Changed 6,000\n of 6,000 records\". A record read twice now counts as moved, and nothing is written.\n- **SvGrid Base: Try again after \"Records moved\" acts only in the view it came from.** Pressed after\n a switch to another view of the table, it pasted into the records at the same positions there,\n which nobody had selected. It now writes nothing if the view, its sort, filter, groups or shown\n fields changed, and says \"That view is no longer open, so nothing was changed.\" or \"The view has\n changed since, so nothing was changed.\" The same holds for the answer to a job's question: past\n 10,000 records, new records past the last row, and a delete.\n- **SvGrid Base: the read count of a range goes up a block at a time.** It read \"2,000\", \"3,999\",\n \"5,998\" records so far: each block after the first read one record of the last again and counted\n one less. Each block now brings 2,000 new records.\n- **SvGrid Base: the notice of a save that stopped part way goes with its Undo.** It stayed after the\n Undo and still said \"Undo takes these back.\"\n- **SvGrid Base: a bulk delete that finds records gone says so in one notice.** \"Deleted 2,598\n records\" with Restore and \"Deleted 2,598 of 2,600 records\" came together; now the notice with\n Restore says \"Deleted 2,598 of 2,600 records\".\n- **SvGrid Base: an edit of a record a teammate deleted reads \"This record no longer exists.\"** It\n read \"The record is no longer there, or you may not change it.\", which suggested a permission\n problem.\n- **SvGrid Base: records that link to each other in a circle delete in one go.** A delete block of\n `n1 -> n3 -> n2 -> n1` went level by level and found no record to start from, so Postgres and SQL\n Server answered \"Other records depend on this one\". Records in a circle now go in one statement,\n which Postgres, SQL Server and SQLite take. MySQL checks each row as it deletes and refuses them;\n the page says so in words, and that the link of one of them must be emptied first. Delete forever\n out of the trash does the same.\n- **SvGrid Base on SQL Server: the trash asked with no limit is its count alone.** It read one record\n where Postgres and MySQL read none.\n- **SvGrid Base on Postgres: a foreign key of two columns links again.** The catalog query paired\n every column of a key with every referenced one, so `(tenant, order_id) -> orders(tenant, id)` read\n as four keys: links showed empty and valid ones were refused. Keys are now read column by column in\n key order on every database, and only within the schema the base reads (a table of the same name in\n another schema, or a key into one, is no longer mixed in, on Postgres, MySQL and SQL Server). A key\n of the tenant column and one more is a link through that one within the tenant, and the tenant\n column stays the read-only tenant instead of an empty link field. Any other key of more than one\n column links nothing and is named in the log. On Supabase, PostgREST does not describe a key of two\n columns, so neither column is a link there.\n- **SvGrid Base: every link through a pair table, and the field that links back, is named for its\n pair table** once the plain name is taken: \"Tags (Ot rev)\", never \"Tags 2\".\n- **SvGrid Base: the All tables menu never squeezes an item.** With many tables, or with the Excel item\n that says jszip is missing, items shrank into each other; the menu now scrolls instead.\n- **SvGrid Base: importing an .xlsx without jszip says so in plain words**, as the Excel download\n does, instead of the developer's install command. `importData` and `readImportMatrix` reject with\n `code: 'no-jszip'`.\n- **Scaffold: dialogs, menus and the rich text editor use the app's font.** The `base`, `sveltekit`\n and `pivot-dashboard` templates set the font on the layout only, and what moves to `<body>` was in\n the browser's serif.\n- **SvGrid Base: a cold automation server judges every record of a block by what it held before.**\n The first write after the server started read the tables after that write, so a pasted block of 45\n \"Won\" ran the rule for one record, and could run it for a record that was Won already. The `audit`\n hook now has a `before` that `createKitHandlers` awaits before each write, so the first read lands\n first. Pass your own audit hook as `automations.audit('tasks', yourHook)` to keep it; a hook you wrap\n yourself loses `before`, and then each record of the first request goes by the fields the write named.\n- **SvGrid Base: a paged view whose read after a teammate's change fails keeps its rows.** One failed\n refresh replaced every loaded row with \"Could not load these rows\". The rows stay, the top bar says\n the table could not be read, and Try again reads them again. New on the server row model:\n `keepRowsOnRefreshError` and `staleBlocks` on each level. The top bar's Try again no longer slides out\n of sight on a narrow window, and a failed block's Retry under the pinned columns' divider takes the click.\n- **SvGrid Base: one feed for a base on an endpoint.** The docs showed one `EventSource` per table and\n junction table. On HTTP/1.1, which a Vite or Node dev server speaks, a browser opens at most six\n connections to a host, so a base of six tables left no connection for a write: \"Still saving\" for\n good. `baseFromEndpoint` takes `feed: (apply) => stop`, one connection whose messages name their\n table, and `createBaseFeed` shares one connection among `baseSource`s.\n- **SvGrid Base: the trash of the person who deleted follows a teammate.** A teammate's Restore and\n Delete forever left the record in the deleter's trash and its count, open or closed.\n- **SvGrid Base: focus stays off another record's Restore.** When a teammate's change took away the item\n holding focus in the trash, or turned its Restore off, focus moved to the next record's Restore, and\n Enter restored a record nobody chose. It goes to the list.\n- **SvGrid Base: a failed write gives the reason the server sent.** An endpoint that answered\n `{ \"error\": \"...\" }` without a code read as \"The server gave no reason\". The page now says \"The server\n said: ...\" for a write and \"The server answered with an error: ...\" for a read. The endpoint's own\n 500 sends `code: 'request-failed'` and keeps the driver's words in its log, as before.\n- **SvGrid Base: a range on a sorted paged view writes nothing when a teammate moves a record under it.**\n A paste, one value, Delete, a fill and a copy read the range a block at a time, by place in the view.\n A teammate's edit that moved one record during the read shifted the next block by one, so lines\n landed on the neighbours and one could land outside the selection, while the page said \"Changed 6,000\n of 6,000 records\". Each block now reads the last record of the one before again and checks it, and\n the record at the far end is read first and checked at the end. If one moved, nothing is written or\n copied, the page says \"Records moved while this was read. Nothing was changed. Try again.\", and Try\n again does the same action on the same cells, on the rows as they stand then. A double click on the\n fill handle that finds records moved says so too; it stopped in silence.\n- **SvGrid Base: new records keep the order they were made in.** `newRecordId()` put two random\n characters before its count, and its clock part ran over every 16.8 hours, so a paste past the last\n row of a paged table showed its new records shuffled. An id is now `rec`, the whole time, a count\n within the millisecond and four random characters: 19 characters, in creation order, and it never\n runs over. Ids made before stay as they are. **Breaking:** ids were 12 characters, so a text id\n column narrower than 19 (`varchar(12)`, `varchar(16)`) now refuses every new record. Widen it to\n 19 or more before you upgrade. The base says so in words: \"Record ids need 19 characters; this\n table's id column holds 16.\"\n- **SvGrid Base: the screen shows a range job when it says it is done.** After a range job or its Undo\n on a paged table, \"Changed 3,000 of 3,000 records\" came while the cells on screen kept their old\n values for a second or two. The rows on screen now take each block as it lands, also the blocks that\n landed before one failed.\n- **SvGrid Base: a record a teammate deleted no longer fails a scoped block on a table without soft\n delete.** Under a `scope`, a key no tenant holds, live or in a trash, is gone: a block skips it and\n finishes, one record by itself answers \"This record no longer exists\", and the key is not sent to\n the source. Another tenant's record, live or in its trash, is still a 403.\n- **SvGrid Base on MySQL: a parent deletes with its children.** Delete, and Delete forever, of a block\n holding records of a table linked to itself and the records that link to them failed on MySQL with\n \"Other records depend on it\". The block now goes a level at a time, the records that link first.\n- **SvGrid Base: a block's answer is used.** A paste whose values a server hook changed read each\n record again, a request a record: 400 for a 400-line paste. The answer is now followed in one step,\n and what a lookup needs of those records is read in one request.\n- **SvGrid Base: a rich text field renders under Trusted Types.** It was a blank page under\n `require-trusted-types-for 'script'`. A base with a field of every type now renders there, with its\n record view and rich text editor (`base-trusted-types.spec.ts`). The paths that still do not run\n under it are listed in SECURITY.md.\n- **SvGrid Base: \"Download this view as Excel\" says why when it cannot run.** Without the `jszip`\n package the item did nothing. It now says the app needs jszip, and a download that fails says so.\n- **SvGrid Base: a checkbox is a real TRUE or FALSE in Excel**, not the words `true` and `false`.\n `exportGrid` with `rawValues` writes a boolean value as a boolean cell too.\n- **SvGrid Base: a \"Could not delete\" notice goes once its records are deleted after all**, instead of\n staying beside \"Deleted 3 records\".\n- **SvGrid Base: two pair tables between the same two tables are two fields with two names.** Both were\n named after the other table, and a third got the second's field id. The second is now named for its\n pair table, as \"Tags (Order tags t)\".\n- **SvGrid Base on Supabase: `date` and `timestamptz` columns are date fields**, read by the Postgres\n type PostgREST names, as on the other databases. They were text.\n\n- **SvGrid Base: a keyboard move in a month calendar stays in sight.** A month drawn in five weeks reads a sixth\n that is not drawn. An event carried into it went out of sight, focus went to another record, Space picked that\n one up, and the first showed on a date it did not hold. The view now goes to the next month with the event, and\n a key on another event while one is in hand acts on the one in hand. An arrow across a change of the clocks moves\n a whole day: a week down used to land an all-day event on the evening before.\n\n- **SvGrid Base: a finger follows the mouse in a cell.** A tap on a link, email, phone or URL cell selects it, and\n the next tap follows the link, so on a tablet such a cell can be selected, copied and edited. A tap beside a\n checkbox's box or on a checklist selects the cell. A tap on the box, a star, a vote or a button acts at once.\n\n- **SvGrid Base: a selected list row or gallery card no longer looks focused.** It has the grid's selected-row fill\n and a tick by its title; the ring is focus alone. Under forced colors the tick stays.\n\n- **SvGrid Base: focus stays with the record.** After Undo of a board move focus goes with the card, not to the card\n that took its place, on a board read from a server too. After a confirmed Delete from a list it goes to the next\n record, not to its group. A note on why a record stays open (a comment not sent or not saved) is said aloud.\n\n- **SvGrid Base: group totals of a paged grouped view are right after each write.** After a\n read that a slow backend had not answered when the next write asked for another, every\n later write kept the old rows and totals on screen for 10 seconds, so the totals showed\n the write before. The block cache now frees a read's slot when it gives that read up (in\n `@svgrid/grid`, below).\n\n- **SvGrid Base: a paged view that cannot read its next part says so in view.** The tab bar\n says \"Could not load *table*\" with the reason and Try again, as a table read whole does,\n and the grid's \"Could not load these rows.\" with Retry sits on the first row wholly on\n screen, not on a row above it.\n\n- **SvGrid Base: a cold automation server sees its first change whatever the row carries.**\n An endpoint's row with a column no field names (a tenant, a sequence) made the first\n change after a start look like a change of nothing, and no rule ran.\n\n- **SvGrid Base: a paged group of records that link to nothing reads \"Empty\"**, as a lookup\n group does, instead of a blank label.\n\n- **SvGrid Base: an open trash follows a teammate's Delete forever and Restore.** It reads\n the table's trash again when the feed reports a change, and a record whose Delete forever\n stopped half way no longer reads \"just now\".\n\n- **SvGrid Base: `queryRecords` counts from the other side of a link again, and its \"is\n empty\" agrees with its groups.** A count or rollup over records that point at a\n customer read 0 when the tables given to `queryRecords` held no such customer. A link\n grouped by id now groups by the id it stores. A view filter \"is empty\" on a link now takes\n a key to a record the linked table does not hold, as the Empty group and the cell do.\n- **SvGrid Base: a paged table whose database stops answering no longer leaves an unhandled\n rejection behind.** The count read beside a grouped view's groups, the record read for a\n feed that sent only the changed fields, and the count read of a dashboard or calendar each\n failed with nothing waiting for them. A feed's record that cannot be read now makes the\n view read it again later, and an Undo of a cell the grid no longer holds says the database\n could not be reached instead of doing nothing in silence.\n- **SvGrid Base: a block write of 20 records or more on a table the page holds offers Try again,\n and shows what the server stored.** A paste, a clear or an Undo of that many records that\n failed said so with no Try again, and two failed blocks of 1,500 records read as \"2 changes\". Now the notice\n counts records and Try again reads them first, sends what is still as it was, and says how\n many cells and records it left. Each block's records are read back once it lands, so a\n trigger's or a hook's value shows at once, and Undo finds it.\n- **SvGrid Base: a new record not stored yet inside a range is no longer passed over in\n silence.** A range job left it out with no word, a paste over it shifted every line up a\n row, a fill from it emptied the cells it reached, and a copy dropped its line. Now the job\n says how many it left, a paste keeps its line for it, a fill takes its value, and a copy\n has its line.\n- **SvGrid Base: a range job on Supabase or PostgREST counts and stops within 200 records,\n and stops when a record fails.** Differing values go one request each there, and a block of\n 1,000 held the count and Stop for 1,000 requests. A failed request now stops the other\n fifteen at once, so the count of what landed is what the database holds, and a record a\n row level policy skipped is a failure, not a write that passed.\n- **SvGrid Base: Delete or Backspace in an open single select empties it.** A click on a\n selected select cell opens it, and the key then did nothing.\n- **SvGrid Base: Undo of a range or Edit fields on a server with a hook takes back what the\n hook stored.** It took the stored value for a teammate's and left it.\n- **SvGrid Base: an undo no longer writes over a teammate's change made a moment after yours.**\n On SQL Server and MySQL a write is read back with a second statement, and a teammate's write\n landing between the two was taken for the answer to yours: Undo then thought the cell still\n yours and put your old value over theirs. SQL Server now writes and reads in one transaction\n of one batch. MySQL does with the source's `transaction`; without one the gap stays, and the\n docs say so.\n\n- **SvGrid Base: day filters and a chart's months follow the viewer.** On `timestamptz` and\n `datetimeoffset` columns \"is 8 March\", \"today\" and a chart's months were the database's days.\n The page now sends each date condition with the moments its days start and end where the\n person is, and a chart reads each part of the span with one UTC offset in that offset. A date\n and a wall clock read as before.\n\n- **SvGrid Base: a table keyed by both ends is a link to many.** A pair table with both key columns foreign keys is a link on SQL and Supabase, under a scope too, and a removed link deletes that one pair. Before, on Supabase it was a tab whose delete took every pair sharing the first column, and on SQL it stopped the whole base. Any other composite key, or no key, leaves that table out with a log line and the rest opens. `keys` names a unique column for such a table or a view. A table or view with no declared key, even with a column called `id`, is now left out until you do.\n- **SvGrid Base: the page shows what the server stored.** After a typed edit, paste, Add row, record edit or undo the cell holds the row the server answered: a hook's value, a stamped tenant, a default. Undo goes back from that value. The tenant column shows and takes no edit, and hiding it no longer makes the table a `403`.\n- **SvGrid Base: a self-linked record and its parent come back together.** The link check of a block counts the keys in that block, and a block writes parents first.\n- **SvGrid Base: lists in `json` and `jsonb` columns.** A multi select, a vote or a sparkline saves on `pg`, `postgres.js`, MySQL, SQL Server and SQLite, cast by what the column is.\n- **SvGrid Base: SQL Server sorts, filters and groups on formulas and lookups in SQL.** What it cannot say is refused at once, where the table used to be read into the page.\n- **SvGrid Base: `maxPageSize` below 1,000 works.** The page follows the server's limit, on reads and on blocks.\n- **SvGrid Base: Edit fields on Supabase is a request per 200 records** for records given one value, and a row level policy that skips a record stops the block and says so.\n- **SvGrid Base: small server fixes.** A bad page or a private field name in a request is a `400`. A MySQL key past 2^53 links. A cross-site browser request is not believed for its `X-Forwarded-Host`. `isSecretColumn` knows `pass`, `userpassword`, `adminpass`, `apikeys`, `secretkey`, `pwhash`, `database_url`, `dsn` and `webhook_url`. The evaluation card builds no HTML string, so Trusted Types pass. A scaffold's theme script is a file, so a strict CSP passes. A PDF preview gives the focus back with the pointer over the file.\n- **Compliance docs.** No GPG fingerprint is claimed, one security address, canvas use and the print popup stated as they are.\n- **SvGrid Base: a table that cannot be read says so and stays said.** Where the records would be: \"Could not load Deals\", why, and Try again. No empty table, Add row or \"Not saved\" on the way.\n- **SvGrid Base: Excel files hold real cells.** Numbers, percents, dates and times are numbers with the field's format, whatever a database driver gave as text: on Postgres, MySQL and SQL Server, held or paged, with lookups and rollups. A yes or no field reads the same word on every driver. A link, a select or a person with a number for a key reads by name, in CSV too.\n- **SvGrid Base: a click to select a cell changes nothing.** The first click on a link, email or phone selects the cell, and the next one follows it. Ctrl or Cmd click follows at once and a double-click edits. A click on a checkbox's box, a star, a vote or a button acts at once; a click on the rest of the cell only selects it. A finger's tap follows the same rules.\n- **SvGrid Base: a calendar's day sizes are the viewer's days.** On a column of moments (`timestamptz`, `datetimeoffset`) the sizes and each day's list follow the page's zone, across a change of the clocks. A date or a wall clock reads the same everywhere and is told as it reads.\n- **SvGrid Base: a big change's automations show and stop.** The Edit fields question says how long the runs will likely take and what a closed tab means. The bar shows how far they are, with Stop, and the tab warns before it closes. Waiting runs no longer hold their records, runs on records in the page leave the tab its turns, and a run that cannot read its record no longer stops the ones after it.\n- **SvGrid Base: memory comes back after a bulk edit.** The page keeps the changes of the latest 20,000 records of a table, and no key per record written.\n- **SvGrid Base: a grouped view on a table read by parts reads like one in the page.** Groups in option order, drawn as the field draws them, with the records they hold at every level (a driver's COUNT as text included), and totals for the fields with a sum, an average, a lowest or a highest in the footer.\n- **SvGrid Base: Undo says what a drag did.** A bar and the bars it pushed are one step, and a card or a bar reads \"Undo move record\". \"Generate for all\" on an AI field asks first when records hold values, and one Undo takes the whole run back.\n- **SvGrid Base: a link kept in a `text[]`, `jsonb` or MySQL `json` column.** An `in` filter on one asks whether any value matches, so opening a Gantt on a self link no longer raises `malformed array literal`.\n- **SvGrid Base: nothing shows under Apply in Edit fields.** The row now covers the drawer's padding.\n- **SvGrid Base: a multi-column paste, clear or fill on a paged table writes every column.** It\n wrote the first column of each row and put the old value back in the others, with no message.\n- **SvGrid Base: a range on a paged table is the whole range.** Paste, one value over a range,\n Delete, fill and Ctrl+Enter reach every record the selection covers, loaded or not, through the\n Edit fields dialog: read and written in blocks, with progress, Stop, a count at the end and one\n Undo step. They used to reach only the rows the page held, and \"All changes saved\" said nothing\n of the rest. Up to 200 cells of loaded rows stay instant, with no dialog.\n- **SvGrid Base: a copy of rows not loaded reads them first.** It wrote blank lines and a $0.00\n for each. Up to 2,000,000 cells it reads with progress; past that it copies nothing, says the\n number, and points to Download.\n- **SvGrid Base: Undo and Redo of a range do not freeze the tab.** A big range is one step that\n reads and writes in blocks, and on rows held in the page a step's cells go back in one pass.\n- **SvGrid Base: a paste past the last row.** A line where no value fits makes no record. Past 10\n new records it asks first, with the number. They go in one block, and Undo takes them out in one.\n- **SvGrid Base: a paste while a view reloads makes no record.** It says the view is loading. A\n write that reads the view again leaves the rows on screen until the new ones arrive.\n- **SvGrid Base: a paste and a clear stamp last modified fields and the record's activity**, as a\n typed edit and a fill do.\n- **SvGrid Base: Undo takes back the Try again of what it undid.** A press on that button wrote\n the undone value, with no step to undo. Esc during a drag of the fill handle or a range now\n cancels it.\n- **SvGrid Base: Delete forever no longer cuts links on a record that comes back.** A teammate's\n Restore during the purge, a server failure at any step or a closed page could leave a restored\n record with its foreign keys empty, and a failed purge could overwrite a teammate's new link. The\n record is now claimed first, so no restore meets a half done purge, the keys are written only\n where they still hold what was expected, and the purge reports the ids it deleted.\n\n- **SvGrid Base: a restore is not a creation.** Restore and Undo of a delete ran \"record created\"\n server automations, which rewrote the record. The first change after the automation server starts\n is seen too.\n\n- **SvGrid Base: a record a teammate trashed no longer fails a block with \"no permission\".** Under a\n `scope`, a block skips it and finishes, as it does with no scope, and one record says it is gone.\n The audit log and `afterDelete` no longer name a record the delete did not change.\n\n- **SvGrid Base: a link to a trashed record is empty everywhere.** \"Is empty\", \"is not empty\", one\n Empty group with no raw key, sort, count and the picker, on held and paged tables. An undo to a\n record deleted forever says so, with no key and no Try again, and a link written to a trashed\n record says why it shows empty.\n\n- **SvGrid Base: your own edit with a lost answer is yours.** A read finds it landed: no failure, no\n Try again and no \"changed by someone else\". Undo of a hard delete no longer asks for a restore the\n table cannot do, and a write that was refused or found its record gone leaves no undo step.\n\n- **Board: a keyboard move writes once, on the drop.** The arrows only show where a card is going, and\n `onCardMove` and `onCardReorder` fire when it is dropped, from where it was picked up. Escape used to put the\n card back on screen while the record kept the new lane. Tab or a click elsewhere puts it back too, focus stays on\n the card, and on a `server` board the drop marker shows the way until the drop.\n- **Scheduler: Escape works past the edge of a view.** A carried event used to leave the view and take Escape with\n it. The view now follows the event, and Escape brings both back.\n- **Scheduler: an all-day bar in Week or Day takes the keyboard.** Space opened it instead of picking it up, and it had\n no name with its date. It now moves as any event does; up and down move it a week.\n- **SvGrid Base: a list read from a server takes Tab.** Flat or grouped it is one stop, on the record you were last\n on, and the arrows walk records, not the row buttons. Grouped lists read from a server also take Home, End, Right\n and Left. Right and Left swap in a right-to-left page.\n- **SvGrid Base: Enter on a list row's button opens its menu**, not the record, and a selected row shows it and\n says \"Selected\".\n- **SvGrid Base: focus no longer falls to the page.** The selection bar goes back to Delete after its question, and to\n where you were after Clear, Escape or a confirmed Delete. A sent form puts focus on its thank-you and says it. A\n dashboard tile moved by its buttons keeps focus on the button pressed, which at either end is off, so a second\n Enter does nothing.\n- **SvGrid Base: a comment's question and Next.** Escape on \"Delete this comment?\" leaves the question, not the\n record, and focus starts on Cancel. Next and Previous wait for a comment being changed, and say so, instead of\n dropping the text. A click outside a record with text not sent says why it stays open.\n- **SvGrid Base: small accessibility fixes.** A link chip in a list row is no longer a button inside a button, Add\n table keeps its 26 px on a phone, a mail link in a record is 24 px tall, and the Menu key cannot open the browser's\n menu over a card's. A record with no name reads \"Untitled\" everywhere.\n- **Edit fields on a big table no longer freezes the page.** It runs in the dialog a delete uses, with progress and\n Stop. Past 10,000 records it asks first, with the exact number. A failure part way stops there, reads the block back,\n says how many records changed, and keeps Undo for those; it used to carry on, say none were saved and leave no Undo.\n- **Bulk writes through `createBaseEndpoint` go in blocks.** `createMany`, `updateMany` and `deleteMany` carry up to\n 1,000 records a request, and records given the same values are one statement. `authorize`, `scope`, `validate`,\n `checkWrite`, hooks and `audit` hold for each record as on a one-record write.\n- **An edit, a paste, an undo or a redo deep in a big table is as quick as one at the top.** Each walked every row\n before it, several times over.\n- **A calendar on a date-with-time field groups by day in the database** (`groupDays`, on every dialect), and a month\n moved past stops sending its days. `groupMonths` works on SQL Server and SQLite too.\n- **`createSqlDataSource`: `updateRows` keeps the time** of a schema `date` over a timestamp on Postgres, where a cast\n dropped it, and writes records given the same values without a CASE. A block's statements carry the tenant.\n- **The link picker's Enter waits for the server's answer.** On a table a server searches it took a row of the last\n search, or made a record whose name was there.\n- **A table read a part at a time asks for 200 records a block by default,** not 100: a sorted read costs a sort of the\n table whatever it returns.\n\n- **`createBaseEndpoint`: a junction table under a tenant `scope`.** A junction holding the tenant\n stays a link to many with `tenantColumn`, and one holding none takes `{ through: 'order_id' }`, so\n a pair belongs to the tenant of that record on every path. A `null` for a junction between scoped\n tables used to share every tenant's pairs: it is now a `403`, and the log says what to do.\n\n- **`createBaseEndpoint`: a link stays inside the tenant.** A foreign key written through the route\n must name a record inside its table's `scope`, in blocks too. Another tenant's record and no\n record get the same `409` with `code: 'link-refused'`. Before, a tenant could point at another's\n record, block its delete, and learn which keys exist.\n\n- **Base: a table keyed by a secret does not start.** A key named like a credential reached every\n reader, because a key cannot be hidden. Such a table is now left out, the log says why, and\n `exposeSecretKeys` serves it by name. A foreign key holding that value stays out too.\n `isSecretColumn` also knows `passcode`, `user_pass`, `session_key`, `activation_key`,\n `remember_digest`, `hashed_pw` and `mfa_seed`.\n\n- **`baseFromSupabase` asks only for what it shows.** It read `*`, so a column `hide` held back still\n reached the page, and a table with a column `anon` may not read failed. `createSupabaseDataSource` takes\n `schemaColumns`, and a base sets it. A record brought back by undo gets a hidden column's default.\n\n- **Escape closes a PDF preview.** Chrome's viewer took the focus as it loaded, and keys pressed there\n never reached the page. The focus now comes back to the dialog, unless a person went into the file.\n\n- **`createKitHandlers` takes JSON from its own site only.** A `POST` that is not\n `application/json` is a `415`, and one from another site a `403`, on `vite dev` as well as in a\n build. A page on another host goes in `origins`. A caller of `handle` that sent no content type\n must now send one.\n\n- **Server logs hold no value a driver quoted.** The handlers log codes and names, and take the value\n out of the text: a SQL Server or MySQL duplicate key, a Postgres `detail`, mysql2's `sql`.\n\n- **A color from data cannot reach a style.** An option's color, a record's value used as a color and\n a label's color now go through `safeColor` on the Board, the scheduler, the Gantt, `SvBoard`,\n `SvRecordDetail`, `SvSchedule` and the automations dialog, and a spreadsheet file's formats through\n `entryToStyle`.\n\n- **`GET` keeps an automation's outbound details from those who do not need them.** A webhook or\n an email is described only to a person `authorize` lets change the table, and an automation that\n runs on the server sends no URL, header or address to any page. `sharedTables` drops them too,\n with what a view said of a hidden field.\n\n- **Print and the HTML export run no script.** A cell's link is `http`, `https`, `mailto`, `tel` or\n `blob` only, a rule's color cannot leave its attribute, and the document carries `script-src 'none'`.\n\n- **`baseToSql` keeps a title inside its comment.** A table named with a line break could start a\n statement of its own in the downloaded file.\n\n- **Compliance pages.** `gdpr.md`, `hipaa.md`, `soc2.md` and the compliance overview now say what\n holds for `@svgrid/grid` and what SvGrid Base adds. The doc site's host was wrong in `gdpr.md`.\n\n- **SvGrid Base: the fill handle fills by the field, not by the look of the stored value.** One\n source cell copies in every field, so a person, an option, a color or a time no longer counts up\n to the next one. Two or more source cells continue a series in number and date fields and repeat\n in every other. A value the field cannot hold is not written, and the page says so. Ctrl+D copies\n down, and is one undo step.\n- **SvGrid Base: a copy writes what the cell shows, and a paste brings the same value back.** Percent,\n duration, currency, date and time cells copy as shown, so Excel gets `50%` and `1:30`, and a paste no\n longer turns 50% into 1%. Inside a page the stored value comes back in full: 12.5% shown as 13%,\n bold in rich text, the right one of two people with one name.\n- **SvGrid Base: one multi-line cell in is one cell out.** Copy quotes a cell holding a line break or\n a tab, and a paste reads quoted text, Excel's included, so the record below is left alone.\n- **SvGrid Base: import keeps what a file says.** A date keeps its calendar day in every time zone, a\n percent reads as the grid reads it, a person matches by name or email, and a name that matches\n nobody is no longer stored as a person. Link fields can be mapped by the linked record's name. The\n dialog says how many cells it could not match. A type change from text to a date or a person reads\n the same way, and a rating stays within the field's maximum.\n- **SvGrid Base: a downloaded view reads back to the same values.** A percent is written `12.5%` and a\n duration `1:30:15`, where the stored 0.125 and 5400 would read as something else in an import, and a\n name with a comma in a list is quoted. In Excel a percent stays a number with a percent format, and\n an import reads an Excel cell as it shows: a percent as a percent, a date cell as that date. A select\n option is known by its label as well as its value, so an import adds no copy of it.\n- **SvGrid Base shows its own notices.** A host with no `<SvToaster />` now sees every failure, refusal\n and Restore. A host that has one keeps it, and nothing shows twice.\n- **SvGrid Base: Add record in a grouped view goes to the new record's first cell.** A new record in a\n group of its own opened no editor, so typing pressed the button again and made blank records.\n- **SvGrid Base: undoing a delete after a restore on another tab keeps both sides of a link.** Roll-ups\n on the other table were left short.\n- **SvGrid Base: smaller things.** Space on a Vote cell votes. A click beside the number in a record's\n number editor keeps the edit. Enter in the new view's name closes the view menu. The undo button and\n the screen-reader message name a fill, a paste or a bulk edit by its cells, and a redo is said as a redo.\n\n- **SvGrid Base: focus no longer falls to the page.** A dialog or menu whose opener went, or\n cannot take focus (a menu item, a row button hidden under the dialog, a field just deleted),\n hands focus to what held it, which is the grid, or to the next card. Pin, Hide field, Delete\n field, Expand record, Comment, a deleted card or calendar event, and Undo or Redo on the last\n step all leave the keyboard in place.\n- **SvGrid Base: Remove on a dashboard tile takes the keyboard to its question,** which is read\n out. A second Enter cancels; it no longer lands on the next tile's buttons.\n- **SvGrid Base: Escape in a comment keeps the text.** It leaves the box, and acts as Cancel on\n a comment being changed. Text not sent holds the record open against Escape, which then says\n why, and against a click outside.\n- **SvGrid Base: cards and list rows keep the record's actions.** At a phone's width, in a\n gallery and in a list, the Menu key, Shift+F10, a long press and a button on the card open\n expand, comment, select, insert, duplicate and delete. Selection shows on the card and in the\n selection bar. `gallery.cardMenu`, `gallery.isSelected` and `gallery.text` are new.\n- **Board:** \"Add card to Lead\" and \"+ Add a card\" show the focus ring. An empty lane no longer\n says it is a list. **Link preview, tooltips:** Escape closes a link preview, a grid tooltip, and\n a calendar or Gantt tooltip with the pointer still on it, and a pointer crossing the gap to a\n link preview keeps it open.\n- **SvGrid Base:** a rating cell's name and a gallery's name are translated. Run now with no\n record writes \"(no record)\" for a field token, not the raw `{Deal}`. A view bar button reached\n at a phone's width comes fully into sight. A percent bar keeps its fill under forced colors.\n\n- **SvGrid Base: a tenant's trash on MySQL and SQLite.** With `scope` and `softDelete` the tenant\n was bound before the ids, and `?` is matched by position, so Undo, Restore and Delete forever\n found nothing: a restore said the record was deleted forever, and a Delete forever reported\n success and left the row.\n- **SvGrid Base: a soft delete leaves foreign keys alone.** It used to clear them in other tables,\n and a restore after a reload or by a teammate never brought them back. The clearing now happens\n at Delete forever, in the tables of the base. A hard delete is as it was.\n- **SvGrid Base: a database's numbers show as stored.** Introspection reads a column's scale:\n `numeric(12,2)` holding 19.99 shows `$19.99`, not `$20`, and an integer shows no decimals. A column\n with no declared scale takes the places its values show. On SQL Server `money` has four. MySQL's\n `tinyint(1)` is a checkbox, a `time` column a time of day and never a select, and a Postgres\n `text[]` a multiple select that can be edited.\n- **SvGrid Base: an insert whose answer was lost.** A read says whether it landed: the record is\n taken, what was typed goes to it, and nothing false is said. A key the database makes is found by\n the record's values, in a table the page holds. A 5xx with no code now says the server did not\n answer, not that the database refused, and a delete or an edit that landed is not told as a failure\n or as a teammate's change.\n- **SvGrid Base: a refused undo shows the value it found.** Without a feed the cell kept the value the\n database no longer held.\n- **SvGrid Base: JSON keeps every digit.** A `json` or `jsonb` column is read and written as text, so\n a number past 2^53 is not rounded by a read, an undo or an edit.\n- **SvGrid Base: a notice goes with its base.** The toast queue belongs to the page, so a \"Deleted ... Restore\" or a Try again\n outlived the base that made it, and the next base on the page drew a button that did nothing. Restore now goes with its base, and a\n failure keeps its words, what was typed included, with no button.\n- **SvGrid Base: smaller things from teams.** Someone who may not delete hears no error from the trash,\n and the panel says why it is empty. A permission refusal offers no Try again and leaves no undo step. A\n paged view with a filter drops a record a teammate moved out of it. A teammate's restore lands at the\n record's place. A Delete forever takes the \"Deleted ... Restore\" notice with it, and a number deleted\n in this page is not given to the next record while it stays open.\n\n- **BREAKING: `scope` refuses an empty tenant.** A value of `''` used to drop the predicate,\n so a person with no organisation read, rewrote and deleted every tenant's rows. Now a\n resolver that returns nothing, or a rule whose value is empty, blank or not a single value,\n is a `403` on every read and write, and `createBaseEndpoint` leaves the table out of `GET`.\n Only `null` means no scope. `scope: (table) => ...` must return `null` for a table every\n tenant shares: returning nothing now closes it.\n- **SQL Server runs.** `dialect: 'mssql'` wrote `LIMIT` and `RETURNING`, so nothing could. It\n now pages with `OFFSET ... FETCH`, reads a key back with `OUTPUT INSERTED ... INTO` (a table\n with triggers works), writes an identity key under `IDENTITY_INSERT`, stays under 2,100\n parameters and 1,000 rows, and keeps `datetime`, `datetimeoffset`, `bigint` and `decimal`\n exact. Run against SQL Server 2022. Formulas are worked out in the page there. Studio's\n connection string for it was a URL the `mssql` driver cannot read.\n- **BREAKING: a PDF preview frames only bytes it read.** A file on another origin went into the\n frame as it was, and could draw a sign-in form under the base's own title and move the tab.\n A file the page cannot read, because its host sends no CORS headers, now offers a download.\n- **An id from a client reaches the source as text.** A number was bound as a number, and on\n MySQL with a loose `sql_mode` `id = 0` matched every id not starting with a digit. An empty\n id is a `400`.\n- **`patch: { id }` no longer changes a record's key.**\n- **A webhook connects to the address it checked.** A name could answer one way for the check\n and another for the connection. On Node the call now goes to the checked addresses, and a\n name that does not resolve is not sent. More ranges are refused: `192.0.0.0/24`,\n `198.18.0.0/15`, the documentation blocks, `fec0::/10`, and IPv6 forms wrapping an internal\n address (`::a.b.c.d`, NAT64, 6to4, Teredo).\n- **`isSecretColumn` knows more names.** Any `token` (`remember_token`, `token_hash`),\n `pass_hash`, `password1`, `pin`, `recovery_codes`, `credit_card`, `iban` and others. It reads\n the name only, and the docs now say that is a default, not a guarantee.\n- **`createBaseEndpoint` takes `validate`, `hooks` and `maxPageSize`.** Writes through it could\n not be checked on the server. `validate: true` asks nothing of a `NOT NULL` column with a\n default, and a schema read from a database no longer marks such a column required.\n- **A refused write says why, through an endpoint too.** A constraint was a `500` and \"The\n database refused it\". It is now a `409` with a `code`, and the person reads which rule it\n was, without the driver's words. An autonumber clash through an endpoint now takes the\n next number, as it does on a direct source.\n- **Thrown text stays on the server.** A message thrown by `authorize` or `scope` reached the\n client, connection strings included. It now reads `forbidden`, unless the `Error` has\n `expose: true`.\n- **A read has a ceiling.** `createKitHandlers` returns at most `maxPageSize` rows (20,000 by\n default) with the true count, so `pageSize: 1000000000` no longer reads a table in one go.\n- **An alert rule's style is a color.** `red;background-image:url(...)` fired a request to\n another host each time a cell painted.\n- **A group key that is an object is refused.** A driver could write one as SQL of its own.\n- **A filter on a number past 15 digits compares exactly.** Rounded, a tenant id could be\n someone else's.\n- **A scaffold's production build says what is closed.** It read \"The database did not answer:\n 403. Is DATABASE_URL set\", which sent you to the wrong fix.\n- **Two dev console warnings are gone**: `state_proxy_equality_mismatch`, once for each field\n of each table, and `ownership_invalid_binding` on `views`.\n- **\"Not in view (1 record)\".** It read \"1 records\".\n- **Forms take the mouse again.** Typed boxes ignored a click or a tap, and so did an\n open editor in a record. A star clicked in a form is now the answer; it used to open\n the editor.\n- **A record's Account picker no longer links what nobody chose.** Enter on the list as it\n opened ticked its first record, and Done saved it. Enter now closes it unchanged, and\n a picker left as it opened writes nothing.\n- **Focus no longer falls to the page.** Comments, a form's pickers, the form builder,\n dashboards, filter conditions, automations, attachments, import, the inbox and the\n field editor keep the keyboard on the logical next control, and comment changes,\n tile moves and form builder moves are said aloud.\n- **Base has the fill handle its docs promise.**\n- **The form builder's field list holds still.** A field turned off keeps its row, and back\n on it takes the place the row shows.\n- **Timeline.** Delete on a bar asks first, as does Delete in a calendar's menu. A bar moves by\n Space and the arrows with one write, stands clear of the task table when focused, opens\n its menu with Shift+F10 with focus in it, reads as a table, keeps its shape in forced\n colors, and every word goes through the base's messages. A drop on a weekend stays\n there.\n- **Attachments: Enter on Remove removes the file.** It opened the preview.\n- **Dashboards ask before a tile is removed.**\n- **Form number boxes read as the field does** (\"25%\", \"$1,250.50\") and edit in its units.\n- **Escape in a number or a rating in a record leaves the box.** It closed the record.\n- **Typing on a focused record field starts the edit**, and Enter in the field type search\n takes the first type.\n- **A double-click on a week's all-day row adds an all-day record.**\n- **Automations:** step buttons name their step, moves are said, and the open automation\n carries `aria-current`. **Delete record** ties its text to the dialog.\n- **SvGrid Base: Edit fields writes to the ticked records and nobody else.** In a sorted,\n filtered or grouped view each value that landed moved its row, and the next one went to\n whatever had taken its place, so records nobody ticked changed and Undo wrote more wrong\n values. Every cell is now worked out first and written by record, in one write and one\n undo step, on tables in the page, on SQL tables and on tables read a part at a time,\n where Edit fields did nothing at all before. Text a field cannot read, such as `abc` in a\n number, stops the form with a message and writes nothing.\n- **SvGrid Base: Delete, paste and fill stay on their records.** A row held under a typed\n edit but moved at once under Delete or a paste, so a second press hit another record.\n The row now holds until the cursor moves on, whatever wrote it.\n- **SvGrid Base: a paste past the last row adds records,** as one undo step with the cells\n it wrote. It used to drop those lines without a word.\n- **SvGrid Base: one rule for pasted values a field cannot read.** A number, a select, a\n date, a person, a link, a rating or a checkbox given text it cannot read keeps its value,\n and one notice says so. An unknown link name no longer wipes the link, and an unknown\n name no longer lands in a person field as a person nobody has.\n- **SvGrid Base: a write with no answer is not given up on.** A write past `writeTimeout`\n stays out: later writes to its record wait behind it, writes to other records go on, and\n the status says so. It can no longer land late over a newer value, a late insert leaves\n no second record behind a Try again, and a late delete keeps its copy in the trash until\n the outcome is known. A rollback now leaves a value written after the failed one.\n- **SvGrid Base: the undo guard fails closed.** A cell undo reads the record first, with a\n feed or without one, and is refused with a message when that read fails, so a feed down\n without a word can no longer take a teammate's value out. The step stays the next to try.\n- **SvGrid Base: Try again checks first.** It reads the record, leaves a value that changed\n since, whoever changed it, and says so, also when the record was deleted.\n- **SvGrid Base: a failed read before a delete stops the delete.** That block stays, and\n the person is told, instead of going ahead on a stale copy.\n- **SvGrid Base: editors in a record are held for.** A teammate's change waits while a field\n of the record view is being edited, as it does for a cell.\n- **SvGrid Base: fewer false \"changed by someone else\".** A number the database rounded, a\n person's own undo on a table read a part at a time, and a feed's word on an earlier write\n of the same record are no longer reported as a teammate's. A read that started before a\n write no longer puts the old value back. Undo refused \"alone\" after 1.234 in a column of\n two decimals goes through.\n- **SvGrid Base: undo and redo run no automation.** A redo of a status change added its\n checklist items again on every press.\n- **SvGrid Base: a second restorer is told nothing is wrong.** Two people restoring the same\n record at once: the second sees it back, not \"Another record already has this value\".\n- **SvGrid Base: wrong words on the history.** A column resize read \"Undo cell edit\", and a\n redo was logged as \"undid\".\n- **SvGrid Base: soft delete keeps a record's junction links.** A delete of a record marked\n with `softDelete` dropped its junction rows and a restore did not put them back, so tags\n were gone after a reload. Junction rows now stay, at either end of the link, and a restore\n writes nothing. A marked record shows in no link. Delete forever removes them with the row.\n- **SvGrid Base: Delete forever is final, and only once the database says so.** A purge the\n database refused left the record marked with no way to reach it, under a notice saying it\n was in the table. The item now stays in the trash until the purge is confirmed, the notice\n says so, and in one Empty trash the free records still go. Ctrl+Z could put a purged record\n back from the page's copy: the undo step now ends, and a stale undo in another window\n refuses. A single Delete forever now asks first, as Empty trash does.\n- **SvGrid Base: the trash of a soft delete table reads the table.** Marked rows show after a\n reload with no trash store, for whoever may delete. **Behaviour change:** those rows appear\n in the trash, those marked by other tools too. They do not age out: `trashDays` still only\n purges what the page's trash holds.\n- **SvGrid Base: `softDelete` on `createBaseEndpoint`, `baseFromDatabase` and\n `baseFromSupabase`.** The endpoint could not use soft delete at all. The column stays out\n of the fields, and the description tells the page.\n- **SvGrid Base: the mark cannot be written.** With the column on the schema a client could\n create a record already deleted, or mark one through an update. `createSqlDataSource`,\n `createSupabaseDataSource` and `createKitHandlers` now leave it out of every write.\n- **SvGrid Base: teammates hear of soft deletes.** A restored record reached a table read a\n part at a time only on a reload-style feed, and a delete through `createKitDataSource` stayed\n on teammates' screens. `createKitHandlers` now says which column marks records and the page\n learns it. `apply({ type: 'update', id })` reads that one record, which is how a feed reports\n a junction change; `baseFromSupabase` listens to junction tables by itself.\n- **SvGrid Base: a new record does not take a marked record's number.** On a table the page\n holds it did, and the restored record was renumbered. Autonumber now reads past marked rows\n on Kit and Supabase too.\n- **SvGrid Base: a very large soft delete purges a block at a time.** Closing the page used to\n leave everything not yet purged marked.\n- **`createKitHandlers`: restore and purge tell the server what happened.** **Behaviour\n change:** the `audit` hook now gets `restore` and `purge` actions, one entry per record that\n really changed, in place of `update` and `delete` with ids joined by commas, and a purge of\n another tenant's id logs nothing. `beforeDelete` and `afterDelete` run on a purge with\n `purge: true`, and `beforeRestore` and `afterRestore` are new. On a table with no `softDelete`\n `restore` and `purge` answer 405.\n- **SvGrid Base: formulas and lookups through `createBaseEndpoint` run in the database.**\n A sort on a formula read the whole table into the page, because the server's source\n was never told the base's fields. It now works from the fields the server holds (the\n `svgrid:` comments, or the new `computed` option for MySQL and SQLite) and says which it\n works out in `serverComputes`. A request names a field and brings no definition. A link\n is followed only for a person who may read the other table, and not into a table\n `scope` covers. A field changed in the page goes back to being worked out in the page.\n- **SvGrid Base: a sort, group or negative filter on a lookup reads the other table once.**\n Sorts and groups use a `LEFT JOIN`, and \"does not contain\", \"is not\", \"is empty\", \"is\n none of\" and \"not equal\" a `NOT EXISTS`, in place of a subquery per record. A record with no\n link, or a link to a record marked deleted, matches as it did.\n- **SvGrid Base: a view counts once per query.** A block read asks for rows alone once the\n view has its count. The count is read again after a write through the page, a change\n from a feed, a block that does not fit it, or 30 seconds. With no feed, a teammate's new\n record reaches the count in that time, at the next read.\n- **SvGrid Base: blocks read their keys first.** A read with text copies or a join finds the\n block's keys, then reads those rows, so a sort or an `OFFSET` no longer carries whole rows.\n- **SvGrid Base: a date condition compares the column to a range** on Postgres and MySQL,\n so an index on it serves a calendar and a filter. `DATE(col)` and `col::date` could not\n use one. SQLite keeps `DATE()`, which reads the offsets in its text.\n- **SvGrid Base: a read nobody waits for stops.** `execute` gets `{ signal }` on reads, no\n further statement of a stopped read is sent, and `createKitHandlers` passes\n `request.signal` on.\n- **SvGrid Base: a chart on dates is one grouped statement**, and group charts, board lanes\n and calendar days no longer run a count they do not use. A dashboard reads its tiles\n once as it opens, and tiles on one query share its count.\n- **SvGrid Base: a view switch shows no old count or rows** under the new view's name.\n- **SvGrid Base: a bulk delete reads by key** in place of a growing `OFFSET`, with a count at\n each end of the read.\n- **SvGrid Base on MySQL: a descending sort reads an index backwards.** Its tie-break on the\n key follows the sort's direction there, so records with the same date now come newest\n key first. `date DESC, id ASC` sorted every row.\n- **`baseFromDatabase` and `createBaseEndpoint` open a MySQL database.** The column read used\n `generated`, a reserved word, as a bare alias, and MySQL refused the statement.\n- **`createBaseEndpoint`: the first description no longer groups a unique column.** A column\n that shows more than 12 different values in a sample of 5,000 rows cannot be a select.\n\n- **Server row model: a drag of the scrollbar reads where it stops.** A jump of more\n than a block waits for the viewport to stand still (`leapPauseMs`, default 60), a\n read the viewport has left stops through `request.signal`, and the rows on screen\n go first in the queue. Placeholder rows keep their row number and line up with the\n columns.\n- **SvGrid Base: Gallery, List, Inbox, Board lanes and the Timeline read ahead of a\n scroll,** further the faster it goes and the slower the source, and a drag of the\n scrollbar reads where it stops.\n- **SvGrid Base: a typed column or view filter waits for a pause,** as the search box\n does; a pick from a list applies at once.\n- **SvGrid Base: a busy calendar asks for its days' sizes in one grouped request.** It\n skips empty days, and after a write reads only the days whose size moved. A record\n with an end date, or a backend that does not group, stays a read per day.\n- **SvGrid Base: reading a table through in the page keeps 4 requests open** past the\n first blocks, instead of waiting for one at a time.\n- **SvGrid Base: a Calendar and a Timeline no longer stay in memory after a view\n switch** on a Svelte older than 5.57.2 by these two routes. Svelte 5.55 to 5.57.1\n keep other parts of a destroyed grid linked as well: use 5.57.2 or later.\n- **SvGrid Base: every record action from the keyboard and on touch.** The Menu key\n and Shift+F10 open the record menu with focus in it, the row number has an actions\n button on a touch screen, a long press opens the menu, and the record view can\n delete its record. The field menu and the table menus take focus, walk with the\n arrows and give focus back. A field moves from its menu or from Hide fields.\n- **SvGrid Base: the rows are one tab stop.** Tab goes on to the header and footer\n controls; it walks cells after a click or in an open editor, and says the way out.\n- **SvGrid Base: forms have real inputs.** Typed fields are labelled inputs with\n required and invalid states and autofill; a failed submit goes to the first error\n and reads it out.\n- **SvGrid Base: what changes is said aloud.** Edits, clears, pastes, undo and redo\n reach a screen reader, and a calendar event moves only once picked up with Space.\n- **SvGrid Base: contrast.** Text on a filled accent control now follows\n `--sg-on-accent`, and accent-colored text in the chrome is plain ink. Text on a colored\n pill or bar stays white on saturated colors, by design; the VPAT gives the ratios.\n- **SvGrid Base: forced colors, right to left, translation.** Switches, the active\n tab and view and the \"+\" keep their state under forced colors. Arrow keys and the\n pin line follow `dir=\"rtl\"`. `setBaseMessages` now reaches the grid's own menus,\n find bar and labels (`grid.*`) and the calendar (`viewbar.calendar.*`).\n- **SvGrid Base: smaller fixes.** Focus stays in the grid after a delete, panels hold\n Tab, Search moves into the find bar, targets are 24 px, an IME composition keeps\n its Enter, and a calendar on a phone keeps its title on one line.\n- **SvGrid Base: a select option has an identity.** Renaming one in the field editor renames it on every\n record, in a multi select too, keeps its cards in their lane, and reaches the view filters, condition colors,\n form logic, dashboard tiles, automations and the field's default that name it. Removing one says how many\n records hold it, next to Save, and clears them. Undo takes it all back. On a table in a database it is a write\n to each record holding it, a block at a time. A card whose value is no option goes to Uncategorized, not off\n the board.\n- **SvGrid Base: views show the data.** A board, calendar, timeline and gallery follow an undo and a record edit\n at once. Calendars and timelines leave undo to the base, so a drag is one Ctrl+Z and not two that cancel out.\n- **SvGrid Base: sort, filter and group apply again after an edit**, holding a grid row only while it is the\n active row, and the count follows. An undo reaches a record the view no longer shows.\n- **SvGrid Base: the bulk bar is part of a base.** Ticking records shows it, with Delete, in every template,\n and never the developer note. `selectionBar={false}` turns it off.\n- **SvGrid Base: a formula reads names.** `{Account}` and `{Owner}` give what the cell shows, not record and\n user ids.\n- **SvGrid Base: deleting a field says what it drives**, by view and automation. A view asks in its own row, goes to\n the trash, and is one step of undo. A field deleted from a view no longer leaves it saying \"Colored by field\".\n- **SvGrid Base: columns resize and move by drag** by default, and a restored field goes back to its place.\n- **SvGrid Base: Hide fields keeps its rows still** while fields are switched.\n- **SvGrid Base: the warning for a type change sits next to Save**, in sight.\n- **SvGrid Base: comments can be edited and deleted by their author.** `update` and `remove` on a comment store.\n- **SvGrid Base: activity stays true.** An undo and a redo are noted, and a local edit shows old to new.\n- **SvGrid Base: Timeline is in Create in every base**, its renderer loading on demand, and the roadmap template's\n Timeline is one.\n- **SvGrid Base: a form builder.** \"Customize form\" sets the title, the words and the fields, their order and which\n are required. An empty form does not send.\n- **SvGrid Base: Space on a group row opens nothing**, where it opened an empty record that lost what was typed.\n- **SvGrid Base: smaller things.** The link picker takes arrows and Enter and keeps its size, and a second click no\n longer closes it. A link cell selects on the first click. One click sets a rating in a form. Next and Previous on a\n card walk the board. Find and the field type search take the focus. A\n header opens its menu on a right click. New groups start open. The row height menu marks the current height. A\n multi-day event's tooltip says its days.\n- **The schema is the whitelist, for reads and writes (breaking).** `createSqlDataSource` names\n the schema's columns in every `SELECT` and `RETURNING`, and drops every key off the schema\n from a write. A group level off the schema is refused with `leafCounts` too. A trusted\n source in the same process as the page can ask for the old behaviour with\n `wholeRecords: true`; `createKitHandlers` refuses to start on one.\n- **Nothing a client sends reaches a private key.** `createKitHandlers` strips `__svgrid*` keys,\n keys off the schema and `readonly` fields before `authorize`, `validate`, hooks and the\n source see the row, so a write can no longer cross a `scope` or swap a validated value. A\n write left with nothing is a 422. A `scope` field off the schema refuses the request.\n- **Handlers no longer send driver text (breaking).** A failure is logged on the server and the\n client gets a plain message. A record that is gone is a 404 with `code: 'row-not-found'`,\n which `createKitDataSource` keeps on its error. A request off the schema is a 400 that says so.\n- **`createBaseEndpoint` starts closed (breaking).** It refuses to start without `authorize` or\n `public: true`, and without `tables` or `allTables: true`, with a message that says what to do.\n `GET` asks `authorize` and describes only the tables it allows. A table `authorize` gives no\n guard for is closed. It takes `scope`. Columns named like a credential stay out of a base\n (`hide`, `isSecretColumn`): never the primary field, never sampled as options, and through the\n endpoint not read or written. A field with `hidden: true` is no field of a base.\n- **A record's text cannot name a secret.** A webhook fills its URL, headers and body in one pass,\n so `{secret:DATABASE_URL}` typed into a field goes out as that text.\n- **Server webhooks stay off internal addresses (breaking).** `createAutomationServer` refuses\n private, local and link-local addresses, by name and by what the name resolves to, and does not\n follow redirects. `allowInternalWebhooks: true` opts out.\n- **A PDF preview runs nothing.** A file is read first and framed as a copy typed by us, only if\n its bytes are a PDF. An HTML file calling itself a PDF offers a download.\n- **Hidden fields and tables stay out of search, Download as SQL and the row number tooltip.**\n `lockViews={false}` on `SvBase` is honoured with `tables`. `permittedRecords` is what to send\n from a server; `permittedTables` keeps hidden values in its records, and the docs no longer\n point at it for that.\n- **A CSV is read for what it is.** The separator and encoding are found and shown to change. A\n quote inside a word is text, and a quote never closed no longer swallows the file. Rows that do\n not fit the header, values a field cannot hold, records that repeat a name, a file that is not\n UTF-8 and one that is not text are all reported before the import. A decimal comma stays a\n decimal, and `1,50` is no longer read as 150. `readImportMatrix` returns `delimiter`,\n `encoding` and `warnings`. Errors from a bad .xlsx or JSON file read as sentences.\n- **Large tables.** Import, summaries, dashboards, chart axes, rollups, pivots and spreadsheet\n ranges no longer throw `Maximum call stack size exceeded` past about 120,000 values (`maxOf`,\n `minOf` in place of `Math.max(...values)`).\n- **Download a view.** The list of tables has CSV and Excel downloads of the shown view: its\n fields and records as they read on screen, formulas, links, people, checklists and attachments\n included. `recordsAsShown` gives the same in code.\n- **Smaller holes.** A protected field is not asked in a form of someone not on its list. A paste\n outside a select's options is not written. Rich text in a base keeps no `class`, and an image on\n another site becomes a link instead of a request. A color from data that is more than a color\n is dropped. JSON import no longer fills `constructor`-like columns from the prototype, and a key\n first seen late in a file is a column.\n- **SvGrid Base: opening a cell and leaving it writes nothing.** A number, bigint,\n time, timestamp, date or JSON value used to be rewritten as the editor showed it.\n- **SvGrid Base: a number is stored as typed.** 19.99 in a field that shows one\n decimal stays 19.99; only an integer field rounds. A number past 15 significant\n digits stays exact as text, in the cell, the box and the write. A rollup with a\n `precision` keeps the linked field's symbol.\n- **SvGrid Base: dates and times no longer change with the browser's zone.** Dates,\n times, bigints and decimals are read as the database writes them, so a naive\n timestamp shows one wall clock everywhere, a typed time keeps it, and an undo or\n restore writes back what was read. On PostgREST a bigint past 2^53 is read again as\n text.\n- **SvGrid Base: dates read as people type them.** `12/24/2026`, `2026-12-23`,\n `Dec 22, 2026` and `12/21` all work in the new date editor (`fieldDate`), and text\n that is no date says so.\n- **SvGrid Base: cell keys.** Enter on a selected cell opens its editor, a key typed on\n a date or link cell starts editing, and Space ticks a checkbox cell.\n- **SvGrid Base: JSON in a text field shows as JSON** instead of `[object Object]`, and\n the Delete key leaves NULL as NULL and empties a number or date column without an error.\n- **SvGrid Base: a record added or deleted survives a view switch.** On a table the page\n holds, the grid went back to its `data` on any option change, so records added (by the\n row, Duplicate, a Kanban card, a form) vanished and deleted ones came back. What the\n grid holds is now what it is handed.\n- **SvGrid Base: Undo and Redo leave a teammate's cell alone.** A cell changed since keeps\n the later value and the page says so, cell by cell in a paste. Without a live feed the\n page reads the record first.\n- **SvGrid Base: a record's way back is whole, and safe.** Undo and restore on a table the\n page holds write every column read, under the record's own key, and the record stays in\n the trash until the database has it; a failed undo can be pressed again. A restored\n record whose Autonumber was given out again takes the next one.\n- **SvGrid Base: the trash gets the record as it is at the delete.** A delete reads the\n records at that moment, also after a confirm dialog, and a record moved out of a bulk\n choice while the question stood stays.\n- **SvGrid Base: typing is safe from teammates.** Changes from others, by feed or by a read,\n wait while a cell editor is open, so what is typed is not cut short. If the record went\n meanwhile, the person is told, with their text.\n- **SvGrid Base: a delete that did not happen is not reported as done.** One a row-level\n policy skipped, or the database refused, stays in the table and out of the trash, and the\n notice comes after the delete, not before.\n- **SvGrid Base: failures a person can act on.** A failed write names the record and the\n field in plain words, shows what was entered and offers Try again; \"Not saved\" stays until\n that record is written. A write with no answer warns and then lets the rest go\n (`writeTimeout`).\n- **SvGrid Base: Ctrl+Z of a new record on `NOT NULL DEFAULT ''`.** An emptied cell goes back\n to what the database gave it, not to NULL.\n- **SvGrid Base: the trash tells the truth.** With nothing keeping it, the panel says items go\n with the page.\n- **Supabase: the find box takes `%`, `_`, commas, quotes and brackets as text**, and a table\n with text keys can add records with default options.\n- **SvGrid Base, smaller things.** The selection bar counts on a table the page holds, a draft\n shows no number until it has one, and a write coming back in the database's words (`5.00`\n for 5) is nobody else's change.\n- **SvGrid Base: a group on a date opens on its records.** On Postgres and MySQL the\n key of a date field's group went through a JS Date, so it lost what is past the\n millisecond and took the page's zone, and outside UTC the group read nothing. The key\n is now text the database writes and compares. A delete can also leave out any group\n at any level: every source reports a group's records as `leafCount` when a base asks.\n- **SvGrid Base: two people adding at once do not share an Autonumber.** They could get the\n same number. Each record is checked once stored and the later one moves on; a UNIQUE\n column refuses the second and the write tries the next number.\n- **SvGrid Base: Undo of an added record keeps a teammate's work.** One someone has\n changed since stays, with a message, and Redo does not add it twice. Redo of a delete\n puts the records in the trash as they are now, not as they were.\n- **SvGrid Base: Undo brings times back exactly.** A delete reads time columns as the\n database writes them, so microseconds and a local time the clocks skipped survive on\n every Postgres and MySQL driver. `dateStrings` is no longer needed.\n- **SvGrid Base: Undo works on a server in any language.** Generated columns are read\n from the schema and left out before a write, and a refusal is told by its error code.\n `createSupabaseDataSource` keeps the code on the errors it throws.\n- **SvGrid Base: Undo leaves other people's work alone.** Undoing a step on a table held\n in the page put back a snapshot of all its records, which took out whatever teammates\n had changed since, and wrote that to the database. A step now undoes what it did and\n nothing else, and a new record closed empty takes only itself out.\n- **`createSqlDataSource` on MySQL.** Writes ended in `RETURNING`, which MySQL does\n not have, unless you set `returning: false`. That is now the default there. A\n record with an auto-increment key now gets the key the insert made.\n- **Formula text in SQL.** A string in a formula could break out of its literal on\n MySQL through a backslash, and a pivot key from stored data the same way. Both are\n closed, on every dialect.\n- **Formulas in SQL say what a cell says, or leave it to the page.** Functions whose\n results differ between a database and a cell (`UPPER`, `FLOOR`, `TODAY` and others,\n listed in the docs) no longer run in SQL.\n- **`createSqlDataSource`:** a filter on a column the plan does not know is an error\n and not a wider answer; `in` with an empty list matches nothing; `notEquals`,\n `notContains`, `endsWith`, `isNotBlank` and `notIn` run as themselves and no longer\n as `contains`; a blank group key matches NULL and ''; `updateRow` on a record that\n is gone is an error with `code: 'row-not-found'`; `createRows` keeps a column's\n default. A filter or sort on a field added at run time works.\n- **SvGrid Base:** a write the database refused no longer reads as saved after a later\n edit; records in the trash leave it only once they are proven back, whole, with the\n columns the base shows no field for and under the key they had; times go back as they\n came on MySQL and on Postgres; junction links come back with a restored record and\n survive a refused delete; a table whose records have no id of their own no longer\n breaks the grid or the trash; writes still queued\n when the page leaves finish, and the browser asks first; a form says so when the\n database refused the record; a field named ID no longer writes over the record's\n key; a formula edited on a big table runs at once, in the database too.\n- **Scrolling a very long list.** In a gallery, list, board lane or timeline past the\n height a browser scrolls, the scrollbar and the list stopped agreeing near both\n ends. A scrollbar step there now moves the list by the same step.\n- **SvGrid Base: a table of a million records scrolls without stalls.** With\n `serverRows`, the view bar counted records by reading every row of the\n server row model, each one made on read, on every block that arrived; the\n record view looked for its record the same way. The count now comes from\n the server (`rowCount`, and the top groups' counts when grouped), and the\n record view looks near the row it was opened from first. Number, currency\n and date cells also keep their `Intl` formatters instead of building one\n per cell. In demo 723 (a dev build, one laptop), 40 wheel steps went from 9.3 s of blocked main\n thread to about 0.1 s, and opening the filtered view from 6.5 s to 0.5 s.\n\n- **SvGrid Base: switching views on a big table no longer freezes the page.**\n A grouped view on `serverRows` set a row height per row and opened its\n groups, and both read every row of the row model; now it keeps one height\n and leaves server groups closed, since each opens with a request. A board\n of more than 200 records draws only the cards in sight in each lane (the\n board measures a card when `cardHeight` is not set). The board also keeps\n each card's key instead of working it out on every read. In demo 723 (a dev\n build, one laptop), the slowest switch went from 3.2 s to 0.4 s.\n\n- **SvGrid Base: a view that left out part of its state kept the last view's.**\n A view declared with only `grouping` took over the sort, hidden fields and\n other state of the view before it. Whatever a view's `state` leaves out\n now starts plain. A saved view is unchanged: it stores all of it.\n\n- **SvGrid Gantt: the task table's header stays at the top.** It scrolled away\n with the rows in any plan taller than the chart.\n\n- **SvGrid Base: a record past what a table read was taken for a deleted one.**\n When a table on a `source` was read again at its limit, a record that had\n moved past the limit was dropped from the page and reported as deleted.\n A read at the limit now keeps the records it did not see.\n\n- **A pivot filtered its rows again for every column.** For each row group,\n `createPivotModel` went through the group's rows once per column-axis leaf.\n The rows are now bucketed by column path in one pass, in the same order,\n and the result is the same. On one laptop, a pivot of 100,000 rows over\n Year x Month went from 3.6 s to 0.6 s, and 1,000,000 rows from 147 s to\n 4 s.\n- **A grouped Excel export slowed down with the square of its rows.** The\n writer looked up each string cell's shared-string index, and checked the\n group rows already written, by scanning an array. Both are hash lookups\n now and the file is byte for byte the same. A 10,000-row export grouped by\n five columns went from 4.8 s to 0.45 s, and 100,000 rows finish in 9 s.\n- **A sheet chart hung below the window was not drawn until a scroll.** The\n sheet measured its objects once after mount, from their anchor cells, and\n the grid now adds most of its overscan rows a frame after its first paint;\n a chart hung from a row in between had no cell to measure. While an object\n has no box, rows the grid adds now measure it again.\n\n- **A wide grid lost columns in the PDF.** pdfmake sizes a column no narrower\n than its longest word, so a 22-column report ran off the right edge and only\n 15 columns printed, without a warning. The columns are now fitted to the\n page (see `pdf.fit` above).\n- **A row taller than a page was missing from the PDF.** Rows are kept whole\n on a page, and pdfmake drops a row that cannot fit on any page. Such a row\n now breaks across pages.\n- **Conditional-format icons printed as empty boxes in the PDF.** The arrow,\n triangle and traffic-light glyphs are not in Roboto; they are drawn as\n shapes now.\n- **The PDF footer could show yesterday's date.** It printed the UTC date, so\n an export after midnight east of Greenwich was dated the day before. It\n prints the local date, and the footer follows custom page margins.\n- **Grouped PDFs dropped conditional formatting and links.** They now carry\n both, as flat exports do.\n- **Booleans printed as \"true\" and \"false\" in the PDF.** They print a tick or\n nothing, like the grid's checkbox. An icon set's `iconOnly` is honoured too:\n it printed the value next to the icon.\n- **A number or date column passed in `columns` without `align` printed\n left-aligned in the PDF.** It right-aligns from its `format`, as on screen.\n- **A pivot chart drew an empty cell as a zero bar.** An average over no rows\n is `null` in the pivot and the bridge read it as 0; it is a gap on the\n chart now. A sum over no rows stays the 0 the engine wrote.\n\n- **PDF export depended on pdfmake's fonts registering themselves.**\n `pdfmake/build/vfs_fonts` registers its font map only as a side effect of\n being imported while a global `pdfMake` already exists, and the export's\n own lookup for the map never matched the shape the module has shipped since\n pdfmake 0.2.8 (the module is the map), so it registered nothing and relied\n on that side effect. Under a bundler the global can be a stale instance (a\n dev server re-optimising its dependencies mid-session is enough), and the\n export then failed with \"File 'Roboto-Medium.ttf' not found in virtual file\n system\" while its promise never settled. The map is now found in every\n shape the module has had (`resolvePdfVfs`) and handed to the instance that\n creates the document (`registerPdfFonts`, through `addVirtualFileSystem`\n and the `vfs` property), and an installed pdfmake whose fonts module has\n no map is reported instead of failing later.\n- **Studio emitted no theme for the fragment export and the CLI `add`\n scaffolds.** Only the full app's `+layout.svelte` carried the `--sg-*` tokens;\n `eject --fragment` (which drops that file) shipped an `app.css` with one\n hard-coded accent, and `svgrid-studio add` / `add --all` wrote none at all. The\n tokens now come from one helper (`themeTokenCss`): `src/app.css` carries them\n for the full app and the fragment, and the `add` scaffolds emit them in\n `<svelte:head>`. Drop-in outputs wrap them in `@layer svgrid-studio`, so a host\n app that already defines its own tokens keeps its look. `add` also honours\n `--theme` / `--dark`, and a project with no theme picked gets Ember (the demo\n theme) instead of Tailwind.\n- **Scheduler column alignment.** The all-day lane, day/time-grid header, and\n body columns now reserve the scrollbar width consistently, so the columns line\n up instead of drifting by the scrollbar's width.\n- **Excel export color-scale conditional formatting.** A color scale's `min` /\n `mid` / `max` are optional, so the export could build an `(string | undefined)[]`\n where a `string[]` was expected; it now falls back to the same `#ffffff` /\n `#000000` defaults the grid renders with. Because `@svgrid/enterprise` ships its\n `src`, this had surfaced as a stray `svelte-check` error in consuming apps - a\n clean app now type-checks with zero errors.\n\n#### Security\n\n- **SvGrid Base: a pair table with the tenant in its key could be read and changed by another tenant.**\n On `order_tags (tenant, order_id, tag_id)` with `tenantColumn` and a scope, the pair id a client sent\n overwrote the scope's tenant, so an editor of tenant B read tenant A's pairs by id and deleted them\n with `delete` and `deleteMany`, on Postgres, MySQL and SQL Server. The scope now wins: an id naming\n another tenant's pair reads nothing, and a delete of it is a `403`, as on a `{ through }` pair.\n- **SvGrid Base: a `{ through }` pair read could hand out another tenant's key.** A pair written\n straight into the database between tenant A's order and tenant B's tag reached A in raw rows. A read\n of a junction (a pair table, or one with its own key) now leaves out a row whose other end is a record\n of another tenant, where that end's table holds its own tenant column. Delete forever still clears\n such a row.\n\n- **A crafted .xlsx could crash the tab.** `importData` and the spreadsheet's\n File > Open filled every gap before a cell with blanks, sized by the cell\n reference in the file: `<c r=\"ZZZZZZZ1\">` asked for about eight billion.\n `importData` now rejects a cell past column XFD; the spreadsheet skips a cell\n outside the sheet, as it does other malformed parts. Both refuse a sheet that\n would need more than ten million blank cells to lay out.\n- **A zip bomb could exhaust the tab's memory.** Every .xlsx and .ods reader\n (`importData`, File > Open, `documentFromXlsx`, `documentFromOds`) unzipped\n each part whole before looking at it, so a small file whose parts inflate to\n gigabytes was read into memory first. Parts are now unzipped as a stream and\n counted as they inflate; a file whose parts pass 256 MB together is refused\n at that point.\n- **CSV and TSV from `exportData` neutralise formula-like text**, through the\n same `@svgrid/grid` serializer (see the grid's Security entry).\n\n### @svgrid/grid-wc\n\n#### Added\n\n- **`<sv-chart>`**, the standalone chart as a custom element, at\n `@svgrid/grid-wc/chart` with React (`/react/chart`), Vue (`/vue/chart`)\n and Angular (`SvChartComponent`) wrappers. Its surface is generated from\n `SvChart`'s own props type the way `<sv-grid>`'s is from the grid's, so the\n 26 props and 7 events stay in step; `spec` is a property, the primitives\n are attributes, and a prop that takes a boolean or a string (`legend`) keeps\n a string attribute where a bare attribute means true. Its own build in\n `dist/chart` carries none of the grid.\n- **Typed event handlers.** Each wrapper's `on<Event>` prop used to be\n `(detail: unknown) => void`, so reading `newValue` meant casting first -\n a poor advertisement for a typed wrapper. The detail type is now lifted from\n the grid's own callback signature, so 16 of the 19 events arrive fully typed\n (`TData` degrades to `Record<string, unknown>`; anything referring to a\n type that only exists inside `@svgrid/grid` stays `unknown` rather than\n naming something a consumer cannot resolve).\n\n- **React, Vue and Angular components**, as subpath imports:\n `@svgrid/grid-wc/react`, `/vue` and `/angular`. Each is generated from the\n same surface as the elements, so all 98 properties and 20 events are typed\n props on every one of them and none can drift from the grid. Each framework is\n an OPTIONAL peer dependency, so a plain-HTML consumer installs none of them,\n and the wrappers are 1.3-3.7 KB gzip because they reuse the one element bundle\n rather than shipping a second copy of the grid.\n\n They exist because a raw custom element is genuinely awkward in each:\n **React 18 and earlier stringify object props onto attributes**, so\n `columns={cols}` silently becomes `\"[object Object]\"` and the grid renders\n empty; **Vue** needs `isCustomElement` build config and a `.prop` modifier\n on every object binding; **Angular** needs `CUSTOM_ELEMENTS_SCHEMA` in every\n component that shows a grid, and gets no typed inputs. The Angular wrapper is\n compiled with ng-packagr in partial-Ivy mode, so it is consumable by a normal\n Angular build, and its selector is the element's own tag - `<sv-grid>` and\n `<sv-grid-shadow>` - with no template of its own, so there is no extra wrapper\n element in the DOM and the tag is the same one you write in plain HTML.\n\n All three also handle two ordering problems a hand-written wrapper meets:\n the element renders BEFORE a framework assigns properties in an effect, and\n `apiready` fires once during that first mount - before React can bind a\n listener at all. The handle is parked on the element and each wrapper replays\n it.\n- **The package ships TypeScript declarations.** It previously had none, so\n `import '@svgrid/grid-wc'` was an error under `moduleResolution: bundler`\n and `document.querySelector('sv-grid')` came back as a bare `Element`. The\n generated `.d.ts` augments `HTMLElementTagNameMap` and\n `HTMLElementEventMap`, so the element and its events are typed without\n hand-written declarations.\n- **`<sv-grid-shadow>` - the grid in an open shadow root**, so a host page's\n CSS cannot reach it. Same properties, attributes and events as `<sv-grid>`;\n a separate element rather than an attribute because Svelte resolves\n `customElement.shadow` at compile time, so no runtime flag can switch it.\n Isolation is one-directional and the docs say so: page CSS stops at the\n boundary, but the grid's own stylesheet is still injected into the document,\n because about twenty overlay surfaces portal to `document.body` on purpose\n to escape ancestor clipping. `--sg-*` theme tokens reach in unchanged,\n including a theme file's dark variant - custom properties are inherited, and\n inheritance crosses a shadow boundary.\n- **The elements now expose the whole grid.** `<sv-grid>` declared **7 props\n and 2 events** by hand against a `Props` type with 100 props and 19\n callbacks, so grouping, pagination, pinning, tree data, master/detail, board,\n scheduler and every Enterprise feature were unreachable from a non-Svelte\n host - while the docs said they \"all come along\". The surface is now\n **generated** from `<SvGrid>`'s own types: **98 properties** (72 with\n kebab-case attributes, 26 property-only) and **20 events**, with CI failing if\n the two drift. Costs 1.6 KiB gzip.\n- **Every grid callback is a DOM `CustomEvent`**, `detail` being the\n callback's argument - or, for the one callback taking two, an object keyed by\n its parameter names. `onApiReady` also parks the imperative handle on the\n element as `el.api`, because an event fires once and a host binding a\n listener later would otherwise never reach it.\n\n#### Fixed\n\n- **Importing the package on a server threw `customElements is not defined`.**\n Svelte registers each element with a bare top-level\n `customElements.define(...)`, so any server-rendered React app\n (TanStack Start, which Lovable now generates, Next.js, Remix) crashed on\n `import { SvGrid } from '@svgrid/grid-wc/react'`. The build now guards the\n registration; on a server the import is inert and SSR emits a plain\n `<sv-grid>` that upgrades on hydration. Guarded by `test/ssr.test.ts`.\n- **The React wrapper lost props the grid reads once at mount.** It wrote\n props in a passive `useEffect`, which runs after paint; the element mounts a\n microtask after insertion, so it had already mounted with defaults and\n `pageSize={25}` showed 10 rows a page (the wrapper's own pagination example\n did). Props are now written in a layout effect, before the element mounts.\n The same change in the `<sv-chart>` wrapper.\n- **`style` on the React wrappers took `Record<string, string | number>`,** so\n passing a React `CSSProperties` object failed to type-check. It is\n `CSSProperties` now.\n- **The React wrapper re-did all its work on every parent render.** Both of its\n effects had no dependency array, so any state change anywhere in the parent\n reassigned all 98 properties and rebound all 20 event listeners - 40 listener\n mutations per keystroke. It now writes only values that actually changed and\n binds listeners once through a handler ref. Measured: 20 no-op re-renders\n produce 0 listener adds, 0 removes and 0 property writes.\n\n- **The element threw when it rendered before its props were assigned.**\n Generating the prop list replaced the elements' own `data = []` /\n `columns = []` defaults with a bare `$props()`, so a consumer that assigns\n properties AFTER the element upgrades - which is what React and Angular both\n do, in an effect - got\n `TypeError: Cannot read properties of undefined (reading 'map')` and a grid\n that never rendered. Only a real framework app was late enough to reproduce\n it; an HTML fixture sets the properties in the same tick and never sees it.\n\n#### Changed\n\n- **Only arrays, objects and functions changed shape: they are property-only.**\n An HTML attribute is a string, so `columns='[object Object]'` cannot work\n and no attribute is offered for the 26 props that are not primitives. The\n published `rowclick` and `selectionchange` events keep the exact\n `detail` they shipped with, and `selectable` keeps meaning row-selection\n checkboxes - `<SvGrid>`'s prop of that name is an alias of\n `enableCellSelection`, and forwarding it would have silently repointed a\n published attribute at a different feature.\n- **Styles arrive with the code that needs them.** The build used to put every\n stylesheet in the bundle, lazy chunks included, into the entry, so a page\n that never opened the chart, a date picker or a menu still downloaded their\n CSS. Each chunk now carries its own styles and adds them to the document\n and to every open shadow root when it loads. `<sv-grid>`'s entry went from\n 107 KiB to 98.5 KiB gzip while the chart grew.\n\n### Tooling & docs\n\n#### Added\n\n- **Every chart gallery demo switches between Chart and Grid.** A segmented\n control first in each demo's toolbar flips every pane to a sortable grid\n of the rows the chart was drawn from, and back, the way the board demos\n switch Board and Table. The from-the-grid page shows the pattern in a\n runnable.\n\n- **Charting a pivot is documented.** The pivot page has a section on\n `pivotToChartSpec` (the mapping, every option, a runnable that pivots\n orders and charts them by quarter) and on the designer's Chart view\n (`chartable`, `defaultView`); the charts hub and the from-the-grid page\n link to it, the /api pivot section lists the function, and the generated\n reference covers `pivot-chart.ts`. Demo 359's data now puts every\n country in every quarter and category, so its groups are four bars, not\n one.\n\n- **27 runnable framework examples**, nine each for React, Vue and Angular:\n a first grid, sorting and filtering, editing and saving, row selection,\n grouping and totals, pagination, server-side data, theming, and Excel export\n from the Enterprise pack. Every one has\n an **Open in StackBlitz** button that boots a full editable project - no local\n install - and a new\n [React, Vue or Angular](/docs/help/web-components/frameworks) landing page\n puts the thirty-second path first.\n\n They are real apps in the repository (`packages/grid-wc/examples/`), not\n snippets, and CI compiles all of them with each framework's own compiler -\n including `ngc --strictTemplates`, so a wrong Angular binding fails the\n build. The doc listings are generated from those files, so a page cannot\n describe a version of an example that no longer exists.\n\n The Vue templates get an extra check of their own: Vue treats an unknown\n attribute as a legal fallthrough, so `<SvGrid sortabel />` type-checks\n cleanly and silently does nothing. Verified - the same typo fails the React\n and Angular compilers and passes `vue-tsc` - so the Vue examples are also\n checked against the generated surface.\n\n- **A Web Components docs category.** Eight pages under\n `help/web-components/` - quick start, a generated `<sv-grid>` reference,\n shadow DOM, React, Vue, Angular, TypeScript and limitations - replacing a\n single page that was filed under \"Layout & Styling\". The reference tables are\n generated from the grid's types, and a guard checks that the counts quoted in\n prose still match the surface.\n- **The docs run.** A `svelte` code fence tagged `{runnable}` is extracted at build time\n into a real component, so a doc page shows the source and the working result\n together rather than highlighted text. 504 examples across the corpus, taking\n the median doc page from 1 example to 3; each one is compiled by `svelte-check`\n in CI, so an example that stops working fails the build instead of rotting in\n place.\n- **`@svgrid/ui` recipe scaffolder** - `npx @svgrid/ui add <component>` drops a\n ready-to-edit starter (e.g. `add calendar`) into your app, complementing the\n shadcn-style component pages (Preview / Code, install tabs) across the UI kit.\n- **See a component before you wire it in** (`@svgrid/ui` 0.3.x).\n `npx @svgrid/ui try <component>` spins up a zero-setup Vite + Svelte sandbox and\n opens the component in your browser; `add --preview` writes a\n `src/routes/preview/<id>` route (plus a `/preview` index) in a SvelteKit app.\n `add` now installs `@svgrid/grid` by default (`--no-install` to opt out) and\n prints the exact `try` command for whatever you just added, so the \"see it\"\n step is never a dead end.\n- **AI coding Skill** (`skills/svgrid`, installable with\n `npx skills add sv-grid/sv-grid`) - an always-on house-style guidance layer\n that complements `@svgrid/mcp`.\n- **The rest of the toolkit is findable.** `packages/grid/README.md` gained an\n \"Also in the box\" section, and the site footer gained a **Toolkit** column\n (SvGrid UI, Kanban board, Scheduler, Dock layout, Studio, Web components).\n The 84 SvGrid UI components have always been exported from `@svgrid/grid`\n itself - `SvButton`, `SvCalendar`, `SvDateTimePicker`, `SvDockLayout`,\n `SvToaster` and the rest are already installed for every grid user - but\n nothing on the highest-traffic surfaces said so.\n\n#### Changed\n\n- **Doc examples show code and result at once.** The runnable card used to be a\n Preview / Code toggle with an in-page editor; it is now the source above the\n live example, both visible, with Copy. A tab hides half of a teaching example\n and makes comparing the two a click, and dropping the editor took the\n on-demand TypeScript compiler chunk off every docs page with it.\n- **`npm create @svgrid` starts on Ember** (the demo theme) instead of Tailwind,\n for every template; `--theme <id>` still picks any preset. The `sveltekit`\n template is now listed on the starters page, and the `sv add @svgrid` demo route\n imports the Ember stylesheet so it no longer renders unthemed.\n\n#### Fixed\n\n- **The base scaffold lists `jszip`**, so \"Download this view as Excel\" works in a new app. Its README no\n longer says a new table appears on reload with the Postgres setup: that route reads the tables once\n per server process, so restart it or call `refresh()`.\n\n- **14 doc examples drew an empty or broken box under their code.** Every\n `{runnable}` block was run against the published packages in a clean Vite\n app. Five needed a SvelteKit load, an `/api` server or a WebSocket (two\n showed \"This example failed to load.\" on svgrid.com) and one is a component\n file that needs props; they are now `{norun}`, a new fence flag: still\n type-checked, never mounted, never promoted back to runnable. The other\n eight now run: the custom header, the real-time cell flash, the\n inline-snippet cell renderer, and the alert, loading-overlay, scroll-area,\n sparkline and toaster examples. The toaster page also said hidden toasts\n appear as room frees; their timers keep running while hidden, which the\n page now says.\n- **`build:lib` in the enterprise package deleted the node bundle.**\n `svelte-package` wipes `dist/`, so a library build alone removed\n `dist/node/studio.js` and the MCP server failed to start until\n `build:node` was run by hand. The library build rebuilds the bundle it\n removed.\n\n- **Docs images and tutorial media 404'd under a site base.** The markdown\n writes them root-absolute (`/docs-media/...`, `/tutorials/...`), which is\n right at svgrid.com and wrong under the `/sv-grid/` base the e2e server\n uses: every docs image and tutorial poster was missing there. The page\n prefixes the base in the HTML before it paints, and an e2e spec loads a\n page with both and checks nothing under those roots 404s.\n\n- **Twenty-one links on the chart pages went to GitHub 404s.** The split\n wrote every cross-link with one `../` too many, and the site's link\n resolver falls back to a GitHub URL for a page it does not know. Fixed on\n the pages, and every relative link under docs/ is now a test\n (`tools/docs-relative-links.test.ts`); the link checker existed but only\n ran by hand before a release.\n- **Links into the generated reference tree went to GitHub too.** The tree\n is hidden from the docs routes; a link to it lands on the /api page, on the\n module's section where it has one.\n- **The chart gallery said \"the source is under each example\" and it was\n not.** Each gallery card now carries the demo's own script and markup on a\n Code tab, written by `tools/gallery-doc-code.mjs` from the demo file and\n guarded by a drift test, so the code beside a preview is the code that\n renders it.\n- **A demo page flattened a bulleted banner into one paragraph**, dashes and\n backticks included. The page reads the pitch as blocks: a list where the\n banner wrote one, `<code>` where it wrote backticks, on the prerendered\n and the hydrated page alike.\n- **`demo-doc-embed` skipped a demo the page only linked to**, so a\n placement for it never landed, and it opened a new \"More examples\" section\n below \"See also\", which the coverage test requires to be last. Present\n means embedded, and the section opens above \"See also\".\n- **Regenerating the doc snippets took the dev server's file watcher with\n it.** The extractor removed and recreated its output directory, and on\n Windows the watch did not re-attach, so an edited runnable kept serving\n its old chunk while the source beside it showed the new code. It writes in\n place and removes only what no longer belongs.\n- **The Getting started chart page had no live example.** Its code blocks\n were `ts` and a Svelte block referencing rows it never declared; four of\n them run now, and the hub's first example runs with its own rows and a\n filter row, so the chart is seen following the grid.\n\n- **The e2e suite adopted any server on :5180.** Playwright reuses whatever\n answers on the port, and a Vite dev server returns the app shell for any\n path, so a website dev server started by hand (base `/`) passed the\n readiness check and the fourteen path-routed and module-importing specs\n failed with nothing pointing at the server. The config now probes the port\n first (`tests/e2e/lib/site-server-probe.mjs` reads the base off Vite's\n client script tag), reuses a server that serves the site under `/sv-grid/`,\n skips one that serves anything else with a line saying so, and starts its\n own on the next free port; `SVGRID_E2E_PORT` pins one.\n- **The website dev server re-optimised dependencies mid-session.** The\n search index, the docs renderer, the /api runner's compilers, the PDF\n exporter and the grid's `esm-env` sit behind dynamic imports the start-up\n scan does not reach, so the first page to touch each one triggered\n \"optimized dependencies changed. reloading\" on every open page. In the e2e\n suite that reload landed mid-test on a cold cache and failed whichever spec\n was on a page at the time. They are pre-bundled at start-up now\n (`optimizeDeps.include`), and `esm-env` is a declared devDependency of the\n website so it resolves from there.\n- **The /api runner check sampled an example's outcome two frames after\n Run.** An example awaiting an export or a fetch was still running, so a\n failure that arrived later, or one thrown inside a library callback the\n runner's try/catch cannot see, passed. The check now waits for the runner\n to report an outcome and counts an uncaught error or rejection during the\n run as a failure of that example.\n- **181 doc pages had content stranded below \"See also\".** Authoring passes\n appended new sections to the end of the file, which buried them under what\n reads as the page footer: on `help/filtering/number-filter`, for instance, the\n `between` example sat below the link list and a reader would never reach it.\n The footer is back at the bottom on every page, and a test now fails if a\n section lands under it.\n- **The documented way to lock one column's width never worked.**\n `docs/help/columns/column-sizing.md` matched headers on `data-col-id`, which is\n the attribute on *body* cells - the selector matched nothing. Header cells\n carry `data-svgrid-header-col`. (`ColumnDef.resizable` is now the real answer.)\n- **`blockKey` was missing from `tools/lib/md-snippets.d.mts`.** The function was\n exported from the `.mjs` and worked at runtime, so nothing caught it until the\n website was type-checked.\n- **`@svgrid/mcp` reports its real version.** The MCP server's\n `serverInfo.version` was hardcoded to `0.1.0`; it now reads the package version.\n- **`@svgrid/create-studio` scaffold builds out of the box.** The generated app\n shipped an orphaned `/products` route referencing an entity that was never\n defined; the `Product` schema, seed, data source, nav, and home card are now\n complete, so `npm run build` succeeds on a fresh scaffold.\n- **API-docs accuracy pass across all 74 UI-component pages.** Corrected the\n `SvDockLayout`, `SvDockManager`, and `SvDateTimePicker` prop tables (added 13\n missing dock-layout props; removed a `size` prop `SvDateTimePicker` never had),\n and filled scattered gaps (`loading`, `onClosed`, `min` / `max`, `dir`,\n `width` / `height`, `ariaLabel`, and the shared field-contract on\n `SvFileUpload`).\n\n### @svgrid/mcp\n\n#### Changed\n\n- **A cold `npx -y @svgrid/mcp` installs 20 packages instead of 111.** The\n server speaks MCP over stdio itself instead of through\n `@modelcontextprotocol/sdk`, whose HTTP and OAuth dependencies (Express, Hono,\n JOSE, AJV) a stdio server never runs. On an empty npm cache the install took\n 6 to 8 seconds instead of 12, which had left little room under an MCP client's\n 30 second connect timeout. The protocol is unchanged: version negotiation,\n errors and every tool, resource and prompt answer as before.\n\n## [1.0.0] - 2026-06-16\n\nInitial public release - the Svelte 5-native data grid: `@svgrid/grid`\n(MIT core) and `@svgrid/enterprise` (the paid feature pack).\n\n### @svgrid/grid\n\n- **New** `onCellValueChange` callback on `<SvGrid>`. Fires after every\n inline edit commits with `{ rowIndex, columnId, oldValue, newValue,\n row }`. The recommended hook for server-side persistence and cascading\n recomputes; see demo `18-cascade-editing` and\n [Saving values](./help/editing/saving-values.md).\n- **New** `externalSort` + `externalFilter` props. Grid records the UI\n state but does NOT re-order / filter rows - the consumer owns the\n pipeline. Paired with `onSortingChange` / `onFiltersChange` for\n server-side data and tree-data scenarios.\n- **New** `onSortingChange(sorting)` / `onFiltersChange(filters)`\n callbacks. Fire on every change with the consolidated payload shapes\n documented in the [SvGrid reference](./reference/SvGrid.md).\n- **New** `onRowSelectionChange(selection, rows)` and\n `onActiveCellChange(cell)` callbacks.\n- **New** `fitColumns` prop. Scales column widths to fill the viewport,\n with rounding-residue absorbed in the last scalable column. Shrinks\n down to 85 % of natural widths; beyond that a horizontal scrollbar\n appears.\n- **New** `showRowNumbers` prop. Leading 1-based row-number column,\n rendered before any selection column.\n- **New** `pageSize` prop on `<SvGrid>`. Initial page size for the\n built-in pager.\n- **Improved** column virtualizer now detects per-item size changes,\n not just total size. Fixes a regression where resizing a column under\n `fitColumns` left other columns stale.\n- **Improved** the wrapper-managed filter pipeline no longer applies\n filters to the pre-paginated view - the filter UI always sees the\n full dataset, not just the visible page. (Removed\n `paginatedRowModel` from the engine pipeline; pagination is applied\n AFTER filters by the wrapper.)\n\n### @svgrid/enterprise\n\n- **New** package. The paid feature pack adds:\n - **Export** to xlsx / pdf / csv / tsv / html, with theme-matched\n styles, multi-sheet workbooks, header + footer with logo, embedded\n cell images. See [Export](./help/export.md).\n - **Print** with repeat-on-page headers, optional cover page,\n page-size + orientation.\n - **Import** from xlsx / csv / tsv / json with column mapping +\n per-row validation. See [Import](./help/import.md).\n - **AI assistant**: provider-agnostic helpers for natural-language\n filter, smart fill, summarize, and classify. Bring your own model\n adapter. See [AI](./help/ai.md).\n - **Pivot tables** via `createPivotModel(data, config)` and\n `pro.pivot.build(config)`. Row + column axes, 8 built-in\n aggregators or custom, grand-total row + column, subtotals,\n custom axis sort. See [Pivot tables](./help/pivot.md).\n- **New** `installEnterprise(api)` augments a `SvGridApi` with `exportData`,\n `print`, `importData`, `ai.*`, and `pivot.*` methods.\n- **Soft-gated licensing**: features run unlicensed but the grid shows\n a small watermark + a one-time console nudge. `setLicenseKey('SVENTERPRISE-…')`\n at app startup clears both.\n\n### Examples gallery\n\n- **New demos** (56-59): theme-matched export, branded export with\n header + footer + logo, export with cell images, multi-sheet export.\n- **New demos** (52-pivot-table, 51-ai-assistant, 53-excel-import):\n reference implementations for each Enterprise feature.\n- **New demos** for industry verticals: stock market (live ticks),\n HR team, finances ledger, industrial IoT, localization, CSP\n compliant, accessibility, cascade editing, server-side rendering,\n industrial dashboard, healthcare EMR, logistics, compliance queue,\n field service, gantt, scheduler, CRM, admin dashboard, seller panel.\n- **New** \"Source\" button on every demo opens the raw `.svelte` file\n in a modal with a Copy-to-clipboard button. Wired via\n `import.meta.glob('../demos/*.svelte', { query: '?raw' })` so it\n picks up new demos automatically.\n\n### Docs\n\n- **New** [Enterprise feature pack](./enterprise/README.md) landing page.\n- **New** [API reference](./reference/index.md) with hand-curated\n pages for `<SvGrid>`, `SvGridApi`, `ColumnDef`, features, and the\n full Enterprise surface.\n- **New** [Why headless?](./why-headless.md) explains the layered\n architecture and when to drop down to the headless core.\n- **New** [Tailwind integration](./help/tailwind.md) walks the\n `--sg-*` token surface, dark-mode wiring via `data-theme`, and the\n override hooks for the stable `.sv-grid-*` class names.\n- **Split** [Getting started](./getting-started.md) into six short\n pages (install / first-grid / data-and-columns / features /\n theme-and-density / going-to-production); the old single-page\n version is preserved at\n [getting-started-full.md](./getting-started-full.md).\n- **Cleaned** every code sample against the live library surface.\n Removed phantom APIs: `state={...}`, `rowModels={...}`,\n `initialState={...}`, `manualFiltering`, `manualSorting`,\n `manualPagination`, `onColumnFiltersChange`, the wrapper `getRowId`\n prop, per-column `enableSorting` / `enableColumnFilter` /\n `enableGrouping` flags, the never-shipped\n `@svgrid/grid/themes/default.css` import.\n- **Em-dashes globally swept to hyphens** (`-` → `-`) across all\n source-controlled text. Codified as a rule for new content.\n\n## How we version\n\n| Tier | What it means |\n| ----------------- | -------------------------------------------------------------------------------------------------------------- |\n| **Major** (1.x) | Breaking change to a Stable API. Includes prop / method renames, type-narrowing, default-behaviour changes. |\n| **Minor** | New Stable API. New Experimental API. New demo. Doc rewrites that touch the public-facing claims. |\n| **Patch** | Bug fixes. Type-only fixes. Internal refactors with no public surface change. Doc typo fixes. |\n\nThe [API stability page](./help/api-stability.md) annotates each export\nwith its current tier (`Stable`, `Experimental`, `Internal`). As of 1.0,\n`Stable` APIs follow semver; `Experimental` APIs may still change in a\nminor release.\n"
4607
+ "markdown": "# Changelog\n\nThe user-facing log of what shipped. Format follows\n[Keep a Changelog 1.1.0](https://keepachangelog.com/en/1.1.0/);\nversioning follows [Semantic Versioning](https://semver.org/).\n\nEvery release has up to six sections:\n- **Added** - new features\n- **Changed** - existing functionality, behaviour changes\n- **Deprecated** - soon-to-be-removed features\n- **Removed** - features removed in this release\n- **Fixed** - bug fixes\n- **Security** - vulnerability patches (with CVE ids when applicable)\n\nPre-release entries live under `## [Unreleased]`. They graduate to a\ndated heading on publish via `pnpm changeset version` (see\n[.changeset/README.md](../.changeset/README.md)).\n\nFor machine-readable releases, fetch\n[`/changelog.json`](/changelog.json) - same content, parseable shape.\n\n## [Unreleased]\n\n### @svgrid/grid\n\n#### Added\n\n- **`SvRichText` for pages and articles.** Four tools to list in `tools`: `image` (the file\n goes to `onImageUpload`, which resolves to the URL to insert; without it, a URL is asked\n for), `table`, `hr` and `source` (the HTML in a text area). `name` posts the value with a\n native form as a hidden input, and `sanitize` replaces the sanitizer for content the app\n trusts. The default toolbar is unchanged. `sanitizeHtml` now keeps tables (with `colspan`,\n `rowspan` and `scope`) and `<hr>`.\n\n- **`trustedHtml(html)`.** `sanitizeHtml`, ready for an HTML sink on a page that requires Trusted\n Types: there it is a `TrustedHTML` of the `svgrid` policy, elsewhere the same string.\n\n- **`toggleOnSecondClick` on a checkbox column.** The first click selects the cell, a click on the selected cell toggles it, and a finger toggles on the tap. Off by default.\n\n- **`parseClipboardText`, and more for the fill and paste hooks.** `parseClipboardText(text)` splits a\n paste into rows of cells, for a consumer whose copy quotes a cell. `processCellForFill` also gets\n `sources` and `index`, and `processCellFromClipboard` gets `cell`, the place in the pasted block.\n Left out, a grid fills and pastes as it did.\n\n- **`keepFocus`.** When the focused control of a region is removed or disabled, focus\n moves to the control now in its place, not to the body. `createFocusTrap` and\n `menuFocus` use it, so dialogs, drawers and menus have it. `use:keepFocus` gives it\n to any other region.\n- **`describedBy` on `SvModal`.** The id of what a dialog says, read with its title.\n\n- **`holdRowsOnWrite`.** A paste, a clear or a fill changes rows where they stand, as a\n typed edit does, for an app that orders the view itself when the person moves on.\n- **`onPasteOverflow`.** The lines of a paste past the last row, split into cells, inside\n the paste's undo step. Without it they are dropped, as before.\n- **`gridLabel`.** The accessible name of the grid, on its table element. SvGrid Base\n passes the table's name.\n- **Very long grids scroll true at both ends.** Past the height a browser paints\n cleanly the grid now scales its scroll, with a pixel staying a pixel near each end,\n so the rows past about 560,000 of a million draw as they should and Ctrl+End\n reaches the last row.\n- **`onColumnVisibilityChange` and `onColumnPinningChange`.** Hiding,\n showing, pinning or unpinning a column, from the column menu, the api\n setters or `api.setState`, now calls back with the hidden ids or the\n `{ left, right }` pinning, the same shapes `api.getState()` reports. Both\n are deduped on the value and do not fire on mount. An app saving the\n layout no longer has to poll `getState()` to notice either. The\n `<sv-grid>` element fires `columnvisibilitychange` and\n `columnpinningchange`.\n\n- **`SvGridProps` is exported.** The full prop type, for a wrapper that\n forwards a typed options bag to `<SvGrid>`. Without it a wrapper had to\n hand-write a `declare module \"@svgrid/grid\"` stub, which shadows the\n package's own types.\n\n- **`chartSpecToTable`.** The rows and columns a spec amounts to, in the\n shape `SvGrid` takes: one row per category with a column per series (a\n grouped axis adds a Group column, a histogram From and To, candles their\n four prices, a box plot its five numbers), per point for a scatter, per\n link for a sankey or chord, per leaf for a tree map or sunburst with a\n column per level, per day for a calendar, one row for a gauge. Number\n columns carry the chart's format (a right-axis series its axis's) and a\n date axis is a date column. The gallery demos' Chart | Grid switch is this\n behind a sortable grid.\n\n- **The Chart panel charts the pivot in pivot mode.** With `pivot` and\n `charting` on one grid, the pivot bar carries the Chart toggle and the\n panel draws the pivot on screen: row groups as categories (nested rows a\n grouped axis), column groups as series, the measures' format on the axis,\n totals left out. The data pickers step aside with a note and the Type\n select keeps the shapes a pivot can take; Stacked, 100%, Horizontal, the\n Format tab, the builder, saved charts, export and Describe still apply. A\n click on a bar filters the innermost row dimension when the grid has a\n column for it, which re-runs the pivot. It used to chart the source rows by\n the source columns whatever was on screen. The mapping is\n `pivotResultToChartSpec` (with `pivotChartType`, `pivotFilterColumn`)\n in the lazy chart engine; enterprise's `pivotToChartSpec` is the same\n function. `bucketsToChartSpec` (server aggregates as a spec) moved there\n from the controller, out of the base bundle.\n\n- **Scatter points select.** A click or Enter on a scatter or bubble point\n selects it the way a bar selects: the point's `label` (or its series) is\n the category, its y the value, and `selectable`, `bind:selected`,\n `onSelectionChange` and `onSelect` all apply. The dots had no click\n handler, so on a scatter chart those props did nothing.\n- **`ChartPointRef.index`.** A selected bar carries its category index\n beside its label, and two refs with an index match only on the same one.\n A grouped axis repeats its leaf labels (Q1 to Q4 under each year), so a\n click on the second Q2 used to select the first Q2 with it. A ref written\n without an index still matches by label.\n- **Data labels on a 100% chart read the share.** Under `stacked100` the\n axis is already a share whatever `valueFormat` says; the labels on the\n bars now say \"18%\" as well, rather than \"$18\", or \"1800%\" when the format\n was percent. A `dataLabels.formatter` still receives the raw value and the\n tooltip keeps it.\n- **A series on the right axis reads in that axis's format.** The tooltip,\n the screen-reader table, the live region and the data labels format a\n series with `axis: 'right'` through `y2Axis.format` or `formatter`, so a\n margin of 0.29 beside a revenue in dollars reads \"29%\" and not \"$0.29\".\n Without a format of its own the right axis is plain numbers, as its ticks\n are.\n- **The breadcrumb survives `toolbar={false}`.** A drilled sunburst, tree\n map or pie with the toolbar off still shows its breadcrumb in a strip of\n its own; it used to have no way back up.\n- **Chart axis model.** `xAxis`, `yAxis` and `y2Axis` on a spec take\n `min`, `max`, `nice`, `tickCount`, `tickInterval`, `format` or a\n `formatter`, `title`, `gridLines`, `labelRotation`, `reversed` and, for\n the x axis, `type: 'number'` for a numeric axis that spreads categories by\n value. `referenceBands` shade a range on either axis, `referenceLines` can\n be vertical, `title` / `subtitle` / `caption` frame the chart. Series\n gained `marker` shapes and per-point `markers` / `colors`, `strokeWidth`,\n `dash`, `opacity`, `gradient`, `step` lines, `connectNulls` and a\n `nullAs` policy; `dataLabels` takes a placement, a formatter and overlap\n hiding; the component takes a `tooltip` snippet, `tooltipFormat`,\n `tooltipMode: 'single'`, a `legend` side and a `legendItem` snippet, and\n `autosize`. `rowsToChartSpec` reduces with `min`, `max`, `median`,\n `first`, `last`, `countDistinct` and `pN` and buckets a date dimension by\n `day` / `week` / `month` / `quarter` / `year`.\n- **Decimation for long series.** `decimate: 'auto'` (the default) thins a\n line or area with more points than pixels through LTTB, which keeps the\n shape; `{ method: 'minmax' }` keeps every spike. Zoom is applied first, so\n zooming into 300 of 100,000 points shows all 300. `lttb`, `minMaxIndices`,\n `decimateSpec` and `pickCategories` are exported.\n- **Thirteen more chart types**: `histogram` (with `binValues` and\n `rowsToHistogramSpec`), `range-bar`, `range-area`, `lollipop`,\n `dumbbell`, `pareto` (`paretoSpec`), `stream` (`stackOffset: 'wiggle'` or\n `'silhouette'`), `sunburst`, `radial-bar`, `radial-column`,\n `nightingale`, `chord` and `bullet`; plus hollow and Heikin-Ashi candles\n (`candleStyle`) and pyramid / cone funnels (`funnelShape`). All of them are\n in the grid panel's picker, gated on what the current columns can feed.\n- **Zoom, pan and presets.** `zoomable` takes an object: `wheel` (`true` or\n `'modifier'` for Ctrl + wheel), `pinch`, `pan` (`'shift'`, `'mode'` or\n both) and `axis` (`'x'`, `'y'`, `'xy'`). The window is a bindable\n `zoom` prop of category indices with `zoomTo` / `resetZoom` on the\n component and an `onZoom` callback; `rangePresets` puts 1W / 1M / 3M / 6M /\n YTD / 1Y / All in the toolbar on a time axis.\n- **Synchronized charts.** Charts with the same `syncGroup` share their\n crosshair and zoom window, matched by label and then by date. The grid\n panel's `syncTabs` keeps one window across its chart tabs.\n- **Context menu, animation, drilldown, selection, keyboard.** `contextMenu`\n (built-in export / zoom / series items, your own `MenuItem`s, or a function\n of the clicked point; Shift + F10 opens it), `animate` (a data-update tween\n and `'fade'` / `'grow'` / `'wipe'` enter effects, off under reduced\n motion), `drillable` + `drillPath` for tree maps and sunbursts with a\n breadcrumb, `selectable` + `selected` + `onSelectionChange` for point\n selection with dimming, Up / Down stepping through series with a polite live\n region (`announce`).\n- **Financial toolkit.** `bollingerBands`, `rsi`, `macd`, `vwap`, `atr`,\n `stochastic`, `wma` and `obv`, NaN-padded like the moving averages;\n `'wma:N'`, `'bb:N:K'` (a shaded band) and `'vwap'` overlays;\n `indicatorPane` and `SvChartPanes`, which stacks indicator panes under a\n price chart with one x axis, one gutter and one crosshair; `rowsToOhlcSpec`\n and `resampleOhlc` (daily bars to weeks / months); `lastPriceLine`;\n annotation `shape` (`'flag'`, `'pin'`, `'square'`) with a tooltip\n `text`; `volumes` on a series.\n- **Drawing tools.** `drawable` puts trend line, horizontal ray, Fibonacci\n retracement, rectangle, arrow and note tools in the toolbar; the drawings\n are data points (`spec.drawings`, `onDrawingsChange`) so they survive a\n zoom, a resize or new data. A selected drawing has draggable handles and\n Delete removes it.\n- **PDF and print** with no PDF library: `chartToPdfBlob`, `downloadChartPdf`\n (a one-page A4 or Letter PDF with the chart as a JPEG and the titles as\n text) and `printChart`, in the toolbar, the context menu and the panel's\n export menu.\n- **Chart builder in the grid panel.** The Build button opens a modal with a\n type gallery (a live thumbnail per type), the data pickers and a Format tab\n (titles, legend and label placement, axis title / min / max / format / grid\n lines, per-series colour / type / axis / marker / width). The format is\n `ChartFormatState` on the chart tab, applied by `applyChartFormat` to every\n derived spec, saved with the view and reachable through\n `configureChart({ format })`. A link toggle unlinks a chart from the grid\n (`configureChart({ frozen: true })`) so it keeps its data while you filter.\n `charting.onChartCreated` / `onChartChanged` report the panel's lifecycle;\n `charting` also takes `zoom`, `rangePresets`, `syncGroup`, `syncTabs`,\n `contextMenu` and `animate`.\n- **Candlesticks from the grid panel.** Pick Candlestick or OHLC bars and the\n panel finds the Date, Open, High, Low, Close and Volume columns by name;\n Indicators chips add panes and overlays; `configureChart({ ohlc,\n indicators })` and the AI chart plan reach the same thing.\n- **A scatter series' `overlay` is a regression of y on x.** Until now the\n scatter layout ignored `overlay` (the fits ran on the index, which a\n scatter has no use for). The regressions (`linearFit`, `polynomialFit`,\n `exponentialFit`, `logarithmicFit`, `powerFit`) take an optional `xs`\n and return `predict(x)`; `regressionFit(values, overlay, xs)` names one by\n its overlay string; a scatter series with `overlay` gets the curve drawn\n across the plot (sampled evenly in log space on a log axis) and its\n equation and R-squared in the point tooltip.\n- **Chart gallery: fourteen full-size demos, one per family, and a docs page\n for them.** Bars, lines, areas, pie / donut / sunburst, scatter and bubble,\n combination, histogram and box plot, heat map and calendar, radar and\n radial, tree map and sunburst, sankey and chord, waterfall / funnel /\n pareto, gauges and bullets, and an annotated chart (demos 438 to 451),\n each with its own data and the switches that family needs, embedded on\n `help/charts/gallery` and the type pages.\n- **Grid chart panel: a deeper Format tab, one set of pickers, a\n localized panel and saved charts.** The builder's Format tab adds series\n labels, the crosshair's axis pills, a compact rule below a width\n (`compactBelow`, through `CHART_RESPONSIVE_PRESETS.compact`), a stack\n group per bar or area series and the chart's font, size, background and\n text colour (`ChartFormatState.seriesLabels` / `crosshairLabels` /\n `compactBelow` / `series[].stack` / `style`). The panel's pickers and the\n builder's Data tab are one component (`SvGridChartPickers`), so the two\n cannot disagree about what a type needs. Every string the panel and the\n builder show is a `chart*` key on `localization.text`\n (`GridChartPanelMessages`, loaded with the panel; `GridMessages` is now\n the chrome map and the panel map together). `api.saveChart(name)`,\n `applySavedChart`, `removeSavedChart`, `getSavedCharts` and\n `configureChart({ saved })` keep a tab's whole configuration under a name;\n the panel has a popover for them and `getState()` carries `savedCharts`\n when there are any.\n- **Chart diagnostics, a plain-language description, live feeds and a JSON\n schema.** `validateChartSpec` returns what is off about a spec (unknown\n keys with the nearest name, an unknown type, a series short of the\n categories, a log axis at zero, an overlay the engine does not read, an\n anchor naming no category, and more); `SvChart` logs the findings once per\n spec in development, `aiChart` attaches the errors to its plan, and the\n MCP server's `svgrid_check_code` applies the rules to a static spec\n literal with a rename for a misspelt key. `chartSummary` reads a chart in\n a sentence (trend, extremes, shares, correlation, last close) and\n `SvChart` hands it to assistive technology as `aria-description` and the\n table caption (`describe`), the context menu and the panel copy it, and\n `aiExplainChart` grounds an Explain button on it. `appendPoints` builds\n the next spec of a feed with every per-category array kept in step and a\n rolling window, and `live` on the chart skips the tween and enter effect.\n `docs/schemas/chart-spec.json` is generated from the types, served at\n svgrid.com/schemas and by `svgrid_get`, and a test keeps it equal to the\n validator's lists.\n- **Series and interaction depth.** Areas join `stack` groups like bars\n (`stacked100` and `stackOffset` run per pile); a numeric x axis takes\n `scale: 'log'` (even decades, 2 / 5 minors, non-positive categories\n dropped, decimation in log positions); regression overlays `poly:N`,\n `exp`, `log` and `power` next to `linear`, each carrying `r2` and\n `equation` on the overlay line and read in the tooltip, with\n `linearFit` / `polynomialFit` / `exponentialFit` / `logarithmicFit` /\n `powerFit` / `rSquared` / `computeOverlayFit` / `overlayName` exported\n and `charting.trend` accepting the new names; hovering a mark dims the\n other series (`hoverHighlight`, on by default, snapping within 18px) and\n `onHover` reports the point once per change; `tooltipPosition` parks the\n tooltip in a plot corner and `tooltipSticky` pins it on a click; data\n labels take `rotation` and `connector` (pushed clear of a collision with\n a leader line); `spec.style` sets a chart's font, size, background and\n text / grid colours over the tokens (`chartStyleVars`), honoured by the\n exports.\n- **Accessibility pass on the chart.** Every non-cartesian family (pie,\n sunburst and the radial charts, heat map, calendar, tree map, funnel,\n sankey nodes and links, radar, scatter) is one Tab stop with a roving\n focus: arrows walk the marks (a heat map by row and column, a calendar by\n day and week), Home / End / PageUp / PageDown jump, the focused mark\n raises its tooltip, Enter / Space select or drill, and every mark is named\n and has a focus ring. Zoom from the keyboard: `+` / `-` / `0` on a\n focused category and Shift + arrows to pan, and the brush is a `slider`\n (arrows pan and resize, Home / End, `0`). The toolbar is a `toolbar`, the\n tooltip a `tooltip`, decorative icons are hidden from the tree. Forced\n colors (Windows High Contrast) keep the data colours and repaint the\n chrome with the system palette. `localeText` on `SvChart` (and\n `charting.localeText` for the grid panel) translates every string the\n chart renders itself, with `ChartMessages`, `defaultChartMessages`,\n `resolveChartMessages` and `chartMessage` exported. A new\n `tests/e2e/chart-a11y.spec.ts` runs axe-core against the rendered chart\n on three demos and drives the keyboard model in a real browser.\n- **Stack groups, series labels, responsive rules, pie callouts.** `stack`\n on a bar series names the pile it joins, so plan and actual stack side by\n side in each category and the axis fits the tallest stack; `seriesLabels`\n names each line at its last point in a reserved gutter, pushed apart where\n lines end together; `responsive` patches the spec by the rendered size\n (axes merge one level deep, a rule's `legend` moves or hides the\n component's legend; `resolveResponsive` and `matchResponsiveRules` are\n exported); `dataLabels: { placement: 'outside' }` on a pie draws callout\n labels with leaders instead of percentages, and a spec can carry its own\n `dataLabels`; `visible: false` starts a series hidden with its legend chip\n dimmed; the crosshair reads off the axes with a category pill and a value\n pill (`crosshairLabels={false}` turns them off).\n- **`moveCells` - drag a selected range to move or copy it.** Grab a selected\n range by its border and drop it somewhere else. Move by default, copy with\n Ctrl / Cmd, and the modifier is read at DROP time so it can be pressed or\n released mid-drag. On whenever cell selection is on; `moveCells={false}`\n opts out. A drop is refused outright - nothing changes - when it would land\n off-grid or when any source or destination cell is read-only, rather than\n moving the part that fits and silently dropping the rest.\n- **`selectionBar` - a floating bar over the grid while rows are selected**,\n showing the count and the actions that apply to the whole selection. `true`\n gives the count and a Clear button; an array is shorthand for `{ actions }`;\n the object form also sets `position` (`'bottom'` default, or `'top'`),\n `maxVisible` and `hideClear`. The prop and its types are free; the bar\n itself is an Enterprise renderer, and without it the grid shows a short upsell\n in its place, the same way the scheduler and board views do.\n- **Edge auto-scroll for every drag that extends a rectangle** - the fill\n handle, the range move, and plain drag-select. All three were previously\n capped at whatever was already on screen when the drag began, which on a grid\n built for 100k rows meant the gesture mostly did not work. One shared rAF loop\n scrolls the container when the pointer is inside a 40px edge band, on a linear\n ramp to 20px/frame, in both axes. It keeps scrolling with the pointer parked\n still, which is the whole point of holding at an edge.\n- **Row and column resizing, both opt-in** (`rowResize`, `columnResize`).\n `columnResize` gives every header a drag handle: drag the edge, double-click\n it to size the column to its content, or focus it and use Left/Right (Shift\n for a 1px step). `rowResize` does the same vertically and brings the\n row-header column with it, since that is where the grip belongs; the grid\n stores the heights itself, so neither needs a function-valued `rowHeight`.\n Both load their code on demand, so a grid that leaves them off downloads\n neither.\n- **`ColumnDef.resizable`** - per-column opt-out, consulted only when the grid\n has `columnResize` on. `resizable: false` removes that column's handle\n entirely, so the drag, the arrow keys and double-click-to-autosize all go with\n it, and the column menu drops its Autosize item. For the columns whose width\n is part of the layout: a row-number gutter, a checkbox column, a fixed icon\n column. Programmatic sizing (`api.setColumnWidth`, `api.autosizeColumn`,\n `fitColumns`) still applies.\n- **`SvGridViewState` is exported.** It is the return type of the public\n `api.getState()`, so writing a typed \"named views\" feature previously meant\n either re-declaring the shape or reaching into an internal path.\n- **Form layer depth in `SvForm` / `createForm`.** Schema-driven forms gained:\n conditional / dependent fields (`visible` and `disabled` accept a\n `(values) => boolean` predicate; hidden fields drop out of validation and the\n submit payload), grouped **sections** and a multi-step **wizard** (`stepper`,\n with per-step validation gating), repeatable **field arrays**\n (`type: 'array'` with `itemFields`, plus `addItem` / `removeItem` / `moveItem`),\n **async validation** (debounced, stale-response-guarded, per-field\n `validating` state), **dirty tracking + `reset()`**, and **submit states**\n (`isSubmitting`, an async `onSubmit`, and `setErrors()` for server-side field\n errors). New demos: dynamic form, wizard, and field array.\n- **Public headless cores for the menu / select family** (`createMenu`,\n `createPopoverSelect`, shared `list-nav`), so the popup pickers can be driven\n headlessly and composed into custom UI.\n- **`color` on `ListOption`** - render a small color swatch before the label in\n `SvListBox` and `SvDropDownList`.\n- **`block` on `SvDateTimePicker`** - stretch the field to fill its container\n (100% width), so it fills a grid cell when used as an editor.\n- **`onClosed` on `SvDrawer`** - fires after the exit animation ends and the\n drawer leaves the DOM.\n- **`loading` on `SvField`**, and **`locked`** on `SvDockLayout` /\n `SvDockManager`.\n- **`summary` on `<SvGrid>`** - shortcut alias for `enableRowSummaries`,\n which turns on the footer row that aggregates every filtered row. It wins\n when both are set, the same precedence `selectable` has over\n `enableCellSelection`.\n- **`summary` on a column** - choose that column's footer aggregate\n (`'sum'`, `'avg'`, `'min'`, `'max'`, `'count'`, `'countDistinct'`,\n `'extent'`, `'first'`, or a custom function), or `false` to leave the cell\n blank. Without it a column keeps the old default: sum a numeric column,\n `Count: N` otherwise.\n- **Development-time configuration checks.** The grid used to fail silently on\n the most common mistakes - a misspelled `field` rendered a column of blank\n cells and printed nothing. It now warns once per problem, in dev builds only:\n\n - a column `field`, `groupBy` entry, `treeData.parentField` / `idField` or\n `treeData.column` that does not exist (with a \"did you mean\" for near misses),\n - two columns resolving to the same id,\n - `pageSize` set when pagination was never turned on,\n - a column marked `sortable` while sorting is not enabled,\n - `initialColumnPinning` while column virtualization is on, which silently\n prevents the columns from sticking,\n - `externalPagination` without `rowCount`, `externalSort` without\n `onSortingChange`, and `externalFilter` without `onFiltersChange`.\n\n The checks live in a lazy chunk behind a dev-only branch, so a production\n build never loads them and the base bundle is unchanged.\n\n#### Changed\n\n- **`showColumnFilters` no longer puts an input in the header.** Set to\n `true`, it also drew a plain \"Filter\" text box in every header cell,\n which with `showFilterRow` made two filter inputs per column. It now\n only turns on the column menu's filter section, as documented. For\n always-visible inputs use the filter row: `showFilterRow` or\n `filterMode=\"row\"`.\n- **The find bar loads on the first Ctrl+F.** It is its own chunk now, so a\n grid nobody searches does not carry it; the first open shows the bar once\n the chunk has loaded.\n- **`alignedGridGroup` loads its module on demand.** The shared group that\n mirrors horizontal scroll and column widths between grids is now a lazy\n chunk, fetched by the first grid that sets the prop, the way managed row\n drag is. A grid outside a group no longer carries it, which kept the base\n bundle under its 100 KB budget after this release's scroll and mount\n changes. Mirroring starts once the chunk has loaded, a frame on a warm\n cache.\n- **BREAKING: column resizing is now off by default.** Every column used to have\n a drag handle unconditionally - there was no prop, no per-column option and no\n way to turn it off, so a `width` you set was only ever a starting value a user\n could drag away. Set `columnResize` to get the handles back; the 241 demos in\n this repo that want them now say so. Sorting and filtering have always been\n opt-in behind `sortable` / `filterable`, and resizing was the odd one out.\n- **BREAKING: the footer summary row is now off by default.**\n `enableRowSummaries` used to default to `true`, so a plain\n `<SvGrid {data} {columns} />` grew an aggregate footer nobody asked for and\n every caller that did not want one had to opt out - 375 call sites in this\n repo passed `enableRowSummaries={false}` against 6 that opted in, and the\n API reference already documented the default as `false`. If you were\n relying on the old behaviour, add `summary` (or `enableRowSummaries`).\n- **Ember is now the default theme preset** (`defaultThemePreset` in\n `@svgrid/grid/themes`). It is what the demo gallery and svgrid.com open on, so\n a scaffolded or Studio-generated app now looks like the demos unless you pick\n a preset. Previously the default was Tailwind.\n- **Popup editors keep themselves on screen.** `SvAutoComplete`, `SvComboBox`,\n `SvDropDownList` and the date pickers now measure the space to the viewport\n edge: the panel flips up when there is not enough room below and clamps its\n `max-height` to the available height, so a picker opened near the bottom of the\n window no longer spills off screen.\n- **Dock manager pop-out to a new window is now off by default**\n (`allowPopout={false}`); opt in explicitly. Added a dedicated pop-out demo.\n\n#### Fixed\n\n- **`fitColumns` with `responsive` fills the row when a `hideBelow` column is hidden.** The fit still\n counted the hidden column's width, so the shown columns stopped short and left a blank strip that\n wide at the right edge.\n- **Cut clears only what reached the clipboard.** Ctrl+X, the Cut menu item and the cut command\n cleared the cells at once, before the clipboard write had finished, so a write the browser refused\n (a page without focus, a denied permission) lost the values. The clear now waits for the write, and\n runs only if the text landed and the selection has not moved meanwhile.\n `copySelectionToClipboard()` returns a promise of whether the text landed.\n- **The pinned-column divider in a contrast theme.** Forced colors keep a fill's alpha, so the\n `pinSplitter` divider stayed a faint line and vanished on `Canvas`. It is drawn in `CanvasText`,\n and in `Highlight` under the pointer, with focus and while dragged.\n- **Chart panel: an Export item that is off looks off.** The PDF item that cannot draw the chart's\n text said why but looked like the items that work; it is dimmed as an off CSV item is.\n- **Scheduler model: a day the clocks change is drawn by the clock.** `layoutDayEvents` placed an\n event by the real minutes since midnight, so on a 25-hour day every event sat an hour below the\n row its time is labelled with, and on a 23-hour day an hour above; an event in the last hour of a\n band fell out of it. It now counts wall-clock minutes (`wallMinutes`, `atWallMinute`, both new\n exports). On a week that starts on a 23-hour Sunday, `monthWeekSegments` put Monday's bars under\n Sunday. `timelineAxis` for `timelineDay` ends the band at its hour by the clock and names each\n real hour by its clock (a 25-hour day shows 1 AM twice), where it left a gap or an overlap.\n- **A menu item that is off keeps the arrow keys.** `menuFocus`, `createMenu` (`SvMenu`,\n `SvContextMenu`, `SvMenuList`) and the column and context menus mark it `aria-disabled` and keep\n it in the arrow-key and first-letter order, as the ARIA menu pattern does, so a screen reader\n hears it and why it is off. Its click does nothing. Before, the arrows skipped it.\n- **Forced colors: the header checkbox of a part-selected grid keeps its dash.** The dash was a\n fill, which a contrast theme removes, so the box read as empty.\n- **A chart's PDF refusal is seen at the toolbar and in the grid's chart panel.** For a title the\n PDF's font cannot draw, the toolbar's PDF button did nothing on screen: the reason went to the\n screen reader only. It is now marked off (`aria-disabled`, the reason as its tooltip) once it is\n pointed at or focused, and a press shows the reason in the button. The chart panel's Export menu\n wrote no file and said nothing; its PDF item now reads the reason and is off.\n- **The chart panel's Export menu takes the keyboard.** Arrow keys, Home, End and first letters move\n in it, focus goes to its first item when it opens and back to Export when it closes (it fell to the\n page). Its CSV item, when the chart has no table of values, stays focusable and says why.\n- **Rich text renders under Trusted Types.** `SvRichCell`, `SvRichText` and the sanitizer passed plain\n strings to `DOMParser`, `{@html}` and `innerHTML`, so a page with `require-trusted-types-for 'script'`\n threw and drew nothing. They now go through one policy, `svgrid`, made only where the browser has\n `trustedTypes`. A CSP with a policy list needs `trusted-types svgrid svelte-trusted-html`\n ([SECURITY.md](https://github.com/sv-grid/sv-grid/blob/main/SECURITY.md#trusted-types)).\n- **Chart Print runs under Trusted Types.** `printChart` built its window with `document.write`; it\n now builds it node by node and draws the chart as a `data:` image.\n- **A chart PDF never writes `?` for a letter.** Its text is in the PDF's built-in Helvetica, which\n has Western European letters only. The euro sign, curly quotes and dashes now come out right. A\n title, subtitle or caption with other letters (Cyrillic, Greek, CJK, Hebrew, emoji) is refused: the\n menu's PDF item says which text and points to PNG or SVG, and `chartToPdfBlob` and `buildChartPdf`\n throw a `ChartPdfTextError` naming it. `pdfUndrawable()` tells beforehand.\n\n- **A block cache that refreshes a block still being read frees that read's slot.** `refresh()`,\n `applyRows` and an insert or remove gave up a read in flight without counting it as done, so\n `loading` stayed true for good, and with `maxConcurrentRequests` reached no block loaded again.\n\n- **The \"Could not load these rows.\" line of a failed run sits on the first row wholly on screen.**\n It sat on the first rendered row, which overscan puts above the screen, so a person scrolled\n into a failed block saw a tinted band and no words.\n\n- **An edit deep in a server row model no longer walks the rows.** `find`, `findIndex`, `some` and `indexOf` on windowed\n data read only what is loaded, a paste, clear or fill over rows held in place copies only those, and `cellEdited`\n hands over `rowId`.\n\n- **A filter's count is said once.** Opening or closing a group under a filter no longer repeats\n it, and a closed group counts the records the filter keeps: it read \"28 of 30\" with 24 shown.\n- **`createFocusTrap` and `menuFocus` never drop focus to the body.** An opener that left the page,\n or turned disabled or hidden, hands focus to what held it. A dialog opened from a menu item goes\n back to what the menu went back to. A tooltip closes with Escape.\n\n- **CSV: a cell that starts with a line feed is guarded.** Tab and carriage return were,\n and a spreadsheet reads `\\n=1+1` as a formula too.\n- **A commit keeps the keyboard in the grid.** A pick in a select or dropdown cell, or\n Escape on a multi select, left focus on the body and arrows dead. Tab in a dropdown\n cell left the grid; it now commits and moves on, as the docs say.\n- **Group rows follow their state.** `aria-expanded`, the chevron and the button's\n label went stale until the row was drawn again.\n- **A grouped grid behaves as a treegrid.** Left and Right close and open a group, Enter\n and Space toggle it, and rows of records carry `aria-level`. Enter on a group's row\n used to move down.\n- **The filter count counts records.** \"28 of 34 rows match\" included group headers.\n- **A menu scrolls to the item the keyboard moved to.**\n- **A first click on the cell seeded on mount selects it.** It used to open the editor.\n- **Moving a column by hand is an undo step**, as a resize is. The api's setter records none.\n- **\"Add condition\" in the filter menu keeps focus**, and closing that menu goes back\n to the funnel.\n\n- **Keyboard and focus.** The Menu key and Shift+F10 open the context menu of the\n active cell. The column and context menus take focus, walk with the arrows, Home,\n End and first letters, and give focus back. `menuFocus` gives a menu of your own\n the same. The find bar takes focus when opened from a focused control.\n- **Focus stays in view.** Moving the active cell, or tabbing to a header control,\n stops clear of pinned columns on either side and above a sticky footer.\n- **Right to left.** The left and right arrows go where they point, horizontal\n scrolling to the active cell works, and the pin splitter follows the pinned columns.\n- **A grouped grid is a `treegrid`**, and a group row's cell is the active descendant.\n- **Menus, filter panel and find bar translate** through `localization.text`, as do\n the row number, select row, column menu, filter and resize names. `setModalCloseLabel`\n names the x on every `SvModal`.\n- **Ember**'s muted text reaches 4.5:1 on a pinned header. The column menu button is\n 24 px.\n- **IME.** A key of a composition no longer moves the cell or commits the editor.\n- **Smaller base bundle.** What a fill or a range move writes when the drag ends loads\n with the drag (`range-commit.ts`).\n- **`SvRichText` sanitizes its value.** It wrote `value` to `innerHTML`, so a stored string ran\n its handlers. `value` now goes through `sanitizeHtml` on the way in and out, and `sanitizeHtml`\n keeps a text alignment, rebuilt from scratch, so the alignment buttons survive it.\n- **CSV injection through a list.** `escapeFormulas` guarded strings only, so a multi-select value\n `=1+1` written with `rawValues: true` reached the file as a formula. It now guards everything\n but numbers.\n- **Open groups stay open.** In a grid with no `expanded` prop, a change to any other\n prop, such as new `data`, folded every group the person had opened.\n- **A row model that groups on the server is not grouped again in the page.**\n With `rowModel` from `createServerRowModel` and a grouping in the grid's\n state (a view, `groupBy`), the grid bucketed the rows it had loaded under\n its own group rows. A model with `group` accessors now hands over its rows\n as they are; the grouping state stays for whatever shows it.\n\n- **A `rowModel` set after mount still sorts and filters on the server.**\n The grid read `externalSort` and `externalFilter` once, at mount. A model\n handed over later (one loaded on demand) was sorted and filtered again in\n the page, placeholders included: with a million rows behind it, a view\n filter held the page for seconds. The grid now follows the model's\n `setSort` and `setFilter` whenever it arrives.\n\n- **`SvRichText` lists had no bullets or numbers in apps with a CSS reset.**\n Tailwind's preflight sets `list-style: none` on every list, so a bulleted\n list in the editor read as plain lines. The editor now sets `disc` and\n `decimal` itself.\n\n- **A custom column header did not sort when its label was clicked, and did\n not see the sort change.** The header wrapper is itself `role=\"button\"`, so\n the check meant to skip clicks on buttons inside a custom header matched the\n wrapper and skipped every click on the header's own text; only a click on\n empty header background sorted. The `header:` callback also ran once, so\n `ctx.column.getIsSorted()` kept its first value. Both are fixed: a click on\n the label sorts, a button inside the header still gets its own click, and\n the callback runs again when the column's sort changes.\n- **Grouping five million rows ran out of memory.** Every group row carried\n ten closures and a Map of its ancestors' values, and a million rows grouped\n five deep make about a million group rows. Group rows now share their\n methods and keep their state in one object. Grouping 1,000,000 rows that\n way adds about 500 MB instead of 785 MB, and 5,000,000 rows group in 56 s\n on a laptop within Chrome's default heap.\n- **Every selection change serialized the whole selection.** Deciding\n whether to call `onRowSelectionChange` ran `JSON.stringify` over the\n selection state; selecting half of a 5,000,000-row grid built a string of\n tens of megabytes per change. The two states are now compared key by key.\n- **A mount built rows nobody could see, and laid the grid out five\n times.** The first paint now draws the rows on screen and at most two\n overscan rows, and the rest of `overscan` follows a frame after that paint:\n a 520 px grid built 26 rows to show 14. Reading the scroll position of the\n container the grid had just created, and each scrollbar measuring its\n track on connect and on every attribute change, forced four layouts of the\n half-built grid before it painted. The position is now confirmed on the\n next frame and the track is measured by the scrollbar's resize observer;\n the one forced layout left is the container size the windows need.\n- **The scrollbar thumb trailed the rows by a frame.** The custom scrollbar\n read the scroll position on the frame after the scroll flush, so the thumb\n moved one frame late and each scroll step paid a second style pass and\n paint for it. The flush now hands the scrollbar the position it already\n has, and the scrollbar keeps its measured track instead of measuring it\n inside the flush.\n- **A vertical scroll frame repainted every rendered row.** Body rows were\n not positioned, so on any frame that moved or rewrote a row Chrome painted\n every rendered row again cell by cell. Flex-layout body rows are now\n `position: relative` (no z-index, no clip), which lets the browser reuse an\n unchanged row's paint whole. In six interleaved rounds of the 100,000-row\n browser bench, a 60 px scroll step went from 4.05 to 3.68 ms of\n main-thread work (median). Sticky frozen and group rows keep\n `position: sticky`.\n- **Expanding one group re-grouped every row.** A toggle re-ran the row\n pipeline, and the grouping stage bucketed all rows again although only the\n expansion had changed. It now keeps its groups while its input rows,\n grouping and columns are the same. On a 100,000-row grid grouped by five\n columns, one expand went from 521 to 98 ms, and expanding or collapsing\n every group from 473 to 82 ms.\n- **A calendar date with `timeZone: 'UTC'` showed the day before.** A\n bare date string such as `'2026-10-07'` is read as local midnight, so a\n date `format` whose `options` carried `timeZone: 'UTC'` (the usual fix\n from when such strings were read as UTC) showed October 6th anywhere east\n of UTC. A calendar date has no zone: `timeZone` is now ignored for one,\n on screen and in exports.\n- **A cell note's corner marker sat at the grid's corner.** The marker is\n positioned against its cell, but a cell with a note was not positioned, so\n it anchored to the grid. A cell with a note is now a positioning context,\n like the active cell.\n- **A sort or filter kept every column's value on every row.** Reading one\n column built and cached an array of all of a row's values, so sorting a\n 1,000,000-row grid by one column left 1,000,000 of them behind. A plain\n `field` column is now read from the data; a computed (`fieldFn`) column\n still builds the array. The heap a sorted and filtered 1,000,000 x 9 grid\n adds went from 196 MB to 128 MB in the browser harness.\n- **`applyTransaction({ remove })` scanned the data once per row object.**\n Removing 10,000 row objects from 1,000,000 rows took about 1.8 s to find\n them; past eight objects the transaction now builds one identity map.\n- **Group footer and grand-total rows drew checkboxes.** With\n `showRowSelection`, each totals row had a row-selection checkbox, and\n ticking it selected the synthetic row. A checkbox column with no `aggregate`\n showed a checkbox too, holding a leaf's value carried over from the group\n row. Totals rows keep the selection cell for alignment but leave it empty,\n and a boolean column without an aggregate renders nothing there.\n- **Toggling a checkbox cell did not fire `onCellValueChange`.** A click or\n Space on a checkbox cell wrote the row and stopped, so an app that saves\n edits from `onCellValueChange` lost every checkbox change, and Ctrl+Z could\n not take one back. The toggle now takes the inline editor's commit path:\n one undo step, then `onCellValueChange` with the old and new value and the\n row's position in `data`.\n\n- **Arrow keys lagged whenever the active cell had to scroll.** On the\n 10,000 x 53 large-dataset demo one ArrowDown at the bottom of the viewport\n cost about 37 ms of main-thread work and painted over two frames; it is now\n about 8 ms and paints once. Five things added up. Every rendered row was\n measured with `getBoundingClientRect` on each scroll tick, on a grid that\n had no `autoRowHeight` to use the number for, which forced a table layout\n per row (about 30 per key press). The scroll geometry the custom scrollbar\n reads was pulled inside a derived at the start of the flush, forcing a\n second full table layout that the row writes then invalidated. The\n virtualizer keyed rows by their position in the window, so a one-row\n scroll rewrote all ~30 rows x every rendered column; rows are now keyed\n `index mod windowLength`, so the rows that stay keep their `<tr>` untouched\n and the one that entered is moved with a single `insertBefore`. The row\n window only updated after the browser's `scroll` event and a\n `requestAnimationFrame`, a frame after the key press; keyboard navigation\n now syncs the virtualizer inline, since it knows where it scrolled. And\n each cell carried three selection-dependent deriveds plus the fill handle's\n `{#if}`, re-marked on every move; they are one `{@const}` now, `null` for\n a plain cell. Also closed on the way: `key in editedCellValues` on a deep\n `$state` proxy created a signal per rendered cell that never went away, and\n the header checkbox's tri-state walked every row on each arrow key.\n\n- **A heat map drew a series legend in colours its cells never use.** The\n ramp on the right is its key; the row of palette chips under it (and a\n calendar's) is gone.\n\n- **A time axis put its ticks on nothing in particular.** `xType: 'time'`\n stepped a fixed 30 or 365 days from the Unix epoch, so a tick labelled\n \"2025\" stood weeks from New Year, a \"Jun 25\" tick sat mid-June, and a\n thirty-month axis carried two labels on a plot with room for eight. Ticks\n now land on calendar boundaries in UTC (days, Mondays, the first of a\n month, quarter or year, or every few years) at about one label per 90px,\n and each label is sized to the gap between ticks rather than the axis span,\n so a quarterly axis no longer reads \"2024, 2024, 2024, 2025\". The labels\n are read in UTC too: a local-time read of a 'YYYY-MM-DD' category in the\n Americas labelled every tick with the day before.\n- **The tooltip echoed the ISO date.** On a time or ordinal-time axis the\n tooltip title, the crosshair pill and the live region wrote \"2025-06-01\";\n they write \"Jun 2025\" when every category is the first of a month, \"2025\"\n when every one is New Year, and \"Jun 1, 2025\" otherwise, through\n `xAxis.formatter` when one is set. A calendar cell names its day the same\n way. The category in `onHover`, `onSelect` and `selected` is unchanged.\n- **The screen-reader table and the CSV read \"NaN\".** A NaN gap (the way a\n forecast series marks the months before it starts) wrote the word \"NaN\"\n into the hidden data table and the CSV export, and a `null` fell back to\n 0, which told a screen reader the outage month had no sign-ups. Both are\n empty cells now.\n- **\"Double-click again to clear\" left a legend chip isolated and hidden.**\n A double-click arrives as click, click, dblclick: with a chip isolated the\n first click cleared the isolation, the second hid the chip and the dblclick\n isolated it again. A click that clears an isolation is remembered for a\n moment, and the double-click that follows it on the same chip clears; on\n another chip it moves the isolation there.\n- **A radar ignored `yAxis.min` / `max`.** The rim was always the data\n maximum, so scores out of 100 filled the dial to whatever the best score\n was and two radars could not be compared. The rim honours a pinned axis,\n a value past it sits on it, and a pinned minimum is the centre.\n- **A pie sized its gutters by the widest label on either side.** Small\n slices with long names on the left kept the same room on the right for a\n \"Chrome 62%\", and the pie drew at half the radius the pane had. The\n gutters are per side, the pie sits centred in what they leave, the floor\n radius is a quarter of the short side, and each side truncates to its own\n room.\n- **A waterfall tooltip read \"$0\" on a total.** A total bar holds 0 in the\n data and its running sum in the geometry; the tooltip and the\n screen-reader table now read the drawn value. Its x labels also tilt by\n the same fit rule as a bar chart's, with the angle the renderer needs:\n they were flagged as rotated with no angle and drew truncated instead.\n- **Histogram edges read \"1.3k, 1.3k\".** Bin edges are written at the\n precision their width needs (whole numbers with separators for a 44ms\n bin, two decimals for a 0.2 bin) rather than in the compact tick form that\n gave two neighbouring edges one label.\n- **A heat map ignored `dataLabels` in the spec.** The cell label read the\n component prop alone, so `dataLabels: { show: true }` in the spec drew\n nothing. Colour legends on heat maps and calendars also round their steps\n to the precision the range earns (\"20, 94, 167\", not \"20, 93.5, 167\").\n- **The range area read as three lines.** Its band filled at the wash an\n area under a line gets, with both edges at full stroke. The band is the\n mark: it fills at 0.35 and its edges are thin.\n- **Annotate mode pinned a tooltip instead of a note** when `tooltipSticky`\n was on too; the annotate click now owns the click.\n- **Drawings on a time axis warned on every point.** The drawing tools write\n a timestamp for every pointer position, and the spec validator matched\n those against the category list. It skips the check on a time or number\n axis, as it already did for reference lines and annotations.\n- **The panel's export menu and saved-charts dialog ignored Escape.** Both\n close on Escape and hand focus back to the button that opened them.\n- **A gauge wrote \"99.2 %\".** A symbol unit (%, degrees) hugs the number; a\n word unit (\"ms\") keeps its space.\n- **The right axis title was clipped.** Its glyphs point outward under the\n rotation and ran past the edge of the SVG.\n- **Pie callout labels ran off the chart in a narrow pane.** The pie kept a\n fixed 34px margin for its outside labels whatever their length, so a\n \"Samsung Internet 3%\" in a 380px pane was clipped to \"%\" and \"Cl\". The\n radius now makes room for the widest label on each side, keeps at least a\n fifth of the short side, and hands the renderer a per-side character\n budget to truncate to when even that is not enough.\n- **The brush strip inherited the chart's titles, axis labels and marks.**\n A title and subtitle took 45 of the strip's 88 pixels and the x labels\n another 30, leaving the mini-map a five-pixel smear under an empty band.\n The brush spec now mutes titles, axis titles and labels, reference lines\n and bands, annotations, drawings, series labels and data labels.\n- **A category chart tilted its x labels by count, not by fit.** More than\n eight categories, or any label over nine characters, tilted every label\n to -40 degrees even on a thousand-pixel chart with room to spare. `auto`\n now tilts only when the widest label does not fit its slot.\n- **A heat map wrote every column label whatever the cell width**, so\n twenty-four hours at 33px a cell read \"00:0001:0002:00\". The column\n labels thin out the way a bar chart's do.\n- **A calendar labelled the padding days of the next year.** The grid pads\n the range to a full week, and a December ending on a Thursday picked up a\n \"Jan\" tick for the two padding days. Month labels stay inside the range.\n- **A waterfall could not open on a value.** A total bar always read the\n running sum, so \"Revenue 4300\" marked as a total drew as zero and every\n bar after it hung below the axis. A total with a value now sets the sum;\n a total with 0 still reads it.\n- **`rowsToChartSpec` with `topN` duplicated a category called \"Other\".**\n A real \"Other\" in the data came out beside the folded one, two categories\n with one name, which the renderer threw on as a duplicate key. The tail\n now folds into that category.\n- **A pin annotation showed \"O…\" and nothing else.** The head held\n `truncate(label, 2)`, an initial plus an ellipsis, and no label was written\n beside it. The head carries the initial and the label sits beside the pin\n like a dot's does.\n- **A chart's hidden data table took up to 17,000px of scrollable space.**\n The screen-reader table (one row per category, capped at 1,000) carried the\n visually-hidden recipe itself, but on a `<table>` `height: 1px` and\n `width: 1px` are minimums and `overflow: hidden` does not clip, so the\n table was invisible and as tall as its rows, and every scroll container\n around a chart (the demo pages first of all) scrolled on through that much\n empty space. The clip now sits on a wrapping div; the table keeps its id,\n which the svg still points at.\n- **Two ARIA faults the axe audit found.** The chart's SVG was\n `role=\"img\"`, whose children are presentational to assistive technology,\n so the focusable marks inside it (category zones, slices, a drawing) were\n a nested-interactive violation and could be skipped by a screen reader; an\n interactive chart is a labelled `group` now (a thumbnail with\n `interactive={false}` stays an image). Sankey links carried an\n `aria-label` with no role, which ARIA prohibits; they are named images in\n the focus order.\n- **A tree map ignored `spec.tree`.** The field is documented as an alias of\n `treemap` and the sunburst honoured it; the tree map read only `treemap`.\n- **An autosized chart with a side legend grew without end in a plain block.**\n `legend=\"left\"` / `\"right\"` lays the chart out as a three-row grid, and a\n grid puts a gap between every track whether or not the toolbar and brush\n rows have anything in them. The autosize maths counted a gap only for the\n chrome that existed, so in a parent that does not pin the height, the host\n measured 8 or 16px taller than the plot it was told to fit, the plot grew\n into it, and the next measurement grew again - a few hundred pixels a\n second until the page ran out. Hiding the legend at runtime had the\n opposite fault: a dimension binding keeps its last value once its element\n is gone, so the vanished legend went on being subtracted and the plot\n ratcheted down to its 120px floor. Both fixed; a fixed-height parent hid\n them, which is why the docs' sizing example never showed either.\n- **`block` did nothing on eight editors.** The prop is declared once in\n `SvEditorProps`, and its own doc comment describes the symptom: \"in a form\n grid a row of inputs each stopping at a different width reads as broken\".\n `SvDropDownList`, `SvComboBox`, `SvMultiSelect`, `SvTreeSelect`,\n `SvGridSelect`, `SvAutoComplete`, `SvTextArea` and `SvDateRangeInput`\n accepted it by type and ignored it, so they stopped at their own 200-280px\n default while the text inputs beside them filled the row. `<SvForm>` was\n also not passing it to seven of the controls it mounts - both halves had to be\n fixed for a form row to line up.\n- **`cellDataType: 'dateString'` stored a timestamp after an edit.**\n Committing ran through the shared date coercion (`new Date(v).toISOString()`),\n so picking Christmas stored `2026-12-25T01:00:00.000Z` and the cell showed a\n timestamp beside neighbours showing plain dates - on a calendar day that\n depended on the user's timezone, since the conversion went through UTC. The\n picked date is now kept as the picked date.\n- **The filter row mounted a native date input** while the cell editor mounted\n the grid's own picker, so one column looked like two different products\n depending on where you touched it. Both use the grid's picker now.\n- **`editorType: 'number'` dropped the character you were typing.** The\n editor mounted `<input type=\"number\">`, whose value sanitization reports\n `\"\"` for anything not already a valid float - so `12.` read as empty, and\n typing `12.5` produced `125`. Same for a lone `-` and `1e`. It is a\n text input with `inputmode=\"decimal\"` and a digit filter now; the trade is\n the native spinner.\n- **Cell context menus and the comment editor floated at stale coordinates\n after a scroll.** Both are positioned at raw cursor coordinates and rendered\n `position: fixed`, but neither was in the scroll effect's gate, so they hung\n over unrelated cells. The comment editor SAVES on scroll rather than\n discarding - backdrop-click and Escape are the gestures that mean \"throw this\n away\"; a scroll is not, and must not destroy typed text.\n- **The footer summary row server-rendered as empty cells.** The totals were\n assigned from an `$effect`, and effects never run during SSR - so the server\n emitted a summary row that took up space and drew its border but held no\n numbers. They appeared only after hydration (a visible pop-in), and never at\n all for a reader with JavaScript off. A grid under the aggregation cell\n limit now derives its totals synchronously, so they ship with the HTML; a\n very large grid keeps the rAF-deferred effect so it still paints before it\n totals. `pnpm ssr:check` asserts the real aggregate now.\n- **The grid painted one unmeasured frame on every mount.** `hasMeasured`\n gates the custom scrollbars, the 16px scrollbar gutter on a trailing\n right-aligned column and the top pager (which carries a `border-top`), but\n it only flipped inside the `ResizeObserver` callback - and that callback is\n deliberately deferred by one `requestAnimationFrame`, so the browser was\n guaranteed to paint a frame with those missing and correct it on the next.\n On a first load that was the \"flashing scrollbar\"; in an app that recreates\n the grid per navigation (a `{#key}` around it, or a route that remounts) it\n showed up as the grid flashing on every sort / filter / page change, with a\n stray border line. The container is now measured synchronously in the mount\n effect - before the first paint - and the observer still handles later\n resizes. Covered by an e2e regression test, since jsdom has no layout to\n miss a paint with.\n- **Pager page-size trigger rendered as a stock button until first click.** The\n closed trigger is a placeholder until the lazy dropdown chunk loads, but its\n box styling only existed in that chunk's scoped CSS, so server-rendered and\n freshly-mounted pagers showed a UA-bordered, left-hugging `<button>` inside the\n bordered box. The placeholder now gets the same rules from `SvGrid.css`.\n- **Virtualized lists no longer flash blank on a fast scrollbar-thumb drag.**\n A fast drag can move the viewport into the windowed list's off-screen padding\n faster than JS can re-render the rows there; the padding now paints faint row\n skeletons, so `SvListBox` / `SvDropDownList` stay visually filled at any scroll\n speed. Also removed an app-wide `flushSync` from the scroll handler that could\n stall the very first scroll.\n- **Dock manager auto-hide now shows every tab of a multi-tab leaf.** Sending a\n leaf with two or more tabs to auto-hide created a single edge tab; it now\n creates one edge tab per pane.\n- **Ticking one row in a large grid took over a second.** Once any row was\n selected, every change walked all rows through the lazy row list to work out\n the header checkbox. At 1,000,000 rows (jsdom) one tick went from 1.3-1.8 s\n to 4-36 ms, and select-all from 4.2 s to 0.4 s.\n- **Tab never left the grid.** Every Tab was consumed and the cursor clamped\n at the last cell, so no key moved focus out (WCAG 2.1.2). Tab on the last\n cell and Shift+Tab on the first now leave the grid.\n- **No focus indicator when the grid is reached by Tab, or under forced\n colors.** The active-cell ring stayed hidden until a click, and Windows High\n Contrast removed it (it is a box-shadow). It now shows on keyboard focus, and\n forced colors get a `Highlight` outline.\n- **The cell editor had no accessible name.** The text, number and multi-line\n editors are now labelled with their column's header.\n\n#### Security\n\n- **CSV / formula injection in exports.** `exportCsv`, `exportTsv` and the\n chart's CSV download wrote text such as `=HYPERLINK(...)` as is, which a\n spreadsheet runs as a formula. Text starting with `=`, `+`, `-`, `@`, a tab\n or a carriage return is now written with a leading `'`; numbers and signed\n numbers in text form (`-12.50`) are unchanged. `escapeFormulas: false` turns\n it off; `copyToClipboard` takes `escapeFormulas: true` to turn it on.\n\n### @svgrid/enterprise\n\n#### Fixed\n\n- **Export on a page with no global `Smart`.** A bundler that inlines dynamic imports ran the\n vendored exporter at page load, before its namespace shim, and the whole bundle threw\n \"Smart is not defined\". The exporter now loads through an entry that installs the shim first.\n\n#### Added\n\n- **SvGrid Base: Delete forever can be one transaction.** `transaction` next to `execute` on\n `createSqlDataSource`, `createBaseEndpoint` and `baseFromDatabase`, and `linkedFrom` on\n `createKitHandlers`, make it one server step. See Undo, trash and import for each driver.\n\n- **SvGrid Base: Edit fields says what it sets off.** Past 10,000 records its question gives the\n number of automation runs the change can start, and \"Also run automations\" can be unticked.\n\n- **SvGrid Base: a comment not sent stays with its record.** Stepping to another record no longer\n carries it into that record's box, and closing the record keeps it and says so.\n\n- **SvGrid Base: `baseFromEndpoint` takes a live feed.** `realtime: (table, apply) => stop` is called per\n table and per junction table.\n\n- **Gantt: `pickUpToMove` and `text`.** Keyboard moves in the scheduler's way, and every\n word of the view for a page in another language. Base uses both.\n\n- **SvGrid Base: soft delete.** `softDelete: 'deleted_at'` on `createSqlDataSource` and\n `createSupabaseDataSource` marks a record instead of removing it. Reads, counts, groups,\n totals, lookups and rollups skip it, a restore clears the mark and rewrites nothing, and\n emptying the trash is the real delete. `createKitHandlers` takes `restore` and `purge`,\n inside the tenant's `scope`.\n- **SvGrid Base: a view always gets its answer.** When a data source cannot run a\n filter, sort, group or total on a computed field, or fails on one, the page reads\n the table a block at a time, works the field out with the cells' code and serves\n the view by id, with the count in the tab bar. A record that changes is updated\n where it stands, and taking the sort or filter away stops the read. `computes` on\n `baseSource` says what the source can work out itself; a `queryRecords` backend\n needs `computes: () => true` to keep those requests in the database.\n- **SvGrid Base: \"+\" opens a draft.** On a big table a new record is stored at its\n first value and not at the click, so one closed empty leaves nothing behind. It\n starts with the values of the view's filter, what the database fills in comes back\n into it, and one the database refuses keeps its values and tries again. \"+\" is in\n every view of records, and Ctrl+Z takes a stored record out again.\n- **SvGrid Base: add a field from the header.** `columnAdder` is on by default for\n whoever may change the schema, and the \"+\" stays at the right edge of a wide\n table. `createSqlDataSource` with `schemaChanges` adds the column on Postgres,\n MySQL and SQLite, `addField` on `baseSource` does it on your own source, and a\n source that cannot shows the `ALTER TABLE` to run. A new field never takes over a\n column already in the table.\n- **SvGrid Base: Undo keeps what it cannot write.** A record goes back with every column\n read, and a column the database generates is left to the database, on any engine.\n- **SvGrid Base, in use.** A record's position is the view's own (\"1 of 1,000,000\"),\n and Next and Previous read past what is loaded. A click or Enter opens a Board card.\n Ctrl K works from anywhere on the page. A group of mixed ticks shows as mixed, group\n headers show what the cell shows, and a field deleted from a view's filter leaves it\n (undo brings the condition back). A record outside the view's filter says so.\n- **SvGrid Base: select all means all.** On a big table the header checkbox selects\n every record of the view as a rule, and the count says so. The selection bar's\n Delete reads what the rule covers, asks with the exact number, and deletes in\n blocks with a Stop button. Up to 10,000 can be undone, in one write.\n- **Block writes.** `ServerDataSource` gains optional `createRows` and `deleteRows`;\n `createSqlDataSource` and `createSupabaseDataSource` implement both.\n- **`filterModel.searchIn`.** The fields a global search may look in. A base sets it\n so that a field hidden by a share or a permission cannot be probed by searching.\n- **Server row model:** `pin` and `unpin` show a row above the list that no block\n holds, outside the count, and `leafTotal` gives a grouped view its true size.\n- **Computed SQL:** dates, `MAX` and `MIN` on MySQL and SQLite, `FIND` on MySQL, and\n `ROUND` and `INT` on SQLite.\n- **SvGrid Base: a table has no size limit.** A table used to stop at `limit`\n records, 10,000 by default, and a table on `serverRows` kept that many in the\n page for everything but its grid views. Now no read stops short. A table that\n outgrows the page (`holdInPage`, the new name of `limit`) is read from the\n database a part at a time, by itself, and the page holds none of its records.\n Links, lookups and rollups read records by id, and a rollup over a link stored\n on the big table is added up by the server: in demo 723, a table of 200\n customers shows each one's orders and revenue over a million. The link picker,\n search across tables and the grid's search box search the whole table.\n Automations run on any record, and a scheduled one on every match. Share\n links show the whole table. A new record's autonumber is the table's next.\n New in the data source contract: `groupRawKeys`, for a link grouped by id.\n\n- **SvGrid Base: `createSqlDataSource` runs a view's filter.** It used to leave\n `filterModel.expression` out, so a filtered view on a SQL table showed every\n record. It also filters, sorts, groups and adds up on formulas, lookups and\n rollups now: `computedSql` (`@svgrid/enterprise/base/compute`) turns them into\n SQL, or says it cannot rather than answer differently from the cell.\n\n- **SvGrid Base: formulas in SQL say what a cell says.** More functions translate\n (`SWITCH`, text slices, `MOD`, `POWER`, `SQRT`, days and weeks on dates), and\n each is checked against a cell on real Postgres, MySQL and SQLite over blanks,\n negatives and quotes. `generatedColumnSql` writes a formula as a generated\n column for Supabase, and `generated` makes a filter read it.\n\n- **SvGrid Base: `schemaChanges` on `createSqlDataSource`.** On Postgres a field's\n type change is one `ALTER` and a table's copy one `CREATE`. Off by default.\n Without it, a big table's values are rewritten sixteen writes at a time.\n\n- **SvGrid Base: Supabase with aggregates off.** How a project starts. Kanban\n lanes and link counts now read their size with plain counts, and a total that\n needs aggregates says how to turn them on (`AggregatesOffError`).\n\n- **SvGrid Base: a link edits from either side of a big table.** A customer's\n orders tick in the picker, and each tick writes that order. Names pasted into\n a link cell are found on the server.\n\n- **SvGrid Base: changing a field's type on a big table converts its values.**\n It asks first, then uses `convertField` or rewrites them a block at a time.\n\n- **SvGrid Base: Download as SQL writes every record of a big table,** a block\n at a time. Duplicate table copies records that fit in the page, or in the\n database with the `duplicate` option.\n\n- **SvGrid Base: a Kanban lane keeps its order.** Sorted by a number field, a\n card dropped elsewhere in its lane writes that number.\n\n- **SvGrid Base: every view reads a whole table of millions.** Views other than\n grids used to show the first records of a big table.\n Now a Kanban lane shows its true size and reads its cards as they scroll by;\n a calendar reads the dates on screen, with each busy day's size and a \"+N\n more\" that lists all of that day; a Gantt reads its rows by start date; a\n gallery, list and inbox draw and read what is in sight; a chart and a\n dashboard have the server add up. It is the same `getRows` request a grid view\n sends, so a source that answers a grid view answers these. `queryRecords` now\n answers `aggregations` and `needsGrandTotal`. In demo 723 (a dev build, one\n laptop), each view of the million opens in under a second.\n\n- **SvGrid Base: a Rich text field type (`fieldType: 'richText'`).** It edits\n in a panel over the cell with a formatting toolbar: bold, italic, underline,\n strikethrough, bullet and numbered lists, links and clear formatting.\n `fieldProps: { toolbar: 'full' }` adds headings, quotes, code blocks, undo\n and redo. Ctrl+B / I / U format the selection and Ctrl+K adds a link;\n Ctrl+Enter, Tab or a click outside saves and Escape cancels. The value is\n HTML, sanitized when it is stored and again before it is drawn. A grid row\n shows the formatting on one line, a taller row height as lines, and the\n expanded record and the form show all of it. Search, the filter, sorting,\n copy, lookups and formulas read the words without the markup. Converting\n from long text renders its Markdown; converting back keeps a line per\n paragraph and list item. Demo 722 (release notes) shows both toolbars, a\n filter view and a formula over a note; demo 511 (content calendar) uses it\n for the brief. The long text option that renders Markdown is now labelled \"Format\n with Markdown\", to tell the two apart.\n\n- **SvGrid Base in your language.** Every word a base shows (the tab bar,\n the view bar and its panels, the record view, forms, the field editor and\n the field types, menus, notices, color and icon names) comes from one table\n of about 1,100 messages. `setBaseMessages(words)`, `<SvBase messages>` or\n `<sv-base>`'s `messages` property hand over the words you have; the rest\n stay English. Counts take `.one` / `.other` forms, and a `plural` rule\n for languages with more. `BASE_MESSAGES` lists every key with its English.\n A trash item for a table with one record now reads \"1 record\", not\n \"1 records\".\n\n- **SvGrid Base: Vote and Sparkline field types.** A vote field keeps who\n voted (user ids): a click on the cell, Enter or Space adds the signed-in\n person's vote or takes it back, through the grid's edit path, so undo and\n protected fields hold. It sorts by the count and filters as people (\"has any\n of me\"). A sparkline field draws a list of numbers as a line, area or column\n chart, typed and pasted as `3, 7, 4, 9`, sorted by the last value.\n\n- **SvGrid Base: a trash kept in your backend.** `<SvBase trashStore={{ list, add,\n remove }}>` reads the trash when the base opens and tells the store of every\n item deleted, restored, undone or emptied, so it survives a reload and is\n the same for everyone.\n\n- **SvGrid Base on a phone: boards and calendars.** A Kanban lane takes the\n table's width and snaps into place as you swipe; a calendar opens in its\n agenda. Gallery cards and forms were one column already.\n\n- **SvGrid Base: tables of millions of records.** `baseSource(ds, { serverRows: true })`\n has a table's grid views read it a block at a time through the server row\n model, with the view's filter (`filterModel.expression`), sort, groups and\n search sent to the server, and its edits, new records and deletes written by\n id. Kanban, calendar, gallery and the other views show the first `limit`\n records, and say so. Grouped views get one row per key with its count from\n the server, and the view bar counts the server's records. `queryRecords`\n answers such a request over records a Node server holds,\n `fieldFilterToSql` turns a view's filter into a `WHERE` clause for\n Postgres, MySQL or SQLite, and `fieldFilterToPostgrest` /\n `supabaseFieldFilter` into a PostgREST filter: `createSupabaseDataSource`\n has a new `expression` hook, and `baseFromSupabase` a `serverRows`\n option. A link edited there changes the field that links back for that\n record only (`linkChanged`), not for the records the page never loaded.\n Demo 723 pages through 1,000,000 orders.\n\n- **SvGrid Base: formulas on a server (`@svgrid/enterprise/base/compute`).**\n `computeRecords(tables, tableId)` works out every formula, lookup, rollup\n and count with the grid's own engine, and `evaluateFilter` /\n `filterRecords` test records against a view's filter, with no DOM and no\n Svelte, so an API, an AI tool or a report sees the values people see. A\n test works out a base both ways and compares them record by record.\n\n- **SvGrid Base: permissions.** `<SvBase permissions>` (and a table's own\n `permissions`) take `canEdit`, `canCreate`, `canDelete`,\n `canEditSchema`, `canEditViews`, `canComment`, `fields` (`'hidden'` or\n `'readOnly'` per field) and `viewsOnly`. The screen offers only what the\n person may do: no \"+\" row or Insert record without `canCreate`, no field\n editor or table menu without `canEditSchema`, view changes kept in the page\n without `canEditViews`. A hidden field takes the formulas, lookups and\n rollups that read it along, and the stored records and columns are never\n changed by what a person is not shown. `<sv-base>` takes the same as its\n `permissions` attribute.\n\n- **SvGrid Base: realtime changes update only what changed.** The\n `realtime` callback's `apply(changes)` takes inserts, updates (the whole\n record or just the changed columns) and deletes, or Supabase Realtime events\n as they come, and updates those records in place, a burst at once, with the\n same highlight and credit as before. `baseFromSupabase` uses it. `apply()`\n with nothing still reads the table again.\n\n- **SvGrid Base: a record history that lasts.** `setRecordHistory({ list, add })`\n reads a record's history from your audit log each time it is opened (a\n promise is fine) and hands it each change made in the page.\n\n- **Change review in the record view.** An expanded record lists its pending\n proposals under their fields, with Accept and Reject per field and for the\n record. `review.acceptRecord`, `rejectRecord`, `pendingIn`, `sources`,\n `pendingFrom` and `revertSource` (everything one agent did, across its\n sets) are new, and the panel has a From picker when more than one source\n has proposals.\n\n- **SvGrid Base: undo for tables, form footers and link allowlists.** Adding,\n renaming, duplicating, deleting and importing a table are undo steps.\n `formPrefill` takes a list of the fields a link may fill\n (`prefillFields` on `SvBaseForm`), and a form view takes `formFooter`\n text or `<SvBase formFooter>` as a snippet. `<SvBase onMention>` hears\n mentions. `<sv-base>` fires `cellchange`, `recordadd`, `recordsdelete`,\n `trashchange` and `mention`, and takes a `read-only` attribute.\n\n- **A grouped grid prints its groups.** `print()` printed a grouped grid's\n leaf rows only. It now prints a header row above each group, \"Region: EU\n (12)\", indented by level; `groupRows: false` leaves them out.\n\n- **`print({ open: false })` returns the printable document.** `print()` and\n `printGrid()` now resolve to the HTML they built, and `open: false` skips\n the print window and dialog, for a preview pane or anything that times the\n printable view. It is a plain table, not the grid.\n\n- **PDF: `pdf.fit`, `pdf.repeatColumns` and `pdf.font`.** `fit` chooses what\n happens when the columns do not fit across the page (shrink the text, split\n the columns over pages, or both, the default); `repeatColumns` says how many\n leading columns repeat on each page of a split. `font` prints with a font\n you supply, base64 or a URL, for scripts Roboto lacks (Chinese, Japanese,\n Korean, Arabic, Hebrew, Thai, Indic, emoji); without one the export now\n warns when it sees them.\n- **PDF export in a worker: `pdf.worker` and `@svgrid/enterprise/pdf-worker`.**\n pdfmake renders in one synchronous run, about a second per thousand rows on\n a desktop machine, and on the page that froze the tab. With\n `pdf.worker: () => new Worker(...)` and a worker script that calls\n `servePdfWorker()`, the document goes to the worker as data and the page\n stays responsive: a 5,000-row export took about two seconds while the page\n kept drawing at 60 fps, with progress through `onProgress` and an abort that\n terminates the worker. A PDF of more than 2,000 rows rendered on the page now\n logs a warning that points here. `buildPdfDocData` and `hydratePdfDoc` split\n the document builder for this; `buildPdfDocDefinition` works as before.\n- **Grouped PDFs total like the grid.** Subtotals use each column's\n `aggregate` (avg, min, max, count, countDistinct, a function) instead of\n always summing, a total row closes the table (`pdf.grandTotal: false` to\n leave it out), and group names are formatted as on screen. `getColumns()` in\n `@svgrid/grid` now reports each column's `aggregate` for this.\n- **PDF: `pdf.labels`, `pdf.watermark`, `pdf.author`.** Replace the footer's\n page and date text and the subtotal and total labels; print a faint\n watermark across every page; and fill the document properties, so a viewer\n shows the report's title rather than the file name.\n- **PDF column widths follow the text.** Each column gets at least its\n longest word and the rest of the page goes to the columns with the most to\n say, instead of equal widths: a 60-row, 8-column order list went from three\n pages to two.\n- **The pivot designer's Chart view grew up.** It offers every shape a pivot\n can take (bar, line, area, lollipop, pareto, the three radial forms, pie,\n funnel, waterfall, radar, heat map, stream) in the panel's groups, with\n Stacked, 100% and Horizontal where the type uses them, the chart's own\n toolbar (PNG, SVG, PDF, print, copy), its spoken summary and a legend\n under the plot. It had six types and a Stacked box.\n\n- **`pivotToChartSpec` carries the measure's format and name.** Pass the\n value chip's `format` and the chart reads \"$470k\" on the axis and\n \"$469,662\" in the tooltip, in the measure's currency and locale; percentage\n points get a formatter that appends the sign. With one measure the value\n axis is named after it (`yAxisTitle` overrides, `null` for none). The\n designer's Chart view passes the format when its measures agree, and\n formats plainly when a layout mixes a currency and a count.\n\n- **PDF export with charts and a KPI strip.** `pdf.charts` prints charts with\n the table: a rendered chart's element is rasterised on export through the\n grid chart's PNG path, or pass an image data URL; each takes a title, a\n caption and a width, and goes above the table or below it\n (`chartsPosition`). `pdf.kpis` prints a strip of headline numbers, each a\n label over a big value with an optional coloured delta line. Roadmap\n discussion 36.\n- **The selection bar renderer, plus bulk edit.** `enableSelectionBar()`\n (or `installEnterprise`) registers the renderer behind the free\n `selectionBar` prop. The bar carries a count chip, your actions, an overflow\n menu past `maxVisible`, and Clear; at phone width it collapses to icons and\n keeps an accessible name on each. **Edit fields** opens a drawer that edits\n one or many rows at once: a field the selection disagrees on opens blank and\n stays per-row unless you touch it, so applying to 40 rows does not flatten the\n values you did not mean to change.\n- **Scheduler booking rules and conflict detection.** New model helpers\n (`overlapCount` for per-resource overlaps, `overlapsBands` for working-hours\n bands) back resource double-booking checks and business-hours enforcement.\n New demos: booking rules, financial trading hours, and HR shift coverage.\n\n#### Fixed\n\n- **SvGrid Base: one Undo or Redo at a time.** Redo pressed while a range Undo on a paged table was\n still writing read the records before the Undo's later blocks landed: the data ended undone or half\n pasted while the page said \"Redone: edit 6,000 records\". Now a second Undo or Redo while one runs is\n refused with \"Undo is still running.\" and the buttons show it busy. An Undo or Redo of more than one\n block runs in the paste's dialog with its count and Stop, leaving the page asks first, and what it did\n is said when it ends, with the records or cells it changed (not those a teammate kept, a deleted\n record or a refused write). A small paste's count is said once its writes have answered.\n The dialog covers every Undo and Redo of more than a block: a delete (its Undo says the records\n it put back, not those a teammate restored first), the records a paste added, and more than\n 1,000 cells of a table the page holds. A cell undo of a record the view no longer shows holds its\n turn until that record's write has answered.\n- **SvGrid Base: Undo of a new record before its insert answers.** The step was a cell edit until the\n database answered, so Undo wrote an empty name and the record stayed. The \"add record\" step now\n exists from the moment the row does, what is typed into the record folds into it, and Undo waits for\n the insert, then takes the record out. A refused new record's Activity no longer shows what was typed.\n- **SvGrid Base: Try again from a record that fell to the end.** It wrote the first line into that\n record and offered to add the rest as new records. Now it writes nothing and says how many rows stand\n below the record against how many the first try covered; the dialog of a Try again names the rows it\n writes, which it covers on screen.\n- **SvGrid Base: a range over 1,000 records on Supabase or PostgREST is written.** The ids check read the\n whole range in one request, and a server that answers 1,000 rows at most (`db-max-rows`, the Supabase\n default) gave fewer back: every paste, clear, fill, one value or copy of more than 1,000 records said\n \"Records moved\" and wrote nothing. A short answer now reads on, a page of the server's own size at a\n time, and only a different id sequence counts as moved; a source with no cap still answers each check in\n one request. These checks select the key alone on Supabase and PostgREST. A download, a duplicated table\n and a view's group reads from a source that sends no count also read on past a short answer.\n- **SvGrid Base: a Cut the browser refused says so.** It did nothing in silence; now \"The browser would\n not put the cells on the clipboard, so nothing was cut.\"\n- **SvGrid Base: a notice that stays no longer covers \"+ Add record\" or a record's Delete and Close.**\n Notices in the bottom-right corner start above the button, or an open record's footer, where they\n meet.\n- **SvGrid Base: a table named by its key keeps its records.** A table with no text column shows\n its key as its first field. A record added there was deleted and created again under a new key at\n the next change of any record, a tab switch or a sort, on Postgres, MySQL and SQL Server: a foreign\n key with `ON DELETE CASCADE` took its children with it, \"record created\" automations ran each time\n and a table that marks deletes filled its trash. Edit field on the key could change its type, which\n wrote every record again under a new key, and offered \"Who can edit its values\". With a feed, a new\n record's key stayed blank and the page counted its own insert as a teammate's change. The key field\n now holds the key the database gave the record, in the key's own type, from a read, a write's answer\n or a feed; no edit, paste, fill, undo or field change writes it. Edit field on it changes its name\n and description only, and Duplicate field makes an ordinary field. A key column named `id` is the\n field `id_key` in the table's description, which `rename` maps back to `id`.\n- **SvGrid Base: a table with `customer` and `customer_id` opens.** The link from `customer_id`\n took the name `customer`, so two fields shared one id: the tab threw, showed the previous\n table's records, and the link and its field back were lost. Two keys that read as one name\n (`owner` and `owner_id`) did the same. Now a link whose name is taken keeps its column's name as\n its id and is headed \"Customer (link)\", or by its column when both are links (\"Owner id\"); both\n fields show, and each link keeps its field back.\n- **SvGrid Base: a client's filter on the tenant column is combined with the scope.** The scope\n replaced it, so `tenant_id = 2` asked by tenant 1 answered all of tenant 1's rows, and an\n `updateWhere` with that filter changed them. The two are combined now: another tenant's value\n finds no rows and changes none.\n- **SvGrid Base: a refused write names the record as stored.** A refused edit of the name field\n named the record by the value that was refused (\"Could not save Label of boom\"). A refusal by a\n foreign key of three columns named no field when the edit was to one of them: it names the key's\n fields now. The top bar's \"Not saved\" names them too.\n- **SvGrid Base: the status line opened in full shows no transport text.** It showed\n `svgrid kit: 409 Conflict - {json}` after the reason. It shows the reason, and the server's own\n words when it sent some that the reason does not say. A line with nothing more still opens, to read a\n reason the bar cuts short. Notices that quoted an error (a delete, a copy, a bulk edit, the\n trash) say the reason in words in place of the endpoint's or the REST API's status and body.\n- **SvGrid Base: a new record of a table whose database makes the key is read by no one before it\n has its key.** On a paged table, a draft (or a record whose insert was on its way) was asked for\n by its page id, and so were the records linking to it: Postgres refused the read on an integer\n key (\"invalid input syntax for type integer\"), and that text showed as a notice. Such a record is\n left out of those reads until the database has given it its key. A read that fails says so in\n words; the driver's text goes to the browser console.\n- **SvGrid Base: Duplicate field on a database table the base cannot add a column to copies\n nothing.** It copied the values into a column that was not there, so their writes could not land.\n It now says the copy needs a column and gives the statement that adds it.\n- **SvGrid Base on a REST API: the API's reason is shown, and a refusal of the values offers no Try\n again.** `createRestDataSource` passes `{ error }` or `{ message }` on as the reason, so a `422`\n says \"The server said: Amount must be 0 or more.\" where it said that no reason came. A server's\n words that do not end a sentence get a full stop, so \"You entered: -500\" after them reads apart. A `400`,\n `409` or `422` from any source offers no Try again, which would send the same values.\n- **SvGrid Base on Postgres: a foreign key to a table of another schema on the search path is a\n link.** `public.notes.task_id -> app.tasks` stayed plain text with both schemas on the path and\n both tables in the base. A key now links when its table is the one its bare name reaches. A key\n to a table that an earlier schema's table of that name hides stays plain, in its own schema too.\n- **SvGrid Base: a grouped paged view shows your own changes in its open groups.** Round 7 made a\n teammate's change read every open group again; every other refresh still read only the group\n rows. With no feed, a record you deleted stayed listed in its open group (and typing into it said\n it no longer exists), a record you restored from the trash or brought back with Undo did not show,\n and a new record, an edit that moved a record to another group, a renamed option, a converted field\n and the read after \"Records moved\" left the open groups as they were, while the counts changed.\n The server row model's `refresh()` with no route now reads the group rows and the loaded records of\n every open group on a grouped view, as `refresh({ all: true })` does, so every caller gets it.\n `refresh({ route: [] })` still reads the top level alone. On a grouped view or tree data an app's\n own `refresh()` now also reads the loaded blocks of each open group or node.\n- **SvGrid Base: a teammate's change to a sorted field takes its place in a paged view.** A feed that\n sends the row patched it where it stood, so a record whose sorted field changed stayed out of order\n until something else read the view. A change to a field the view sorts or groups by now reads the\n view again (every open group, when grouped); a change to any other field still patches the row and\n reads nothing. A view sorted or grouped by a formula or lookup reads again on any change, since a\n change to a field it reads moves it too.\n- **SvGrid Base: a face is never another person's while it loads.** User, created-by and\n last-modified-by cells reuse their faces by position on a scroll, and the browser kept drawing the\n last person's picture beside the new name until the new picture loaded. Another picture now gets\n a new `<img>`, empty until it loads, as attachment thumbnails do; the same picture keeps its\n element. The attachment preview's Next and Previous do the same.\n- **SvGrid Base: a checkbox cell is named by its field.** It read \"Checked\" or \"Not checked\" in\n English whatever `setBaseMessages` set, and a screen reader heard the state twice. Its name is the\n field's name (`fields.typeCheckbox` when the field has none), and `aria-checked` says the state.\n- **SvGrid Base: an image in rich text stays on the cell's line.** Under a reset that makes `img` a\n block (Tailwind's preflight), an image in a rich text cell took a line of its own: the text after it\n was cut off, or ran into the next row.\n- **Scheduler: a move keeps the event's length by the clock.** A drag, a keyboard carry, a timeline\n drag, the other selected events of a move, a series edit, Duplicate and a new event at a slot\n added real minutes to the start. On the night the clocks go back a one-hour event dropped on 1 am\n read and was stored as 1am - 1am, and a 45-minute one dropped on 1:30 ended at 1:15. The end is\n now the start plus the event's length by the clock, the length it had when picked up. An end the\n clocks skip lands after the skip, as a start does: on the night they go forward a one-hour event\n dropped on 1:30 reads 1:30am - 3:30am, one hour in real time. A timeline month drag over a\n 25-hour day ended an hour past midnight; a new all-day event on a timeline month was 24 hours,\n not one day. Both now keep whole days.\n- **Scheduler: the Agenda and the tooltip say when an event over several days ends.** A timed event\n from Monday to Friday read \"12am - 12am\"; it reads \"12am - Oct 30, 12am\". An all-day one over several\n days shows its first and last day instead of \"all day\".\n- **Scheduler: a keyboard carry lands at the time carried.** Each arrow stepped from where the last\n one landed, so an event picked up on Saturday at 2:30 am showed 3:30 am on the Sunday the clocks\n skip 2:30, and stayed at 3:30 am on Monday and after. Every step now counts from where the event\n was picked up (days by the calendar, slots by the clock), and a slot step always goes the way\n pressed. With 15-minute slots a step down from 1:45 am on that night landed at 3:45 am; it lands\n at 3:00 am.\n- **Scheduler: the event in hand stays in view.** A carry by the arrow keys never scrolled the\n calendar: the event went under the sticky header and out of sight. The day, week and timeline\n views now scroll with it, and a Gantt chart scrolls with a bar moved by the arrows.\n- **Scheduler: the top edge of an event that ends at midnight resizes it.** Dragged from 11 pm to\n 10 pm, an 11pm - 12am event became 12am - 12am, a whole day, and was written so. Its end is now\n read as 24:00 of its day.\n- **Scheduler: Duplicate can copy the event.** It called `onEventAdd` with the times alone, so the\n copy was a blank \"New event\". `onEventDuplicate(source, start, end, resourceId, allDay)` gets the\n event it copies. In a base Duplicate on a calendar event copies the record, as a card's Duplicate\n does. A whole-day event with no end field copied onto its own day; it copies onto the day after.\n- **SvGrid Base: keyboard focus shows on menu items under forced colours.** The \"All tables\" menu,\n the table menu and a summary's menu drew focus as a tint, which a contrast theme removes. They draw\n a `Highlight` outline there.\n- **SvGrid Base: a checkbox change is said in words.** A screen reader heard \"Done changed to true\";\n it hears \"Done changed to checked\" (`viewbar.valueChecked` and `viewbar.valueUnchecked`), and the\n record's Activity and the Inbox read the same words. A checkbox cell is `aria-readonly` only when it\n cannot be changed.\n- **SvGrid Base: the tenant column never names a record.** A table under `tenantColumn` with no\n `name` column took its first text column, the tenant, as every record's name: link chips all read\n \"A\" and errors said \"Could not save Site id of A\". The primary field is now the first text column\n that is not the tenant, not a key column and not a foreign key column (one that links nothing, a key\n of three columns or into another schema, included). With none, the record's key names it and shows\n as the first column, read-only, pinned and never hidden, moved or deleted, as any primary field is;\n links, errors and the record view read that name. An integer `tenant_id` placed first is passed over\n the same way, and the tenant is never the first column. `EntitySchema.foreignKeyColumns` lists every\n foreign key column a database read found.\n- **SvGrid Base: an empty database opens as \"No tables yet\".** `createBaseEndpoint` with `allTables`\n answered a database with no tables with a `403`, which the scaffolded page read as \"set\n `authorize`\", in development too. It now answers `200` with no tables and reads the database again\n on the next request, so tables made by `db:setup` show on a reload. On Postgres `allTables` listed\n only `public` while every query reached tables through the search path: it now lists the tables of\n `current_schemas(false)`, the first of a name winning as `to_regclass` resolves it, and the svgrid\n comments, the key default and the column types of a table come from that same table.\n- **SvGrid Base: a refused write says what is known.** A foreign key refusing a write (a value naming\n a record that is not there) read \"Other records depend on it.\" with Try again: it now reads \"The\n record Site id points to is not there.\" with no Try again, and the endpoint sends `409` with\n `code: 'link-missing'` and the key's `fields` when the database says which. A delete keeps\n `linked`. A write the server answered with an error (`request-failed`, a trigger's `RAISE\n EXCEPTION`) read \"The server gave no reason, so it may not have been stored.\"; it now reads \"Not\n saved: The server answered with an error. Whoever runs it can see why in its log.\" Only a write\n with no answer says \"may not have been stored\". New messages `shell.why.linkMissing`,\n `shell.why.linkMissingField`, `shell.why.linkMissingFields` and `shell.why.serverError`;\n `shell.why.unknown` has new words.\n- **SvGrid Base: a new record the database refused takes no second create.** Add row opens the new\n record's first cell; when its insert failed (an id column too narrow, say) while that cell was open,\n the row waited for the editor, and what was typed into it sent a create again (in a browser on\n MySQL, two more), each failing with its own notice. What is typed now goes with the failed record,\n for its Try again where there is one, the notice says \"Could not add\" with the name typed and the\n reason, and the row goes when the editor closes. A refused new record that stays listed (while another\n editor is open, or a draft on a table read a part at a time whose refusal no retry can change, such\n as an id column too narrow) takes no edit: a click, a double\n click, a typed key or a paste on it shows its notice instead of an editor, so nothing typed into it\n is dropped. An editor open on it when the refusal comes is closed with what it holds, which the\n notice names. A draft refused for a reason a retry may get past (no answer, a server error) keeps\n taking edits, and its next edit sends it again, as before.\n- **SvGrid Base: a query with no `pageSize` is a `400`.** It was a `500` (\"planToSql: limit must be a\n non-negative integer\"). `pageSize`, and `startRow` and `endRow` when given, must be whole numbers of\n 0 or more, or their digits; the answer says so.\n- **SvGrid Base: two foreign keys to one table give fields back named for their keys.** Customers got\n \"Orders\" and \"Orders 2\" for `customer_id` and `rev_customer_id`; they are now \"Orders (Customer)\"\n and \"Orders (Rev customer)\". One key alone keeps the plain name.\n- **`npx @svgrid/create`: the base template asks for current packages and has a favicon.** Its\n `package.json` asked for `@svgrid/grid ^3.0.11` and `@svgrid/enterprise ^3.1.16`, and it had no\n favicon, so the browser's request for one was a 404 on every page. It now has the SvGrid icon\n (`favicon.ico`, 16 and 32 pixel PNGs, and an Apple touch icon). The pivot dashboard and\n SvelteKit templates' floors moved up too.\n- **SvGrid Base: three records that move at once during a range read stop it.** On a sorted paged\n view, one teammate's edit that moved a record from the part read to past the end, one from the part\n not read into the part read, and one from past the end into the part not read, left every block's\n seam, the far end and the count as they were and repeated no record. A 6,000-line paste then wrote\n about 500 lines to the record beside the right one and said \"Changed 6,000 of 6,000 records\". The\n whole id sequence is checked now, in requests that ask for ids alone: a range longer than a block\n reads its ids first, in place of the record at the far end, and each block must bring the same ids\n in the same places; once the last block is read, the ids of the whole range are read again and must\n be the very ids read, in the same order. Any difference writes nothing. This holds for a range of\n one block too, and for copy, clear, one value and fill. A teammate's move inside the part already\n read now stops the job too, where before the paste went ahead. `createSqlDataSource` reads ids\n alone for these requests (`idsOnly`); a source that does not know the hint sends whole rows.\n- **SvGrid Base: Try again after \"Records moved\" goes from the first record selected.** In the same\n view it pasted by position: after a teammate's edit moved a record above the selection, every line\n landed one record down, a selected record got nothing and one outside the selection got a line. Try\n again now finds the first record selected by its id, goes from there over as many rows below it as\n before, as the list stands at the click, and selects those rows first so the screen shows what it\n writes. The notice says so: \"Records moved while this was read. Nothing was changed. Try again\n pastes from \"Acme\" down over the rows as they are now.\" If that record is gone or no longer\n among the rows shown, Try again writes nothing and says so. Opening or closing a group since counts\n as a change of the view. Paste, one value, clear, fill and copy alike. Messages `shell.rangeMoved`\n and `shell.copyMoved` are replaced by `shell.rangeMovedPaste`, `shell.rangeMovedValue`,\n `shell.rangeMovedClear`, `shell.rangeMovedFill` and `shell.copyMovedFrom` (with `{name}`): a\n `setBaseMessages` override of the old keys needs the new ones.\n- **SvGrid Base: Undo and Redo of a range say how many records kept a teammate's value.** Undo of\n a 2,500-record paste where a teammate had changed one record said \"Someone has changed this cell\n since\" before its blocks were written. It now says \"Someone has changed 1 of these records since,\n so it keeps their value.\" once the blocks are done. A small range undone cell by cell says \"1 of\n these cells\" in the same case, after its checks.\n- **SvGrid Base: a grouped paged view shows a teammate's change in its open groups.** A change from\n the feed re-read only the group rows: an open group kept showing old values with no sign they were\n old, a record moved to another group stayed listed under the old one, and with the changed fields in\n the feed the counts did not follow either. The group rows and the loaded records of every open group\n are now read again, so a value shows in place, a record whose group changed moves to its new group,\n and the counts follow. A feed that sends the changed fields patches the row at once and reads the\n groups again only when a grouped field changed. A group closed at the time reads again when it is\n opened. A read that fails keeps every group's rows and says so in the top bar, as an ungrouped view\n does. The server row model's `refresh({ all: true })` is what does it.\n- **SvGrid Base: Retry in a failed block takes the pointer under the pinned column divider.** Over\n the button the divider showed its resize cursor and its line, and the click relied on being passed\n on. The divider now lets the pointer through to the button while it is over it, and is the divider\n again once the pointer leaves. Its line is no longer drawn across a failed block's band, so the\n words and Retry are clear of it; everywhere else the line is as it was, under forced colours too.\n- **SvGrid Base: the top bar's full reason can be read on any device.** A narrow window cut the line\n short with an ellipsis, and the tooltip held only the server's own words, which a touch screen\n cannot show. A line with a reason is now a button: a tap, a click or Enter opens it in full,\n wrapped, with the server's words under it, and the bar grows only while it is open. Another press,\n or the notice going, closes it. The tooltip holds the whole line, then the server's words. A screen\n reader hears the line and the server's words once, from the button, which carries `aria-expanded`.\n- **SvGrid Base: each Restore in the trash is named by its record.** A screen reader heard a list of\n \"Restore\" buttons; it hears \"Restore Task 4\", as Delete forever already said \"Delete Task 4\n forever\". New message `shell.restoreItem`.\n- **SvGrid Base: \"All tables\" says which table is shown.** Its tables are radio items now, the shown\n one checked, so the tick is announced as well as drawn.\n- **Calendar: Day and Week on a day the clocks change.** Events were drawn an hour off their rows on\n a 25- or 23-hour day, in the page and in Base, and a drop or a slot wrote the time of the wrong\n row. They are drawn, dropped and written by the clock now. A carry by the arrow keys steps by the\n clock too: the night the clocks go back, 1:30 and one step is 2:00, not the second 1:00, and the\n event keeps its length. A Pro timeline's working hours on such a day, and the days after it, start\n at their hour; they were an hour early, and on a 25-hour day one day was counted twice.\n- **Calendar: an event over several days says its last day.** Its name gave the first day only, so a\n bar from Monday to Thursday was heard as a Monday event.\n- **SvGrid Base, WebKit with a finger: a tap beside a link selects, as everywhere else.** WebKit sends\n the press of a tap to the cell and its click to a link, chip or star a few pixels away. That\n click followed the link, opened the linked record or set the rating on the first tap. A click\n now acts only when the press began on the same thing; on a cell already selected a tap beside a\n link does what a mouse click there does, which opens the cell's editor, and does not follow it.\n- **SvGrid Base, forced colors: a ticked checkbox shows its tick.** The white tick sat on a fill that\n a contrast theme removes, so every ticked box read as empty. The tick, a given rating icon and\n your vote are drawn in the theme's text color.\n- **SvGrid Base: a menu item that is off is heard with why.** Download as Excel without jszip, and\n Delete table on the last table, are in the arrow-key order of their menu (`aria-disabled`).\n- **SvGrid Base: deleting the last record of a group in a list keeps focus among the records.**\n Focus went up to the group's header. It goes to the record after it, the first of the next group,\n or to the record before it when it was the last of the list.\n- **SvGrid Base: two records that trade places during a range read stop it.** A paste, copy, clear,\n one value or fill on a sorted paged view checked the ends of each block it read. When a teammate's\n edit moved one record into the part read and another out of it, the ends stayed put: one record was\n read twice, the other never, and the paste wrote a line to the wrong record and said \"Changed 6,000\n of 6,000 records\". A record read twice now counts as moved, and nothing is written.\n- **SvGrid Base: Try again after \"Records moved\" acts only in the view it came from.** Pressed after\n a switch to another view of the table, it pasted into the records at the same positions there,\n which nobody had selected. It now writes nothing if the view, its sort, filter, groups or shown\n fields changed, and says \"That view is no longer open, so nothing was changed.\" or \"The view has\n changed since, so nothing was changed.\" The same holds for the answer to a job's question: past\n 10,000 records, new records past the last row, and a delete.\n- **SvGrid Base: the read count of a range goes up a block at a time.** It read \"2,000\", \"3,999\",\n \"5,998\" records so far: each block after the first read one record of the last again and counted\n one less. Each block now brings 2,000 new records.\n- **SvGrid Base: the notice of a save that stopped part way goes with its Undo.** It stayed after the\n Undo and still said \"Undo takes these back.\"\n- **SvGrid Base: a bulk delete that finds records gone says so in one notice.** \"Deleted 2,598\n records\" with Restore and \"Deleted 2,598 of 2,600 records\" came together; now the notice with\n Restore says \"Deleted 2,598 of 2,600 records\".\n- **SvGrid Base: an edit of a record a teammate deleted reads \"This record no longer exists.\"** It\n read \"The record is no longer there, or you may not change it.\", which suggested a permission\n problem.\n- **SvGrid Base: records that link to each other in a circle delete in one go.** A delete block of\n `n1 -> n3 -> n2 -> n1` went level by level and found no record to start from, so Postgres and SQL\n Server answered \"Other records depend on this one\". Records in a circle now go in one statement,\n which Postgres, SQL Server and SQLite take. MySQL checks each row as it deletes and refuses them;\n the page says so in words, and that the link of one of them must be emptied first. Delete forever\n out of the trash does the same.\n- **SvGrid Base on SQL Server: the trash asked with no limit is its count alone.** It read one record\n where Postgres and MySQL read none.\n- **SvGrid Base on Postgres: a foreign key of two columns links again.** The catalog query paired\n every column of a key with every referenced one, so `(tenant, order_id) -> orders(tenant, id)` read\n as four keys: links showed empty and valid ones were refused. Keys are now read column by column in\n key order on every database, and only within the schema the base reads (a table of the same name in\n another schema, or a key into one, is no longer mixed in, on Postgres, MySQL and SQL Server). A key\n of the tenant column and one more is a link through that one within the tenant, and the tenant\n column stays the read-only tenant instead of an empty link field. Any other key of more than one\n column links nothing and is named in the log. On Supabase, PostgREST does not describe a key of two\n columns, so neither column is a link there.\n- **SvGrid Base: every link through a pair table, and the field that links back, is named for its\n pair table** once the plain name is taken: \"Tags (Ot rev)\", never \"Tags 2\".\n- **SvGrid Base: the All tables menu never squeezes an item.** With many tables, or with the Excel item\n that says jszip is missing, items shrank into each other; the menu now scrolls instead.\n- **SvGrid Base: importing an .xlsx without jszip says so in plain words**, as the Excel download\n does, instead of the developer's install command. `importData` and `readImportMatrix` reject with\n `code: 'no-jszip'`.\n- **Scaffold: dialogs, menus and the rich text editor use the app's font.** The `base`, `sveltekit`\n and `pivot-dashboard` templates set the font on the layout only, and what moves to `<body>` was in\n the browser's serif.\n- **SvGrid Base: a cold automation server judges every record of a block by what it held before.**\n The first write after the server started read the tables after that write, so a pasted block of 45\n \"Won\" ran the rule for one record, and could run it for a record that was Won already. The `audit`\n hook now has a `before` that `createKitHandlers` awaits before each write, so the first read lands\n first. Pass your own audit hook as `automations.audit('tasks', yourHook)` to keep it; a hook you wrap\n yourself loses `before`, and then each record of the first request goes by the fields the write named.\n- **SvGrid Base: a paged view whose read after a teammate's change fails keeps its rows.** One failed\n refresh replaced every loaded row with \"Could not load these rows\". The rows stay, the top bar says\n the table could not be read, and Try again reads them again. New on the server row model:\n `keepRowsOnRefreshError` and `staleBlocks` on each level. The top bar's Try again no longer slides out\n of sight on a narrow window, and a failed block's Retry under the pinned columns' divider takes the click.\n- **SvGrid Base: one feed for a base on an endpoint.** The docs showed one `EventSource` per table and\n junction table. On HTTP/1.1, which a Vite or Node dev server speaks, a browser opens at most six\n connections to a host, so a base of six tables left no connection for a write: \"Still saving\" for\n good. `baseFromEndpoint` takes `feed: (apply) => stop`, one connection whose messages name their\n table, and `createBaseFeed` shares one connection among `baseSource`s.\n- **SvGrid Base: the trash of the person who deleted follows a teammate.** A teammate's Restore and\n Delete forever left the record in the deleter's trash and its count, open or closed.\n- **SvGrid Base: focus stays off another record's Restore.** When a teammate's change took away the item\n holding focus in the trash, or turned its Restore off, focus moved to the next record's Restore, and\n Enter restored a record nobody chose. It goes to the list.\n- **SvGrid Base: a failed write gives the reason the server sent.** An endpoint that answered\n `{ \"error\": \"...\" }` without a code read as \"The server gave no reason\". The page now says \"The server\n said: ...\" for a write and \"The server answered with an error: ...\" for a read. The endpoint's own\n 500 sends `code: 'request-failed'` and keeps the driver's words in its log, as before.\n- **SvGrid Base: a range on a sorted paged view writes nothing when a teammate moves a record under it.**\n A paste, one value, Delete, a fill and a copy read the range a block at a time, by place in the view.\n A teammate's edit that moved one record during the read shifted the next block by one, so lines\n landed on the neighbours and one could land outside the selection, while the page said \"Changed 6,000\n of 6,000 records\". Each block now reads the last record of the one before again and checks it, and\n the record at the far end is read first and checked at the end. If one moved, nothing is written or\n copied, the page says \"Records moved while this was read. Nothing was changed. Try again.\", and Try\n again does the same action on the same cells, on the rows as they stand then. A double click on the\n fill handle that finds records moved says so too; it stopped in silence.\n- **SvGrid Base: new records keep the order they were made in.** `newRecordId()` put two random\n characters before its count, and its clock part ran over every 16.8 hours, so a paste past the last\n row of a paged table showed its new records shuffled. An id is now `rec`, the whole time, a count\n within the millisecond and four random characters: 19 characters, in creation order, and it never\n runs over. Ids made before stay as they are. **Breaking:** ids were 12 characters, so a text id\n column narrower than 19 (`varchar(12)`, `varchar(16)`) now refuses every new record. Widen it to\n 19 or more before you upgrade. The base says so in words: \"Record ids need 19 characters; this\n table's id column holds 16.\"\n- **SvGrid Base: the screen shows a range job when it says it is done.** After a range job or its Undo\n on a paged table, \"Changed 3,000 of 3,000 records\" came while the cells on screen kept their old\n values for a second or two. The rows on screen now take each block as it lands, also the blocks that\n landed before one failed.\n- **SvGrid Base: a record a teammate deleted no longer fails a scoped block on a table without soft\n delete.** Under a `scope`, a key no tenant holds, live or in a trash, is gone: a block skips it and\n finishes, one record by itself answers \"This record no longer exists\", and the key is not sent to\n the source. Another tenant's record, live or in its trash, is still a 403.\n- **SvGrid Base on MySQL: a parent deletes with its children.** Delete, and Delete forever, of a block\n holding records of a table linked to itself and the records that link to them failed on MySQL with\n \"Other records depend on it\". The block now goes a level at a time, the records that link first.\n- **SvGrid Base: a block's answer is used.** A paste whose values a server hook changed read each\n record again, a request a record: 400 for a 400-line paste. The answer is now followed in one step,\n and what a lookup needs of those records is read in one request.\n- **SvGrid Base: a rich text field renders under Trusted Types.** It was a blank page under\n `require-trusted-types-for 'script'`. A base with a field of every type now renders there, with its\n record view and rich text editor (`base-trusted-types.spec.ts`). The paths that still do not run\n under it are listed in SECURITY.md.\n- **SvGrid Base: \"Download this view as Excel\" says why when it cannot run.** Without the `jszip`\n package the item did nothing. It now says the app needs jszip, and a download that fails says so.\n- **SvGrid Base: a checkbox is a real TRUE or FALSE in Excel**, not the words `true` and `false`.\n `exportGrid` with `rawValues` writes a boolean value as a boolean cell too.\n- **SvGrid Base: a \"Could not delete\" notice goes once its records are deleted after all**, instead of\n staying beside \"Deleted 3 records\".\n- **SvGrid Base: two pair tables between the same two tables are two fields with two names.** Both were\n named after the other table, and a third got the second's field id. The second is now named for its\n pair table, as \"Tags (Order tags t)\".\n- **SvGrid Base on Supabase: `date` and `timestamptz` columns are date fields**, read by the Postgres\n type PostgREST names, as on the other databases. They were text.\n\n- **SvGrid Base: a keyboard move in a month calendar stays in sight.** A month drawn in five weeks reads a sixth\n that is not drawn. An event carried into it went out of sight, focus went to another record, Space picked that\n one up, and the first showed on a date it did not hold. The view now goes to the next month with the event, and\n a key on another event while one is in hand acts on the one in hand. An arrow across a change of the clocks moves\n a whole day: a week down used to land an all-day event on the evening before.\n\n- **SvGrid Base: a finger follows the mouse in a cell.** A tap on a link, email, phone or URL cell selects it, and\n the next tap follows the link, so on a tablet such a cell can be selected, copied and edited. A tap beside a\n checkbox's box or on a checklist selects the cell. A tap on the box, a star, a vote or a button acts at once.\n\n- **SvGrid Base: a selected list row or gallery card no longer looks focused.** It has the grid's selected-row fill\n and a tick by its title; the ring is focus alone. Under forced colors the tick stays.\n\n- **SvGrid Base: focus stays with the record.** After Undo of a board move focus goes with the card, not to the card\n that took its place, on a board read from a server too. After a confirmed Delete from a list it goes to the next\n record, not to its group. A note on why a record stays open (a comment not sent or not saved) is said aloud.\n\n- **SvGrid Base: group totals of a paged grouped view are right after each write.** After a\n read that a slow backend had not answered when the next write asked for another, every\n later write kept the old rows and totals on screen for 10 seconds, so the totals showed\n the write before. The block cache now frees a read's slot when it gives that read up (in\n `@svgrid/grid`, below).\n\n- **SvGrid Base: a paged view that cannot read its next part says so in view.** The tab bar\n says \"Could not load *table*\" with the reason and Try again, as a table read whole does,\n and the grid's \"Could not load these rows.\" with Retry sits on the first row wholly on\n screen, not on a row above it.\n\n- **SvGrid Base: a cold automation server sees its first change whatever the row carries.**\n An endpoint's row with a column no field names (a tenant, a sequence) made the first\n change after a start look like a change of nothing, and no rule ran.\n\n- **SvGrid Base: a paged group of records that link to nothing reads \"Empty\"**, as a lookup\n group does, instead of a blank label.\n\n- **SvGrid Base: an open trash follows a teammate's Delete forever and Restore.** It reads\n the table's trash again when the feed reports a change, and a record whose Delete forever\n stopped half way no longer reads \"just now\".\n\n- **SvGrid Base: `queryRecords` counts from the other side of a link again, and its \"is\n empty\" agrees with its groups.** A count or rollup over records that point at a\n customer read 0 when the tables given to `queryRecords` held no such customer. A link\n grouped by id now groups by the id it stores. A view filter \"is empty\" on a link now takes\n a key to a record the linked table does not hold, as the Empty group and the cell do.\n- **SvGrid Base: a paged table whose database stops answering no longer leaves an unhandled\n rejection behind.** The count read beside a grouped view's groups, the record read for a\n feed that sent only the changed fields, and the count read of a dashboard or calendar each\n failed with nothing waiting for them. A feed's record that cannot be read now makes the\n view read it again later, and an Undo of a cell the grid no longer holds says the database\n could not be reached instead of doing nothing in silence.\n- **SvGrid Base: a block write of 20 records or more on a table the page holds offers Try again,\n and shows what the server stored.** A paste, a clear or an Undo of that many records that\n failed said so with no Try again, and two failed blocks of 1,500 records read as \"2 changes\". Now the notice\n counts records and Try again reads them first, sends what is still as it was, and says how\n many cells and records it left. Each block's records are read back once it lands, so a\n trigger's or a hook's value shows at once, and Undo finds it.\n- **SvGrid Base: a new record not stored yet inside a range is no longer passed over in\n silence.** A range job left it out with no word, a paste over it shifted every line up a\n row, a fill from it emptied the cells it reached, and a copy dropped its line. Now the job\n says how many it left, a paste keeps its line for it, a fill takes its value, and a copy\n has its line.\n- **SvGrid Base: a range job on Supabase or PostgREST counts and stops within 200 records,\n and stops when a record fails.** Differing values go one request each there, and a block of\n 1,000 held the count and Stop for 1,000 requests. A failed request now stops the other\n fifteen at once, so the count of what landed is what the database holds, and a record a\n row level policy skipped is a failure, not a write that passed.\n- **SvGrid Base: Delete or Backspace in an open single select empties it.** A click on a\n selected select cell opens it, and the key then did nothing.\n- **SvGrid Base: Undo of a range or Edit fields on a server with a hook takes back what the\n hook stored.** It took the stored value for a teammate's and left it.\n- **SvGrid Base: an undo no longer writes over a teammate's change made a moment after yours.**\n On SQL Server and MySQL a write is read back with a second statement, and a teammate's write\n landing between the two was taken for the answer to yours: Undo then thought the cell still\n yours and put your old value over theirs. SQL Server now writes and reads in one transaction\n of one batch. MySQL does with the source's `transaction`; without one the gap stays, and the\n docs say so.\n\n- **SvGrid Base: day filters and a chart's months follow the viewer.** On `timestamptz` and\n `datetimeoffset` columns \"is 8 March\", \"today\" and a chart's months were the database's days.\n The page now sends each date condition with the moments its days start and end where the\n person is, and a chart reads each part of the span with one UTC offset in that offset. A date\n and a wall clock read as before.\n\n- **SvGrid Base: a table keyed by both ends is a link to many.** A pair table with both key columns foreign keys is a link on SQL and Supabase, under a scope too, and a removed link deletes that one pair. Before, on Supabase it was a tab whose delete took every pair sharing the first column, and on SQL it stopped the whole base. Any other composite key, or no key, leaves that table out with a log line and the rest opens. `keys` names a unique column for such a table or a view. A table or view with no declared key, even with a column called `id`, is now left out until you do.\n- **SvGrid Base: the page shows what the server stored.** After a typed edit, paste, Add row, record edit or undo the cell holds the row the server answered: a hook's value, a stamped tenant, a default. Undo goes back from that value. The tenant column shows and takes no edit, and hiding it no longer makes the table a `403`.\n- **SvGrid Base: a self-linked record and its parent come back together.** The link check of a block counts the keys in that block, and a block writes parents first.\n- **SvGrid Base: lists in `json` and `jsonb` columns.** A multi select, a vote or a sparkline saves on `pg`, `postgres.js`, MySQL, SQL Server and SQLite, cast by what the column is.\n- **SvGrid Base: SQL Server sorts, filters and groups on formulas and lookups in SQL.** What it cannot say is refused at once, where the table used to be read into the page.\n- **SvGrid Base: `maxPageSize` below 1,000 works.** The page follows the server's limit, on reads and on blocks.\n- **SvGrid Base: Edit fields on Supabase is a request per 200 records** for records given one value, and a row level policy that skips a record stops the block and says so.\n- **SvGrid Base: small server fixes.** A bad page or a private field name in a request is a `400`. A MySQL key past 2^53 links. A cross-site browser request is not believed for its `X-Forwarded-Host`. `isSecretColumn` knows `pass`, `userpassword`, `adminpass`, `apikeys`, `secretkey`, `pwhash`, `database_url`, `dsn` and `webhook_url`. The evaluation card builds no HTML string, so Trusted Types pass. A scaffold's theme script is a file, so a strict CSP passes. A PDF preview gives the focus back with the pointer over the file.\n- **Compliance docs.** No GPG fingerprint is claimed, one security address, canvas use and the print popup stated as they are.\n- **SvGrid Base: a table that cannot be read says so and stays said.** Where the records would be: \"Could not load Deals\", why, and Try again. No empty table, Add row or \"Not saved\" on the way.\n- **SvGrid Base: Excel files hold real cells.** Numbers, percents, dates and times are numbers with the field's format, whatever a database driver gave as text: on Postgres, MySQL and SQL Server, held or paged, with lookups and rollups. A yes or no field reads the same word on every driver. A link, a select or a person with a number for a key reads by name, in CSV too.\n- **SvGrid Base: a click to select a cell changes nothing.** The first click on a link, email or phone selects the cell, and the next one follows it. Ctrl or Cmd click follows at once and a double-click edits. A click on a checkbox's box, a star, a vote or a button acts at once; a click on the rest of the cell only selects it. A finger's tap follows the same rules.\n- **SvGrid Base: a calendar's day sizes are the viewer's days.** On a column of moments (`timestamptz`, `datetimeoffset`) the sizes and each day's list follow the page's zone, across a change of the clocks. A date or a wall clock reads the same everywhere and is told as it reads.\n- **SvGrid Base: a big change's automations show and stop.** The Edit fields question says how long the runs will likely take and what a closed tab means. The bar shows how far they are, with Stop, and the tab warns before it closes. Waiting runs no longer hold their records, runs on records in the page leave the tab its turns, and a run that cannot read its record no longer stops the ones after it.\n- **SvGrid Base: memory comes back after a bulk edit.** The page keeps the changes of the latest 20,000 records of a table, and no key per record written.\n- **SvGrid Base: a grouped view on a table read by parts reads like one in the page.** Groups in option order, drawn as the field draws them, with the records they hold at every level (a driver's COUNT as text included), and totals for the fields with a sum, an average, a lowest or a highest in the footer.\n- **SvGrid Base: Undo says what a drag did.** A bar and the bars it pushed are one step, and a card or a bar reads \"Undo move record\". \"Generate for all\" on an AI field asks first when records hold values, and one Undo takes the whole run back.\n- **SvGrid Base: a link kept in a `text[]`, `jsonb` or MySQL `json` column.** An `in` filter on one asks whether any value matches, so opening a Gantt on a self link no longer raises `malformed array literal`.\n- **SvGrid Base: nothing shows under Apply in Edit fields.** The row now covers the drawer's padding.\n- **SvGrid Base: a multi-column paste, clear or fill on a paged table writes every column.** It\n wrote the first column of each row and put the old value back in the others, with no message.\n- **SvGrid Base: a range on a paged table is the whole range.** Paste, one value over a range,\n Delete, fill and Ctrl+Enter reach every record the selection covers, loaded or not, through the\n Edit fields dialog: read and written in blocks, with progress, Stop, a count at the end and one\n Undo step. They used to reach only the rows the page held, and \"All changes saved\" said nothing\n of the rest. Up to 200 cells of loaded rows stay instant, with no dialog.\n- **SvGrid Base: a copy of rows not loaded reads them first.** It wrote blank lines and a $0.00\n for each. Up to 2,000,000 cells it reads with progress; past that it copies nothing, says the\n number, and points to Download.\n- **SvGrid Base: Undo and Redo of a range do not freeze the tab.** A big range is one step that\n reads and writes in blocks, and on rows held in the page a step's cells go back in one pass.\n- **SvGrid Base: a paste past the last row.** A line where no value fits makes no record. Past 10\n new records it asks first, with the number. They go in one block, and Undo takes them out in one.\n- **SvGrid Base: a paste while a view reloads makes no record.** It says the view is loading. A\n write that reads the view again leaves the rows on screen until the new ones arrive.\n- **SvGrid Base: a paste and a clear stamp last modified fields and the record's activity**, as a\n typed edit and a fill do.\n- **SvGrid Base: Undo takes back the Try again of what it undid.** A press on that button wrote\n the undone value, with no step to undo. Esc during a drag of the fill handle or a range now\n cancels it.\n- **SvGrid Base: Delete forever no longer cuts links on a record that comes back.** A teammate's\n Restore during the purge, a server failure at any step or a closed page could leave a restored\n record with its foreign keys empty, and a failed purge could overwrite a teammate's new link. The\n record is now claimed first, so no restore meets a half done purge, the keys are written only\n where they still hold what was expected, and the purge reports the ids it deleted.\n\n- **SvGrid Base: a restore is not a creation.** Restore and Undo of a delete ran \"record created\"\n server automations, which rewrote the record. The first change after the automation server starts\n is seen too.\n\n- **SvGrid Base: a record a teammate trashed no longer fails a block with \"no permission\".** Under a\n `scope`, a block skips it and finishes, as it does with no scope, and one record says it is gone.\n The audit log and `afterDelete` no longer name a record the delete did not change.\n\n- **SvGrid Base: a link to a trashed record is empty everywhere.** \"Is empty\", \"is not empty\", one\n Empty group with no raw key, sort, count and the picker, on held and paged tables. An undo to a\n record deleted forever says so, with no key and no Try again, and a link written to a trashed\n record says why it shows empty.\n\n- **SvGrid Base: your own edit with a lost answer is yours.** A read finds it landed: no failure, no\n Try again and no \"changed by someone else\". Undo of a hard delete no longer asks for a restore the\n table cannot do, and a write that was refused or found its record gone leaves no undo step.\n\n- **Board: a keyboard move writes once, on the drop.** The arrows only show where a card is going, and\n `onCardMove` and `onCardReorder` fire when it is dropped, from where it was picked up. Escape used to put the\n card back on screen while the record kept the new lane. Tab or a click elsewhere puts it back too, focus stays on\n the card, and on a `server` board the drop marker shows the way until the drop.\n- **Scheduler: Escape works past the edge of a view.** A carried event used to leave the view and take Escape with\n it. The view now follows the event, and Escape brings both back.\n- **Scheduler: an all-day bar in Week or Day takes the keyboard.** Space opened it instead of picking it up, and it had\n no name with its date. It now moves as any event does; up and down move it a week.\n- **SvGrid Base: a list read from a server takes Tab.** Flat or grouped it is one stop, on the record you were last\n on, and the arrows walk records, not the row buttons. Grouped lists read from a server also take Home, End, Right\n and Left. Right and Left swap in a right-to-left page.\n- **SvGrid Base: Enter on a list row's button opens its menu**, not the record, and a selected row shows it and\n says \"Selected\".\n- **SvGrid Base: focus no longer falls to the page.** The selection bar goes back to Delete after its question, and to\n where you were after Clear, Escape or a confirmed Delete. A sent form puts focus on its thank-you and says it. A\n dashboard tile moved by its buttons keeps focus on the button pressed, which at either end is off, so a second\n Enter does nothing.\n- **SvGrid Base: a comment's question and Next.** Escape on \"Delete this comment?\" leaves the question, not the\n record, and focus starts on Cancel. Next and Previous wait for a comment being changed, and say so, instead of\n dropping the text. A click outside a record with text not sent says why it stays open.\n- **SvGrid Base: small accessibility fixes.** A link chip in a list row is no longer a button inside a button, Add\n table keeps its 26 px on a phone, a mail link in a record is 24 px tall, and the Menu key cannot open the browser's\n menu over a card's. A record with no name reads \"Untitled\" everywhere.\n- **Edit fields on a big table no longer freezes the page.** It runs in the dialog a delete uses, with progress and\n Stop. Past 10,000 records it asks first, with the exact number. A failure part way stops there, reads the block back,\n says how many records changed, and keeps Undo for those; it used to carry on, say none were saved and leave no Undo.\n- **Bulk writes through `createBaseEndpoint` go in blocks.** `createMany`, `updateMany` and `deleteMany` carry up to\n 1,000 records a request, and records given the same values are one statement. `authorize`, `scope`, `validate`,\n `checkWrite`, hooks and `audit` hold for each record as on a one-record write.\n- **An edit, a paste, an undo or a redo deep in a big table is as quick as one at the top.** Each walked every row\n before it, several times over.\n- **A calendar on a date-with-time field groups by day in the database** (`groupDays`, on every dialect), and a month\n moved past stops sending its days. `groupMonths` works on SQL Server and SQLite too.\n- **`createSqlDataSource`: `updateRows` keeps the time** of a schema `date` over a timestamp on Postgres, where a cast\n dropped it, and writes records given the same values without a CASE. A block's statements carry the tenant.\n- **The link picker's Enter waits for the server's answer.** On a table a server searches it took a row of the last\n search, or made a record whose name was there.\n- **A table read a part at a time asks for 200 records a block by default,** not 100: a sorted read costs a sort of the\n table whatever it returns.\n\n- **`createBaseEndpoint`: a junction table under a tenant `scope`.** A junction holding the tenant\n stays a link to many with `tenantColumn`, and one holding none takes `{ through: 'order_id' }`, so\n a pair belongs to the tenant of that record on every path. A `null` for a junction between scoped\n tables used to share every tenant's pairs: it is now a `403`, and the log says what to do.\n\n- **`createBaseEndpoint`: a link stays inside the tenant.** A foreign key written through the route\n must name a record inside its table's `scope`, in blocks too. Another tenant's record and no\n record get the same `409` with `code: 'link-refused'`. Before, a tenant could point at another's\n record, block its delete, and learn which keys exist.\n\n- **Base: a table keyed by a secret does not start.** A key named like a credential reached every\n reader, because a key cannot be hidden. Such a table is now left out, the log says why, and\n `exposeSecretKeys` serves it by name. A foreign key holding that value stays out too.\n `isSecretColumn` also knows `passcode`, `user_pass`, `session_key`, `activation_key`,\n `remember_digest`, `hashed_pw` and `mfa_seed`.\n\n- **`baseFromSupabase` asks only for what it shows.** It read `*`, so a column `hide` held back still\n reached the page, and a table with a column `anon` may not read failed. `createSupabaseDataSource` takes\n `schemaColumns`, and a base sets it. A record brought back by undo gets a hidden column's default.\n\n- **Escape closes a PDF preview.** Chrome's viewer took the focus as it loaded, and keys pressed there\n never reached the page. The focus now comes back to the dialog, unless a person went into the file.\n\n- **`createKitHandlers` takes JSON from its own site only.** A `POST` that is not\n `application/json` is a `415`, and one from another site a `403`, on `vite dev` as well as in a\n build. A page on another host goes in `origins`. A caller of `handle` that sent no content type\n must now send one.\n\n- **Server logs hold no value a driver quoted.** The handlers log codes and names, and take the value\n out of the text: a SQL Server or MySQL duplicate key, a Postgres `detail`, mysql2's `sql`.\n\n- **A color from data cannot reach a style.** An option's color, a record's value used as a color and\n a label's color now go through `safeColor` on the Board, the scheduler, the Gantt, `SvBoard`,\n `SvRecordDetail`, `SvSchedule` and the automations dialog, and a spreadsheet file's formats through\n `entryToStyle`.\n\n- **`GET` keeps an automation's outbound details from those who do not need them.** A webhook or\n an email is described only to a person `authorize` lets change the table, and an automation that\n runs on the server sends no URL, header or address to any page. `sharedTables` drops them too,\n with what a view said of a hidden field.\n\n- **Print and the HTML export run no script.** A cell's link is `http`, `https`, `mailto`, `tel` or\n `blob` only, a rule's color cannot leave its attribute, and the document carries `script-src 'none'`.\n\n- **`baseToSql` keeps a title inside its comment.** A table named with a line break could start a\n statement of its own in the downloaded file.\n\n- **Compliance pages.** `gdpr.md`, `hipaa.md`, `soc2.md` and the compliance overview now say what\n holds for `@svgrid/grid` and what SvGrid Base adds. The doc site's host was wrong in `gdpr.md`.\n\n- **SvGrid Base: the fill handle fills by the field, not by the look of the stored value.** One\n source cell copies in every field, so a person, an option, a color or a time no longer counts up\n to the next one. Two or more source cells continue a series in number and date fields and repeat\n in every other. A value the field cannot hold is not written, and the page says so. Ctrl+D copies\n down, and is one undo step.\n- **SvGrid Base: a copy writes what the cell shows, and a paste brings the same value back.** Percent,\n duration, currency, date and time cells copy as shown, so Excel gets `50%` and `1:30`, and a paste no\n longer turns 50% into 1%. Inside a page the stored value comes back in full: 12.5% shown as 13%,\n bold in rich text, the right one of two people with one name.\n- **SvGrid Base: one multi-line cell in is one cell out.** Copy quotes a cell holding a line break or\n a tab, and a paste reads quoted text, Excel's included, so the record below is left alone.\n- **SvGrid Base: import keeps what a file says.** A date keeps its calendar day in every time zone, a\n percent reads as the grid reads it, a person matches by name or email, and a name that matches\n nobody is no longer stored as a person. Link fields can be mapped by the linked record's name. The\n dialog says how many cells it could not match. A type change from text to a date or a person reads\n the same way, and a rating stays within the field's maximum.\n- **SvGrid Base: a downloaded view reads back to the same values.** A percent is written `12.5%` and a\n duration `1:30:15`, where the stored 0.125 and 5400 would read as something else in an import, and a\n name with a comma in a list is quoted. In Excel a percent stays a number with a percent format, and\n an import reads an Excel cell as it shows: a percent as a percent, a date cell as that date. A select\n option is known by its label as well as its value, so an import adds no copy of it.\n- **SvGrid Base shows its own notices.** A host with no `<SvToaster />` now sees every failure, refusal\n and Restore. A host that has one keeps it, and nothing shows twice.\n- **SvGrid Base: Add record in a grouped view goes to the new record's first cell.** A new record in a\n group of its own opened no editor, so typing pressed the button again and made blank records.\n- **SvGrid Base: undoing a delete after a restore on another tab keeps both sides of a link.** Roll-ups\n on the other table were left short.\n- **SvGrid Base: smaller things.** Space on a Vote cell votes. A click beside the number in a record's\n number editor keeps the edit. Enter in the new view's name closes the view menu. The undo button and\n the screen-reader message name a fill, a paste or a bulk edit by its cells, and a redo is said as a redo.\n\n- **SvGrid Base: focus no longer falls to the page.** A dialog or menu whose opener went, or\n cannot take focus (a menu item, a row button hidden under the dialog, a field just deleted),\n hands focus to what held it, which is the grid, or to the next card. Pin, Hide field, Delete\n field, Expand record, Comment, a deleted card or calendar event, and Undo or Redo on the last\n step all leave the keyboard in place.\n- **SvGrid Base: Remove on a dashboard tile takes the keyboard to its question,** which is read\n out. A second Enter cancels; it no longer lands on the next tile's buttons.\n- **SvGrid Base: Escape in a comment keeps the text.** It leaves the box, and acts as Cancel on\n a comment being changed. Text not sent holds the record open against Escape, which then says\n why, and against a click outside.\n- **SvGrid Base: cards and list rows keep the record's actions.** At a phone's width, in a\n gallery and in a list, the Menu key, Shift+F10, a long press and a button on the card open\n expand, comment, select, insert, duplicate and delete. Selection shows on the card and in the\n selection bar. `gallery.cardMenu`, `gallery.isSelected` and `gallery.text` are new.\n- **Board:** \"Add card to Lead\" and \"+ Add a card\" show the focus ring. An empty lane no longer\n says it is a list. **Link preview, tooltips:** Escape closes a link preview, a grid tooltip, and\n a calendar or Gantt tooltip with the pointer still on it, and a pointer crossing the gap to a\n link preview keeps it open.\n- **SvGrid Base:** a rating cell's name and a gallery's name are translated. Run now with no\n record writes \"(no record)\" for a field token, not the raw `{Deal}`. A view bar button reached\n at a phone's width comes fully into sight. A percent bar keeps its fill under forced colors.\n\n- **SvGrid Base: a tenant's trash on MySQL and SQLite.** With `scope` and `softDelete` the tenant\n was bound before the ids, and `?` is matched by position, so Undo, Restore and Delete forever\n found nothing: a restore said the record was deleted forever, and a Delete forever reported\n success and left the row.\n- **SvGrid Base: a soft delete leaves foreign keys alone.** It used to clear them in other tables,\n and a restore after a reload or by a teammate never brought them back. The clearing now happens\n at Delete forever, in the tables of the base. A hard delete is as it was.\n- **SvGrid Base: a database's numbers show as stored.** Introspection reads a column's scale:\n `numeric(12,2)` holding 19.99 shows `$19.99`, not `$20`, and an integer shows no decimals. A column\n with no declared scale takes the places its values show. On SQL Server `money` has four. MySQL's\n `tinyint(1)` is a checkbox, a `time` column a time of day and never a select, and a Postgres\n `text[]` a multiple select that can be edited.\n- **SvGrid Base: an insert whose answer was lost.** A read says whether it landed: the record is\n taken, what was typed goes to it, and nothing false is said. A key the database makes is found by\n the record's values, in a table the page holds. A 5xx with no code now says the server did not\n answer, not that the database refused, and a delete or an edit that landed is not told as a failure\n or as a teammate's change.\n- **SvGrid Base: a refused undo shows the value it found.** Without a feed the cell kept the value the\n database no longer held.\n- **SvGrid Base: JSON keeps every digit.** A `json` or `jsonb` column is read and written as text, so\n a number past 2^53 is not rounded by a read, an undo or an edit.\n- **SvGrid Base: a notice goes with its base.** The toast queue belongs to the page, so a \"Deleted ... Restore\" or a Try again\n outlived the base that made it, and the next base on the page drew a button that did nothing. Restore now goes with its base, and a\n failure keeps its words, what was typed included, with no button.\n- **SvGrid Base: smaller things from teams.** Someone who may not delete hears no error from the trash,\n and the panel says why it is empty. A permission refusal offers no Try again and leaves no undo step. A\n paged view with a filter drops a record a teammate moved out of it. A teammate's restore lands at the\n record's place. A Delete forever takes the \"Deleted ... Restore\" notice with it, and a number deleted\n in this page is not given to the next record while it stays open.\n\n- **BREAKING: `scope` refuses an empty tenant.** A value of `''` used to drop the predicate,\n so a person with no organisation read, rewrote and deleted every tenant's rows. Now a\n resolver that returns nothing, or a rule whose value is empty, blank or not a single value,\n is a `403` on every read and write, and `createBaseEndpoint` leaves the table out of `GET`.\n Only `null` means no scope. `scope: (table) => ...` must return `null` for a table every\n tenant shares: returning nothing now closes it.\n- **SQL Server runs.** `dialect: 'mssql'` wrote `LIMIT` and `RETURNING`, so nothing could. It\n now pages with `OFFSET ... FETCH`, reads a key back with `OUTPUT INSERTED ... INTO` (a table\n with triggers works), writes an identity key under `IDENTITY_INSERT`, stays under 2,100\n parameters and 1,000 rows, and keeps `datetime`, `datetimeoffset`, `bigint` and `decimal`\n exact. Run against SQL Server 2022. Formulas are worked out in the page there. Studio's\n connection string for it was a URL the `mssql` driver cannot read.\n- **BREAKING: a PDF preview frames only bytes it read.** A file on another origin went into the\n frame as it was, and could draw a sign-in form under the base's own title and move the tab.\n A file the page cannot read, because its host sends no CORS headers, now offers a download.\n- **An id from a client reaches the source as text.** A number was bound as a number, and on\n MySQL with a loose `sql_mode` `id = 0` matched every id not starting with a digit. An empty\n id is a `400`.\n- **`patch: { id }` no longer changes a record's key.**\n- **A webhook connects to the address it checked.** A name could answer one way for the check\n and another for the connection. On Node the call now goes to the checked addresses, and a\n name that does not resolve is not sent. More ranges are refused: `192.0.0.0/24`,\n `198.18.0.0/15`, the documentation blocks, `fec0::/10`, and IPv6 forms wrapping an internal\n address (`::a.b.c.d`, NAT64, 6to4, Teredo).\n- **`isSecretColumn` knows more names.** Any `token` (`remember_token`, `token_hash`),\n `pass_hash`, `password1`, `pin`, `recovery_codes`, `credit_card`, `iban` and others. It reads\n the name only, and the docs now say that is a default, not a guarantee.\n- **`createBaseEndpoint` takes `validate`, `hooks` and `maxPageSize`.** Writes through it could\n not be checked on the server. `validate: true` asks nothing of a `NOT NULL` column with a\n default, and a schema read from a database no longer marks such a column required.\n- **A refused write says why, through an endpoint too.** A constraint was a `500` and \"The\n database refused it\". It is now a `409` with a `code`, and the person reads which rule it\n was, without the driver's words. An autonumber clash through an endpoint now takes the\n next number, as it does on a direct source.\n- **Thrown text stays on the server.** A message thrown by `authorize` or `scope` reached the\n client, connection strings included. It now reads `forbidden`, unless the `Error` has\n `expose: true`.\n- **A read has a ceiling.** `createKitHandlers` returns at most `maxPageSize` rows (20,000 by\n default) with the true count, so `pageSize: 1000000000` no longer reads a table in one go.\n- **An alert rule's style is a color.** `red;background-image:url(...)` fired a request to\n another host each time a cell painted.\n- **A group key that is an object is refused.** A driver could write one as SQL of its own.\n- **A filter on a number past 15 digits compares exactly.** Rounded, a tenant id could be\n someone else's.\n- **A scaffold's production build says what is closed.** It read \"The database did not answer:\n 403. Is DATABASE_URL set\", which sent you to the wrong fix.\n- **Two dev console warnings are gone**: `state_proxy_equality_mismatch`, once for each field\n of each table, and `ownership_invalid_binding` on `views`.\n- **\"Not in view (1 record)\".** It read \"1 records\".\n- **Forms take the mouse again.** Typed boxes ignored a click or a tap, and so did an\n open editor in a record. A star clicked in a form is now the answer; it used to open\n the editor.\n- **A record's Account picker no longer links what nobody chose.** Enter on the list as it\n opened ticked its first record, and Done saved it. Enter now closes it unchanged, and\n a picker left as it opened writes nothing.\n- **Focus no longer falls to the page.** Comments, a form's pickers, the form builder,\n dashboards, filter conditions, automations, attachments, import, the inbox and the\n field editor keep the keyboard on the logical next control, and comment changes,\n tile moves and form builder moves are said aloud.\n- **Base has the fill handle its docs promise.**\n- **The form builder's field list holds still.** A field turned off keeps its row, and back\n on it takes the place the row shows.\n- **Timeline.** Delete on a bar asks first, as does Delete in a calendar's menu. A bar moves by\n Space and the arrows with one write, stands clear of the task table when focused, opens\n its menu with Shift+F10 with focus in it, reads as a table, keeps its shape in forced\n colors, and every word goes through the base's messages. A drop on a weekend stays\n there.\n- **Attachments: Enter on Remove removes the file.** It opened the preview.\n- **Dashboards ask before a tile is removed.**\n- **Form number boxes read as the field does** (\"25%\", \"$1,250.50\") and edit in its units.\n- **Escape in a number or a rating in a record leaves the box.** It closed the record.\n- **Typing on a focused record field starts the edit**, and Enter in the field type search\n takes the first type.\n- **A double-click on a week's all-day row adds an all-day record.**\n- **Automations:** step buttons name their step, moves are said, and the open automation\n carries `aria-current`. **Delete record** ties its text to the dialog.\n- **SvGrid Base: Edit fields writes to the ticked records and nobody else.** In a sorted,\n filtered or grouped view each value that landed moved its row, and the next one went to\n whatever had taken its place, so records nobody ticked changed and Undo wrote more wrong\n values. Every cell is now worked out first and written by record, in one write and one\n undo step, on tables in the page, on SQL tables and on tables read a part at a time,\n where Edit fields did nothing at all before. Text a field cannot read, such as `abc` in a\n number, stops the form with a message and writes nothing.\n- **SvGrid Base: Delete, paste and fill stay on their records.** A row held under a typed\n edit but moved at once under Delete or a paste, so a second press hit another record.\n The row now holds until the cursor moves on, whatever wrote it.\n- **SvGrid Base: a paste past the last row adds records,** as one undo step with the cells\n it wrote. It used to drop those lines without a word.\n- **SvGrid Base: one rule for pasted values a field cannot read.** A number, a select, a\n date, a person, a link, a rating or a checkbox given text it cannot read keeps its value,\n and one notice says so. An unknown link name no longer wipes the link, and an unknown\n name no longer lands in a person field as a person nobody has.\n- **SvGrid Base: a write with no answer is not given up on.** A write past `writeTimeout`\n stays out: later writes to its record wait behind it, writes to other records go on, and\n the status says so. It can no longer land late over a newer value, a late insert leaves\n no second record behind a Try again, and a late delete keeps its copy in the trash until\n the outcome is known. A rollback now leaves a value written after the failed one.\n- **SvGrid Base: the undo guard fails closed.** A cell undo reads the record first, with a\n feed or without one, and is refused with a message when that read fails, so a feed down\n without a word can no longer take a teammate's value out. The step stays the next to try.\n- **SvGrid Base: Try again checks first.** It reads the record, leaves a value that changed\n since, whoever changed it, and says so, also when the record was deleted.\n- **SvGrid Base: a failed read before a delete stops the delete.** That block stays, and\n the person is told, instead of going ahead on a stale copy.\n- **SvGrid Base: editors in a record are held for.** A teammate's change waits while a field\n of the record view is being edited, as it does for a cell.\n- **SvGrid Base: fewer false \"changed by someone else\".** A number the database rounded, a\n person's own undo on a table read a part at a time, and a feed's word on an earlier write\n of the same record are no longer reported as a teammate's. A read that started before a\n write no longer puts the old value back. Undo refused \"alone\" after 1.234 in a column of\n two decimals goes through.\n- **SvGrid Base: undo and redo run no automation.** A redo of a status change added its\n checklist items again on every press.\n- **SvGrid Base: a second restorer is told nothing is wrong.** Two people restoring the same\n record at once: the second sees it back, not \"Another record already has this value\".\n- **SvGrid Base: wrong words on the history.** A column resize read \"Undo cell edit\", and a\n redo was logged as \"undid\".\n- **SvGrid Base: soft delete keeps a record's junction links.** A delete of a record marked\n with `softDelete` dropped its junction rows and a restore did not put them back, so tags\n were gone after a reload. Junction rows now stay, at either end of the link, and a restore\n writes nothing. A marked record shows in no link. Delete forever removes them with the row.\n- **SvGrid Base: Delete forever is final, and only once the database says so.** A purge the\n database refused left the record marked with no way to reach it, under a notice saying it\n was in the table. The item now stays in the trash until the purge is confirmed, the notice\n says so, and in one Empty trash the free records still go. Ctrl+Z could put a purged record\n back from the page's copy: the undo step now ends, and a stale undo in another window\n refuses. A single Delete forever now asks first, as Empty trash does.\n- **SvGrid Base: the trash of a soft delete table reads the table.** Marked rows show after a\n reload with no trash store, for whoever may delete. **Behaviour change:** those rows appear\n in the trash, those marked by other tools too. They do not age out: `trashDays` still only\n purges what the page's trash holds.\n- **SvGrid Base: `softDelete` on `createBaseEndpoint`, `baseFromDatabase` and\n `baseFromSupabase`.** The endpoint could not use soft delete at all. The column stays out\n of the fields, and the description tells the page.\n- **SvGrid Base: the mark cannot be written.** With the column on the schema a client could\n create a record already deleted, or mark one through an update. `createSqlDataSource`,\n `createSupabaseDataSource` and `createKitHandlers` now leave it out of every write.\n- **SvGrid Base: teammates hear of soft deletes.** A restored record reached a table read a\n part at a time only on a reload-style feed, and a delete through `createKitDataSource` stayed\n on teammates' screens. `createKitHandlers` now says which column marks records and the page\n learns it. `apply({ type: 'update', id })` reads that one record, which is how a feed reports\n a junction change; `baseFromSupabase` listens to junction tables by itself.\n- **SvGrid Base: a new record does not take a marked record's number.** On a table the page\n holds it did, and the restored record was renumbered. Autonumber now reads past marked rows\n on Kit and Supabase too.\n- **SvGrid Base: a very large soft delete purges a block at a time.** Closing the page used to\n leave everything not yet purged marked.\n- **`createKitHandlers`: restore and purge tell the server what happened.** **Behaviour\n change:** the `audit` hook now gets `restore` and `purge` actions, one entry per record that\n really changed, in place of `update` and `delete` with ids joined by commas, and a purge of\n another tenant's id logs nothing. `beforeDelete` and `afterDelete` run on a purge with\n `purge: true`, and `beforeRestore` and `afterRestore` are new. On a table with no `softDelete`\n `restore` and `purge` answer 405.\n- **SvGrid Base: formulas and lookups through `createBaseEndpoint` run in the database.**\n A sort on a formula read the whole table into the page, because the server's source\n was never told the base's fields. It now works from the fields the server holds (the\n `svgrid:` comments, or the new `computed` option for MySQL and SQLite) and says which it\n works out in `serverComputes`. A request names a field and brings no definition. A link\n is followed only for a person who may read the other table, and not into a table\n `scope` covers. A field changed in the page goes back to being worked out in the page.\n- **SvGrid Base: a sort, group or negative filter on a lookup reads the other table once.**\n Sorts and groups use a `LEFT JOIN`, and \"does not contain\", \"is not\", \"is empty\", \"is\n none of\" and \"not equal\" a `NOT EXISTS`, in place of a subquery per record. A record with no\n link, or a link to a record marked deleted, matches as it did.\n- **SvGrid Base: a view counts once per query.** A block read asks for rows alone once the\n view has its count. The count is read again after a write through the page, a change\n from a feed, a block that does not fit it, or 30 seconds. With no feed, a teammate's new\n record reaches the count in that time, at the next read.\n- **SvGrid Base: blocks read their keys first.** A read with text copies or a join finds the\n block's keys, then reads those rows, so a sort or an `OFFSET` no longer carries whole rows.\n- **SvGrid Base: a date condition compares the column to a range** on Postgres and MySQL,\n so an index on it serves a calendar and a filter. `DATE(col)` and `col::date` could not\n use one. SQLite keeps `DATE()`, which reads the offsets in its text.\n- **SvGrid Base: a read nobody waits for stops.** `execute` gets `{ signal }` on reads, no\n further statement of a stopped read is sent, and `createKitHandlers` passes\n `request.signal` on.\n- **SvGrid Base: a chart on dates is one grouped statement**, and group charts, board lanes\n and calendar days no longer run a count they do not use. A dashboard reads its tiles\n once as it opens, and tiles on one query share its count.\n- **SvGrid Base: a view switch shows no old count or rows** under the new view's name.\n- **SvGrid Base: a bulk delete reads by key** in place of a growing `OFFSET`, with a count at\n each end of the read.\n- **SvGrid Base on MySQL: a descending sort reads an index backwards.** Its tie-break on the\n key follows the sort's direction there, so records with the same date now come newest\n key first. `date DESC, id ASC` sorted every row.\n- **`baseFromDatabase` and `createBaseEndpoint` open a MySQL database.** The column read used\n `generated`, a reserved word, as a bare alias, and MySQL refused the statement.\n- **`createBaseEndpoint`: the first description no longer groups a unique column.** A column\n that shows more than 12 different values in a sample of 5,000 rows cannot be a select.\n\n- **Server row model: a drag of the scrollbar reads where it stops.** A jump of more\n than a block waits for the viewport to stand still (`leapPauseMs`, default 60), a\n read the viewport has left stops through `request.signal`, and the rows on screen\n go first in the queue. Placeholder rows keep their row number and line up with the\n columns.\n- **SvGrid Base: Gallery, List, Inbox, Board lanes and the Timeline read ahead of a\n scroll,** further the faster it goes and the slower the source, and a drag of the\n scrollbar reads where it stops.\n- **SvGrid Base: a typed column or view filter waits for a pause,** as the search box\n does; a pick from a list applies at once.\n- **SvGrid Base: a busy calendar asks for its days' sizes in one grouped request.** It\n skips empty days, and after a write reads only the days whose size moved. A record\n with an end date, or a backend that does not group, stays a read per day.\n- **SvGrid Base: reading a table through in the page keeps 4 requests open** past the\n first blocks, instead of waiting for one at a time.\n- **SvGrid Base: a Calendar and a Timeline no longer stay in memory after a view\n switch** on a Svelte older than 5.57.2 by these two routes. Svelte 5.55 to 5.57.1\n keep other parts of a destroyed grid linked as well: use 5.57.2 or later.\n- **SvGrid Base: every record action from the keyboard and on touch.** The Menu key\n and Shift+F10 open the record menu with focus in it, the row number has an actions\n button on a touch screen, a long press opens the menu, and the record view can\n delete its record. The field menu and the table menus take focus, walk with the\n arrows and give focus back. A field moves from its menu or from Hide fields.\n- **SvGrid Base: the rows are one tab stop.** Tab goes on to the header and footer\n controls; it walks cells after a click or in an open editor, and says the way out.\n- **SvGrid Base: forms have real inputs.** Typed fields are labelled inputs with\n required and invalid states and autofill; a failed submit goes to the first error\n and reads it out.\n- **SvGrid Base: what changes is said aloud.** Edits, clears, pastes, undo and redo\n reach a screen reader, and a calendar event moves only once picked up with Space.\n- **SvGrid Base: contrast.** Text on a filled accent control now follows\n `--sg-on-accent`, and accent-colored text in the chrome is plain ink. Text on a colored\n pill or bar stays white on saturated colors, by design; the VPAT gives the ratios.\n- **SvGrid Base: forced colors, right to left, translation.** Switches, the active\n tab and view and the \"+\" keep their state under forced colors. Arrow keys and the\n pin line follow `dir=\"rtl\"`. `setBaseMessages` now reaches the grid's own menus,\n find bar and labels (`grid.*`) and the calendar (`viewbar.calendar.*`).\n- **SvGrid Base: smaller fixes.** Focus stays in the grid after a delete, panels hold\n Tab, Search moves into the find bar, targets are 24 px, an IME composition keeps\n its Enter, and a calendar on a phone keeps its title on one line.\n- **SvGrid Base: a select option has an identity.** Renaming one in the field editor renames it on every\n record, in a multi select too, keeps its cards in their lane, and reaches the view filters, condition colors,\n form logic, dashboard tiles, automations and the field's default that name it. Removing one says how many\n records hold it, next to Save, and clears them. Undo takes it all back. On a table in a database it is a write\n to each record holding it, a block at a time. A card whose value is no option goes to Uncategorized, not off\n the board.\n- **SvGrid Base: views show the data.** A board, calendar, timeline and gallery follow an undo and a record edit\n at once. Calendars and timelines leave undo to the base, so a drag is one Ctrl+Z and not two that cancel out.\n- **SvGrid Base: sort, filter and group apply again after an edit**, holding a grid row only while it is the\n active row, and the count follows. An undo reaches a record the view no longer shows.\n- **SvGrid Base: the bulk bar is part of a base.** Ticking records shows it, with Delete, in every template,\n and never the developer note. `selectionBar={false}` turns it off.\n- **SvGrid Base: a formula reads names.** `{Account}` and `{Owner}` give what the cell shows, not record and\n user ids.\n- **SvGrid Base: deleting a field says what it drives**, by view and automation. A view asks in its own row, goes to\n the trash, and is one step of undo. A field deleted from a view no longer leaves it saying \"Colored by field\".\n- **SvGrid Base: columns resize and move by drag** by default, and a restored field goes back to its place.\n- **SvGrid Base: Hide fields keeps its rows still** while fields are switched.\n- **SvGrid Base: the warning for a type change sits next to Save**, in sight.\n- **SvGrid Base: comments can be edited and deleted by their author.** `update` and `remove` on a comment store.\n- **SvGrid Base: activity stays true.** An undo and a redo are noted, and a local edit shows old to new.\n- **SvGrid Base: Timeline is in Create in every base**, its renderer loading on demand, and the roadmap template's\n Timeline is one.\n- **SvGrid Base: a form builder.** \"Customize form\" sets the title, the words and the fields, their order and which\n are required. An empty form does not send.\n- **SvGrid Base: Space on a group row opens nothing**, where it opened an empty record that lost what was typed.\n- **SvGrid Base: smaller things.** The link picker takes arrows and Enter and keeps its size, and a second click no\n longer closes it. A link cell selects on the first click. One click sets a rating in a form. Next and Previous on a\n card walk the board. Find and the field type search take the focus. A\n header opens its menu on a right click. New groups start open. The row height menu marks the current height. A\n multi-day event's tooltip says its days.\n- **The schema is the whitelist, for reads and writes (breaking).** `createSqlDataSource` names\n the schema's columns in every `SELECT` and `RETURNING`, and drops every key off the schema\n from a write. A group level off the schema is refused with `leafCounts` too. A trusted\n source in the same process as the page can ask for the old behaviour with\n `wholeRecords: true`; `createKitHandlers` refuses to start on one.\n- **Nothing a client sends reaches a private key.** `createKitHandlers` strips `__svgrid*` keys,\n keys off the schema and `readonly` fields before `authorize`, `validate`, hooks and the\n source see the row, so a write can no longer cross a `scope` or swap a validated value. A\n write left with nothing is a 422. A `scope` field off the schema refuses the request.\n- **Handlers no longer send driver text (breaking).** A failure is logged on the server and the\n client gets a plain message. A record that is gone is a 404 with `code: 'row-not-found'`,\n which `createKitDataSource` keeps on its error. A request off the schema is a 400 that says so.\n- **`createBaseEndpoint` starts closed (breaking).** It refuses to start without `authorize` or\n `public: true`, and without `tables` or `allTables: true`, with a message that says what to do.\n `GET` asks `authorize` and describes only the tables it allows. A table `authorize` gives no\n guard for is closed. It takes `scope`. Columns named like a credential stay out of a base\n (`hide`, `isSecretColumn`): never the primary field, never sampled as options, and through the\n endpoint not read or written. A field with `hidden: true` is no field of a base.\n- **A record's text cannot name a secret.** A webhook fills its URL, headers and body in one pass,\n so `{secret:DATABASE_URL}` typed into a field goes out as that text.\n- **Server webhooks stay off internal addresses (breaking).** `createAutomationServer` refuses\n private, local and link-local addresses, by name and by what the name resolves to, and does not\n follow redirects. `allowInternalWebhooks: true` opts out.\n- **A PDF preview runs nothing.** A file is read first and framed as a copy typed by us, only if\n its bytes are a PDF. An HTML file calling itself a PDF offers a download.\n- **Hidden fields and tables stay out of search, Download as SQL and the row number tooltip.**\n `lockViews={false}` on `SvBase` is honoured with `tables`. `permittedRecords` is what to send\n from a server; `permittedTables` keeps hidden values in its records, and the docs no longer\n point at it for that.\n- **A CSV is read for what it is.** The separator and encoding are found and shown to change. A\n quote inside a word is text, and a quote never closed no longer swallows the file. Rows that do\n not fit the header, values a field cannot hold, records that repeat a name, a file that is not\n UTF-8 and one that is not text are all reported before the import. A decimal comma stays a\n decimal, and `1,50` is no longer read as 150. `readImportMatrix` returns `delimiter`,\n `encoding` and `warnings`. Errors from a bad .xlsx or JSON file read as sentences.\n- **Large tables.** Import, summaries, dashboards, chart axes, rollups, pivots and spreadsheet\n ranges no longer throw `Maximum call stack size exceeded` past about 120,000 values (`maxOf`,\n `minOf` in place of `Math.max(...values)`).\n- **Download a view.** The list of tables has CSV and Excel downloads of the shown view: its\n fields and records as they read on screen, formulas, links, people, checklists and attachments\n included. `recordsAsShown` gives the same in code.\n- **Smaller holes.** A protected field is not asked in a form of someone not on its list. A paste\n outside a select's options is not written. Rich text in a base keeps no `class`, and an image on\n another site becomes a link instead of a request. A color from data that is more than a color\n is dropped. JSON import no longer fills `constructor`-like columns from the prototype, and a key\n first seen late in a file is a column.\n- **SvGrid Base: opening a cell and leaving it writes nothing.** A number, bigint,\n time, timestamp, date or JSON value used to be rewritten as the editor showed it.\n- **SvGrid Base: a number is stored as typed.** 19.99 in a field that shows one\n decimal stays 19.99; only an integer field rounds. A number past 15 significant\n digits stays exact as text, in the cell, the box and the write. A rollup with a\n `precision` keeps the linked field's symbol.\n- **SvGrid Base: dates and times no longer change with the browser's zone.** Dates,\n times, bigints and decimals are read as the database writes them, so a naive\n timestamp shows one wall clock everywhere, a typed time keeps it, and an undo or\n restore writes back what was read. On PostgREST a bigint past 2^53 is read again as\n text.\n- **SvGrid Base: dates read as people type them.** `12/24/2026`, `2026-12-23`,\n `Dec 22, 2026` and `12/21` all work in the new date editor (`fieldDate`), and text\n that is no date says so.\n- **SvGrid Base: cell keys.** Enter on a selected cell opens its editor, a key typed on\n a date or link cell starts editing, and Space ticks a checkbox cell.\n- **SvGrid Base: JSON in a text field shows as JSON** instead of `[object Object]`, and\n the Delete key leaves NULL as NULL and empties a number or date column without an error.\n- **SvGrid Base: a record added or deleted survives a view switch.** On a table the page\n holds, the grid went back to its `data` on any option change, so records added (by the\n row, Duplicate, a Kanban card, a form) vanished and deleted ones came back. What the\n grid holds is now what it is handed.\n- **SvGrid Base: Undo and Redo leave a teammate's cell alone.** A cell changed since keeps\n the later value and the page says so, cell by cell in a paste. Without a live feed the\n page reads the record first.\n- **SvGrid Base: a record's way back is whole, and safe.** Undo and restore on a table the\n page holds write every column read, under the record's own key, and the record stays in\n the trash until the database has it; a failed undo can be pressed again. A restored\n record whose Autonumber was given out again takes the next one.\n- **SvGrid Base: the trash gets the record as it is at the delete.** A delete reads the\n records at that moment, also after a confirm dialog, and a record moved out of a bulk\n choice while the question stood stays.\n- **SvGrid Base: typing is safe from teammates.** Changes from others, by feed or by a read,\n wait while a cell editor is open, so what is typed is not cut short. If the record went\n meanwhile, the person is told, with their text.\n- **SvGrid Base: a delete that did not happen is not reported as done.** One a row-level\n policy skipped, or the database refused, stays in the table and out of the trash, and the\n notice comes after the delete, not before.\n- **SvGrid Base: failures a person can act on.** A failed write names the record and the\n field in plain words, shows what was entered and offers Try again; \"Not saved\" stays until\n that record is written. A write with no answer warns and then lets the rest go\n (`writeTimeout`).\n- **SvGrid Base: Ctrl+Z of a new record on `NOT NULL DEFAULT ''`.** An emptied cell goes back\n to what the database gave it, not to NULL.\n- **SvGrid Base: the trash tells the truth.** With nothing keeping it, the panel says items go\n with the page.\n- **Supabase: the find box takes `%`, `_`, commas, quotes and brackets as text**, and a table\n with text keys can add records with default options.\n- **SvGrid Base, smaller things.** The selection bar counts on a table the page holds, a draft\n shows no number until it has one, and a write coming back in the database's words (`5.00`\n for 5) is nobody else's change.\n- **SvGrid Base: a group on a date opens on its records.** On Postgres and MySQL the\n key of a date field's group went through a JS Date, so it lost what is past the\n millisecond and took the page's zone, and outside UTC the group read nothing. The key\n is now text the database writes and compares. A delete can also leave out any group\n at any level: every source reports a group's records as `leafCount` when a base asks.\n- **SvGrid Base: two people adding at once do not share an Autonumber.** They could get the\n same number. Each record is checked once stored and the later one moves on; a UNIQUE\n column refuses the second and the write tries the next number.\n- **SvGrid Base: Undo of an added record keeps a teammate's work.** One someone has\n changed since stays, with a message, and Redo does not add it twice. Redo of a delete\n puts the records in the trash as they are now, not as they were.\n- **SvGrid Base: Undo brings times back exactly.** A delete reads time columns as the\n database writes them, so microseconds and a local time the clocks skipped survive on\n every Postgres and MySQL driver. `dateStrings` is no longer needed.\n- **SvGrid Base: Undo works on a server in any language.** Generated columns are read\n from the schema and left out before a write, and a refusal is told by its error code.\n `createSupabaseDataSource` keeps the code on the errors it throws.\n- **SvGrid Base: Undo leaves other people's work alone.** Undoing a step on a table held\n in the page put back a snapshot of all its records, which took out whatever teammates\n had changed since, and wrote that to the database. A step now undoes what it did and\n nothing else, and a new record closed empty takes only itself out.\n- **`createSqlDataSource` on MySQL.** Writes ended in `RETURNING`, which MySQL does\n not have, unless you set `returning: false`. That is now the default there. A\n record with an auto-increment key now gets the key the insert made.\n- **Formula text in SQL.** A string in a formula could break out of its literal on\n MySQL through a backslash, and a pivot key from stored data the same way. Both are\n closed, on every dialect.\n- **Formulas in SQL say what a cell says, or leave it to the page.** Functions whose\n results differ between a database and a cell (`UPPER`, `FLOOR`, `TODAY` and others,\n listed in the docs) no longer run in SQL.\n- **`createSqlDataSource`:** a filter on a column the plan does not know is an error\n and not a wider answer; `in` with an empty list matches nothing; `notEquals`,\n `notContains`, `endsWith`, `isNotBlank` and `notIn` run as themselves and no longer\n as `contains`; a blank group key matches NULL and ''; `updateRow` on a record that\n is gone is an error with `code: 'row-not-found'`; `createRows` keeps a column's\n default. A filter or sort on a field added at run time works.\n- **SvGrid Base:** a write the database refused no longer reads as saved after a later\n edit; records in the trash leave it only once they are proven back, whole, with the\n columns the base shows no field for and under the key they had; times go back as they\n came on MySQL and on Postgres; junction links come back with a restored record and\n survive a refused delete; a table whose records have no id of their own no longer\n breaks the grid or the trash; writes still queued\n when the page leaves finish, and the browser asks first; a form says so when the\n database refused the record; a field named ID no longer writes over the record's\n key; a formula edited on a big table runs at once, in the database too.\n- **Scrolling a very long list.** In a gallery, list, board lane or timeline past the\n height a browser scrolls, the scrollbar and the list stopped agreeing near both\n ends. A scrollbar step there now moves the list by the same step.\n- **SvGrid Base: a table of a million records scrolls without stalls.** With\n `serverRows`, the view bar counted records by reading every row of the\n server row model, each one made on read, on every block that arrived; the\n record view looked for its record the same way. The count now comes from\n the server (`rowCount`, and the top groups' counts when grouped), and the\n record view looks near the row it was opened from first. Number, currency\n and date cells also keep their `Intl` formatters instead of building one\n per cell. In demo 723 (a dev build, one laptop), 40 wheel steps went from 9.3 s of blocked main\n thread to about 0.1 s, and opening the filtered view from 6.5 s to 0.5 s.\n\n- **SvGrid Base: switching views on a big table no longer freezes the page.**\n A grouped view on `serverRows` set a row height per row and opened its\n groups, and both read every row of the row model; now it keeps one height\n and leaves server groups closed, since each opens with a request. A board\n of more than 200 records draws only the cards in sight in each lane (the\n board measures a card when `cardHeight` is not set). The board also keeps\n each card's key instead of working it out on every read. In demo 723 (a dev\n build, one laptop), the slowest switch went from 3.2 s to 0.4 s.\n\n- **SvGrid Base: a view that left out part of its state kept the last view's.**\n A view declared with only `grouping` took over the sort, hidden fields and\n other state of the view before it. Whatever a view's `state` leaves out\n now starts plain. A saved view is unchanged: it stores all of it.\n\n- **SvGrid Gantt: the task table's header stays at the top.** It scrolled away\n with the rows in any plan taller than the chart.\n\n- **SvGrid Base: a record past what a table read was taken for a deleted one.**\n When a table on a `source` was read again at its limit, a record that had\n moved past the limit was dropped from the page and reported as deleted.\n A read at the limit now keeps the records it did not see.\n\n- **A pivot filtered its rows again for every column.** For each row group,\n `createPivotModel` went through the group's rows once per column-axis leaf.\n The rows are now bucketed by column path in one pass, in the same order,\n and the result is the same. On one laptop, a pivot of 100,000 rows over\n Year x Month went from 3.6 s to 0.6 s, and 1,000,000 rows from 147 s to\n 4 s.\n- **A grouped Excel export slowed down with the square of its rows.** The\n writer looked up each string cell's shared-string index, and checked the\n group rows already written, by scanning an array. Both are hash lookups\n now and the file is byte for byte the same. A 10,000-row export grouped by\n five columns went from 4.8 s to 0.45 s, and 100,000 rows finish in 9 s.\n- **A sheet chart hung below the window was not drawn until a scroll.** The\n sheet measured its objects once after mount, from their anchor cells, and\n the grid now adds most of its overscan rows a frame after its first paint;\n a chart hung from a row in between had no cell to measure. While an object\n has no box, rows the grid adds now measure it again.\n\n- **A wide grid lost columns in the PDF.** pdfmake sizes a column no narrower\n than its longest word, so a 22-column report ran off the right edge and only\n 15 columns printed, without a warning. The columns are now fitted to the\n page (see `pdf.fit` above).\n- **A row taller than a page was missing from the PDF.** Rows are kept whole\n on a page, and pdfmake drops a row that cannot fit on any page. Such a row\n now breaks across pages.\n- **Conditional-format icons printed as empty boxes in the PDF.** The arrow,\n triangle and traffic-light glyphs are not in Roboto; they are drawn as\n shapes now.\n- **The PDF footer could show yesterday's date.** It printed the UTC date, so\n an export after midnight east of Greenwich was dated the day before. It\n prints the local date, and the footer follows custom page margins.\n- **Grouped PDFs dropped conditional formatting and links.** They now carry\n both, as flat exports do.\n- **Booleans printed as \"true\" and \"false\" in the PDF.** They print a tick or\n nothing, like the grid's checkbox. An icon set's `iconOnly` is honoured too:\n it printed the value next to the icon.\n- **A number or date column passed in `columns` without `align` printed\n left-aligned in the PDF.** It right-aligns from its `format`, as on screen.\n- **A pivot chart drew an empty cell as a zero bar.** An average over no rows\n is `null` in the pivot and the bridge read it as 0; it is a gap on the\n chart now. A sum over no rows stays the 0 the engine wrote.\n\n- **PDF export depended on pdfmake's fonts registering themselves.**\n `pdfmake/build/vfs_fonts` registers its font map only as a side effect of\n being imported while a global `pdfMake` already exists, and the export's\n own lookup for the map never matched the shape the module has shipped since\n pdfmake 0.2.8 (the module is the map), so it registered nothing and relied\n on that side effect. Under a bundler the global can be a stale instance (a\n dev server re-optimising its dependencies mid-session is enough), and the\n export then failed with \"File 'Roboto-Medium.ttf' not found in virtual file\n system\" while its promise never settled. The map is now found in every\n shape the module has had (`resolvePdfVfs`) and handed to the instance that\n creates the document (`registerPdfFonts`, through `addVirtualFileSystem`\n and the `vfs` property), and an installed pdfmake whose fonts module has\n no map is reported instead of failing later.\n- **Studio emitted no theme for the fragment export and the CLI `add`\n scaffolds.** Only the full app's `+layout.svelte` carried the `--sg-*` tokens;\n `eject --fragment` (which drops that file) shipped an `app.css` with one\n hard-coded accent, and `svgrid-studio add` / `add --all` wrote none at all. The\n tokens now come from one helper (`themeTokenCss`): `src/app.css` carries them\n for the full app and the fragment, and the `add` scaffolds emit them in\n `<svelte:head>`. Drop-in outputs wrap them in `@layer svgrid-studio`, so a host\n app that already defines its own tokens keeps its look. `add` also honours\n `--theme` / `--dark`, and a project with no theme picked gets Ember (the demo\n theme) instead of Tailwind.\n- **Scheduler column alignment.** The all-day lane, day/time-grid header, and\n body columns now reserve the scrollbar width consistently, so the columns line\n up instead of drifting by the scrollbar's width.\n- **Excel export color-scale conditional formatting.** A color scale's `min` /\n `mid` / `max` are optional, so the export could build an `(string | undefined)[]`\n where a `string[]` was expected; it now falls back to the same `#ffffff` /\n `#000000` defaults the grid renders with. Because `@svgrid/enterprise` ships its\n `src`, this had surfaced as a stray `svelte-check` error in consuming apps - a\n clean app now type-checks with zero errors.\n\n#### Security\n\n- **SvGrid Base: a pair table with the tenant in its key could be read and changed by another tenant.**\n On `order_tags (tenant, order_id, tag_id)` with `tenantColumn` and a scope, the pair id a client sent\n overwrote the scope's tenant, so an editor of tenant B read tenant A's pairs by id and deleted them\n with `delete` and `deleteMany`, on Postgres, MySQL and SQL Server. The scope now wins: an id naming\n another tenant's pair reads nothing, and a delete of it is a `403`, as on a `{ through }` pair.\n- **SvGrid Base: a `{ through }` pair read could hand out another tenant's key.** A pair written\n straight into the database between tenant A's order and tenant B's tag reached A in raw rows. A read\n of a junction (a pair table, or one with its own key) now leaves out a row whose other end is a record\n of another tenant, where that end's table holds its own tenant column. Delete forever still clears\n such a row.\n\n- **A crafted .xlsx could crash the tab.** `importData` and the spreadsheet's\n File > Open filled every gap before a cell with blanks, sized by the cell\n reference in the file: `<c r=\"ZZZZZZZ1\">` asked for about eight billion.\n `importData` now rejects a cell past column XFD; the spreadsheet skips a cell\n outside the sheet, as it does other malformed parts. Both refuse a sheet that\n would need more than ten million blank cells to lay out.\n- **A zip bomb could exhaust the tab's memory.** Every .xlsx and .ods reader\n (`importData`, File > Open, `documentFromXlsx`, `documentFromOds`) unzipped\n each part whole before looking at it, so a small file whose parts inflate to\n gigabytes was read into memory first. Parts are now unzipped as a stream and\n counted as they inflate; a file whose parts pass 256 MB together is refused\n at that point.\n- **CSV and TSV from `exportData` neutralise formula-like text**, through the\n same `@svgrid/grid` serializer (see the grid's Security entry).\n\n### @svgrid/grid-wc\n\n#### Added\n\n- **`<sv-chart>`**, the standalone chart as a custom element, at\n `@svgrid/grid-wc/chart` with React (`/react/chart`), Vue (`/vue/chart`)\n and Angular (`SvChartComponent`) wrappers. Its surface is generated from\n `SvChart`'s own props type the way `<sv-grid>`'s is from the grid's, so the\n 26 props and 7 events stay in step; `spec` is a property, the primitives\n are attributes, and a prop that takes a boolean or a string (`legend`) keeps\n a string attribute where a bare attribute means true. Its own build in\n `dist/chart` carries none of the grid.\n- **Typed event handlers.** Each wrapper's `on<Event>` prop used to be\n `(detail: unknown) => void`, so reading `newValue` meant casting first -\n a poor advertisement for a typed wrapper. The detail type is now lifted from\n the grid's own callback signature, so 16 of the 19 events arrive fully typed\n (`TData` degrades to `Record<string, unknown>`; anything referring to a\n type that only exists inside `@svgrid/grid` stays `unknown` rather than\n naming something a consumer cannot resolve).\n\n- **React, Vue and Angular components**, as subpath imports:\n `@svgrid/grid-wc/react`, `/vue` and `/angular`. Each is generated from the\n same surface as the elements, so all 98 properties and 20 events are typed\n props on every one of them and none can drift from the grid. Each framework is\n an OPTIONAL peer dependency, so a plain-HTML consumer installs none of them,\n and the wrappers are 1.3-3.7 KB gzip because they reuse the one element bundle\n rather than shipping a second copy of the grid.\n\n They exist because a raw custom element is genuinely awkward in each:\n **React 18 and earlier stringify object props onto attributes**, so\n `columns={cols}` silently becomes `\"[object Object]\"` and the grid renders\n empty; **Vue** needs `isCustomElement` build config and a `.prop` modifier\n on every object binding; **Angular** needs `CUSTOM_ELEMENTS_SCHEMA` in every\n component that shows a grid, and gets no typed inputs. The Angular wrapper is\n compiled with ng-packagr in partial-Ivy mode, so it is consumable by a normal\n Angular build, and its selector is the element's own tag - `<sv-grid>` and\n `<sv-grid-shadow>` - with no template of its own, so there is no extra wrapper\n element in the DOM and the tag is the same one you write in plain HTML.\n\n All three also handle two ordering problems a hand-written wrapper meets:\n the element renders BEFORE a framework assigns properties in an effect, and\n `apiready` fires once during that first mount - before React can bind a\n listener at all. The handle is parked on the element and each wrapper replays\n it.\n- **The package ships TypeScript declarations.** It previously had none, so\n `import '@svgrid/grid-wc'` was an error under `moduleResolution: bundler`\n and `document.querySelector('sv-grid')` came back as a bare `Element`. The\n generated `.d.ts` augments `HTMLElementTagNameMap` and\n `HTMLElementEventMap`, so the element and its events are typed without\n hand-written declarations.\n- **`<sv-grid-shadow>` - the grid in an open shadow root**, so a host page's\n CSS cannot reach it. Same properties, attributes and events as `<sv-grid>`;\n a separate element rather than an attribute because Svelte resolves\n `customElement.shadow` at compile time, so no runtime flag can switch it.\n Isolation is one-directional and the docs say so: page CSS stops at the\n boundary, but the grid's own stylesheet is still injected into the document,\n because about twenty overlay surfaces portal to `document.body` on purpose\n to escape ancestor clipping. `--sg-*` theme tokens reach in unchanged,\n including a theme file's dark variant - custom properties are inherited, and\n inheritance crosses a shadow boundary.\n- **The elements now expose the whole grid.** `<sv-grid>` declared **7 props\n and 2 events** by hand against a `Props` type with 100 props and 19\n callbacks, so grouping, pagination, pinning, tree data, master/detail, board,\n scheduler and every Enterprise feature were unreachable from a non-Svelte\n host - while the docs said they \"all come along\". The surface is now\n **generated** from `<SvGrid>`'s own types: **98 properties** (72 with\n kebab-case attributes, 26 property-only) and **20 events**, with CI failing if\n the two drift. Costs 1.6 KiB gzip.\n- **Every grid callback is a DOM `CustomEvent`**, `detail` being the\n callback's argument - or, for the one callback taking two, an object keyed by\n its parameter names. `onApiReady` also parks the imperative handle on the\n element as `el.api`, because an event fires once and a host binding a\n listener later would otherwise never reach it.\n\n#### Fixed\n\n- **Importing the package on a server threw `customElements is not defined`.**\n Svelte registers each element with a bare top-level\n `customElements.define(...)`, so any server-rendered React app\n (TanStack Start, which Lovable now generates, Next.js, Remix) crashed on\n `import { SvGrid } from '@svgrid/grid-wc/react'`. The build now guards the\n registration; on a server the import is inert and SSR emits a plain\n `<sv-grid>` that upgrades on hydration. Guarded by `test/ssr.test.ts`.\n- **The React wrapper lost props the grid reads once at mount.** It wrote\n props in a passive `useEffect`, which runs after paint; the element mounts a\n microtask after insertion, so it had already mounted with defaults and\n `pageSize={25}` showed 10 rows a page (the wrapper's own pagination example\n did). Props are now written in a layout effect, before the element mounts.\n The same change in the `<sv-chart>` wrapper.\n- **`style` on the React wrappers took `Record<string, string | number>`,** so\n passing a React `CSSProperties` object failed to type-check. It is\n `CSSProperties` now.\n- **The React wrapper re-did all its work on every parent render.** Both of its\n effects had no dependency array, so any state change anywhere in the parent\n reassigned all 98 properties and rebound all 20 event listeners - 40 listener\n mutations per keystroke. It now writes only values that actually changed and\n binds listeners once through a handler ref. Measured: 20 no-op re-renders\n produce 0 listener adds, 0 removes and 0 property writes.\n\n- **The element threw when it rendered before its props were assigned.**\n Generating the prop list replaced the elements' own `data = []` /\n `columns = []` defaults with a bare `$props()`, so a consumer that assigns\n properties AFTER the element upgrades - which is what React and Angular both\n do, in an effect - got\n `TypeError: Cannot read properties of undefined (reading 'map')` and a grid\n that never rendered. Only a real framework app was late enough to reproduce\n it; an HTML fixture sets the properties in the same tick and never sees it.\n\n#### Changed\n\n- **Only arrays, objects and functions changed shape: they are property-only.**\n An HTML attribute is a string, so `columns='[object Object]'` cannot work\n and no attribute is offered for the 26 props that are not primitives. The\n published `rowclick` and `selectionchange` events keep the exact\n `detail` they shipped with, and `selectable` keeps meaning row-selection\n checkboxes - `<SvGrid>`'s prop of that name is an alias of\n `enableCellSelection`, and forwarding it would have silently repointed a\n published attribute at a different feature.\n- **Styles arrive with the code that needs them.** The build used to put every\n stylesheet in the bundle, lazy chunks included, into the entry, so a page\n that never opened the chart, a date picker or a menu still downloaded their\n CSS. Each chunk now carries its own styles and adds them to the document\n and to every open shadow root when it loads. `<sv-grid>`'s entry went from\n 107 KiB to 98.5 KiB gzip while the chart grew.\n\n### Tooling & docs\n\n#### Added\n\n- **Every chart gallery demo switches between Chart and Grid.** A segmented\n control first in each demo's toolbar flips every pane to a sortable grid\n of the rows the chart was drawn from, and back, the way the board demos\n switch Board and Table. The from-the-grid page shows the pattern in a\n runnable.\n\n- **Charting a pivot is documented.** The pivot page has a section on\n `pivotToChartSpec` (the mapping, every option, a runnable that pivots\n orders and charts them by quarter) and on the designer's Chart view\n (`chartable`, `defaultView`); the charts hub and the from-the-grid page\n link to it, the /api pivot section lists the function, and the generated\n reference covers `pivot-chart.ts`. Demo 359's data now puts every\n country in every quarter and category, so its groups are four bars, not\n one.\n\n- **27 runnable framework examples**, nine each for React, Vue and Angular:\n a first grid, sorting and filtering, editing and saving, row selection,\n grouping and totals, pagination, server-side data, theming, and Excel export\n from the Enterprise pack. Every one has\n an **Open in StackBlitz** button that boots a full editable project - no local\n install - and a new\n [React, Vue or Angular](/docs/help/web-components/frameworks) landing page\n puts the thirty-second path first.\n\n They are real apps in the repository (`packages/grid-wc/examples/`), not\n snippets, and CI compiles all of them with each framework's own compiler -\n including `ngc --strictTemplates`, so a wrong Angular binding fails the\n build. The doc listings are generated from those files, so a page cannot\n describe a version of an example that no longer exists.\n\n The Vue templates get an extra check of their own: Vue treats an unknown\n attribute as a legal fallthrough, so `<SvGrid sortabel />` type-checks\n cleanly and silently does nothing. Verified - the same typo fails the React\n and Angular compilers and passes `vue-tsc` - so the Vue examples are also\n checked against the generated surface.\n\n- **A Web Components docs category.** Eight pages under\n `help/web-components/` - quick start, a generated `<sv-grid>` reference,\n shadow DOM, React, Vue, Angular, TypeScript and limitations - replacing a\n single page that was filed under \"Layout & Styling\". The reference tables are\n generated from the grid's types, and a guard checks that the counts quoted in\n prose still match the surface.\n- **The docs run.** A `svelte` code fence tagged `{runnable}` is extracted at build time\n into a real component, so a doc page shows the source and the working result\n together rather than highlighted text. 504 examples across the corpus, taking\n the median doc page from 1 example to 3; each one is compiled by `svelte-check`\n in CI, so an example that stops working fails the build instead of rotting in\n place.\n- **`@svgrid/ui` recipe scaffolder** - `npx @svgrid/ui add <component>` drops a\n ready-to-edit starter (e.g. `add calendar`) into your app, complementing the\n shadcn-style component pages (Preview / Code, install tabs) across the UI kit.\n- **See a component before you wire it in** (`@svgrid/ui` 0.3.x).\n `npx @svgrid/ui try <component>` spins up a zero-setup Vite + Svelte sandbox and\n opens the component in your browser; `add --preview` writes a\n `src/routes/preview/<id>` route (plus a `/preview` index) in a SvelteKit app.\n `add` now installs `@svgrid/grid` by default (`--no-install` to opt out) and\n prints the exact `try` command for whatever you just added, so the \"see it\"\n step is never a dead end.\n- **AI coding Skill** (`skills/svgrid`, installable with\n `npx skills add sv-grid/sv-grid`) - an always-on house-style guidance layer\n that complements `@svgrid/mcp`.\n- **The rest of the toolkit is findable.** `packages/grid/README.md` gained an\n \"Also in the box\" section, and the site footer gained a **Toolkit** column\n (SvGrid UI, Kanban board, Scheduler, Dock layout, Studio, Web components).\n The 84 SvGrid UI components have always been exported from `@svgrid/grid`\n itself - `SvButton`, `SvCalendar`, `SvDateTimePicker`, `SvDockLayout`,\n `SvToaster` and the rest are already installed for every grid user - but\n nothing on the highest-traffic surfaces said so.\n\n#### Changed\n\n- **Doc examples show code and result at once.** The runnable card used to be a\n Preview / Code toggle with an in-page editor; it is now the source above the\n live example, both visible, with Copy. A tab hides half of a teaching example\n and makes comparing the two a click, and dropping the editor took the\n on-demand TypeScript compiler chunk off every docs page with it.\n- **`npm create @svgrid` starts on Ember** (the demo theme) instead of Tailwind,\n for every template; `--theme <id>` still picks any preset. The `sveltekit`\n template is now listed on the starters page, and the `sv add @svgrid` demo route\n imports the Ember stylesheet so it no longer renders unthemed.\n\n#### Fixed\n\n- **The base scaffold lists `jszip`**, so \"Download this view as Excel\" works in a new app. Its README no\n longer says a new table appears on reload with the Postgres setup: that route reads the tables once\n per server process, so restart it or call `refresh()`.\n\n- **14 doc examples drew an empty or broken box under their code.** Every\n `{runnable}` block was run against the published packages in a clean Vite\n app. Five needed a SvelteKit load, an `/api` server or a WebSocket (two\n showed \"This example failed to load.\" on svgrid.com) and one is a component\n file that needs props; they are now `{norun}`, a new fence flag: still\n type-checked, never mounted, never promoted back to runnable. The other\n eight now run: the custom header, the real-time cell flash, the\n inline-snippet cell renderer, and the alert, loading-overlay, scroll-area,\n sparkline and toaster examples. The toaster page also said hidden toasts\n appear as room frees; their timers keep running while hidden, which the\n page now says.\n- **`build:lib` in the enterprise package deleted the node bundle.**\n `svelte-package` wipes `dist/`, so a library build alone removed\n `dist/node/studio.js` and the MCP server failed to start until\n `build:node` was run by hand. The library build rebuilds the bundle it\n removed.\n\n- **Docs images and tutorial media 404'd under a site base.** The markdown\n writes them root-absolute (`/docs-media/...`, `/tutorials/...`), which is\n right at svgrid.com and wrong under the `/sv-grid/` base the e2e server\n uses: every docs image and tutorial poster was missing there. The page\n prefixes the base in the HTML before it paints, and an e2e spec loads a\n page with both and checks nothing under those roots 404s.\n\n- **Twenty-one links on the chart pages went to GitHub 404s.** The split\n wrote every cross-link with one `../` too many, and the site's link\n resolver falls back to a GitHub URL for a page it does not know. Fixed on\n the pages, and every relative link under docs/ is now a test\n (`tools/docs-relative-links.test.ts`); the link checker existed but only\n ran by hand before a release.\n- **Links into the generated reference tree went to GitHub too.** The tree\n is hidden from the docs routes; a link to it lands on the /api page, on the\n module's section where it has one.\n- **The chart gallery said \"the source is under each example\" and it was\n not.** Each gallery card now carries the demo's own script and markup on a\n Code tab, written by `tools/gallery-doc-code.mjs` from the demo file and\n guarded by a drift test, so the code beside a preview is the code that\n renders it.\n- **A demo page flattened a bulleted banner into one paragraph**, dashes and\n backticks included. The page reads the pitch as blocks: a list where the\n banner wrote one, `<code>` where it wrote backticks, on the prerendered\n and the hydrated page alike.\n- **`demo-doc-embed` skipped a demo the page only linked to**, so a\n placement for it never landed, and it opened a new \"More examples\" section\n below \"See also\", which the coverage test requires to be last. Present\n means embedded, and the section opens above \"See also\".\n- **Regenerating the doc snippets took the dev server's file watcher with\n it.** The extractor removed and recreated its output directory, and on\n Windows the watch did not re-attach, so an edited runnable kept serving\n its old chunk while the source beside it showed the new code. It writes in\n place and removes only what no longer belongs.\n- **The Getting started chart page had no live example.** Its code blocks\n were `ts` and a Svelte block referencing rows it never declared; four of\n them run now, and the hub's first example runs with its own rows and a\n filter row, so the chart is seen following the grid.\n\n- **The e2e suite adopted any server on :5180.** Playwright reuses whatever\n answers on the port, and a Vite dev server returns the app shell for any\n path, so a website dev server started by hand (base `/`) passed the\n readiness check and the fourteen path-routed and module-importing specs\n failed with nothing pointing at the server. The config now probes the port\n first (`tests/e2e/lib/site-server-probe.mjs` reads the base off Vite's\n client script tag), reuses a server that serves the site under `/sv-grid/`,\n skips one that serves anything else with a line saying so, and starts its\n own on the next free port; `SVGRID_E2E_PORT` pins one.\n- **The website dev server re-optimised dependencies mid-session.** The\n search index, the docs renderer, the /api runner's compilers, the PDF\n exporter and the grid's `esm-env` sit behind dynamic imports the start-up\n scan does not reach, so the first page to touch each one triggered\n \"optimized dependencies changed. reloading\" on every open page. In the e2e\n suite that reload landed mid-test on a cold cache and failed whichever spec\n was on a page at the time. They are pre-bundled at start-up now\n (`optimizeDeps.include`), and `esm-env` is a declared devDependency of the\n website so it resolves from there.\n- **The /api runner check sampled an example's outcome two frames after\n Run.** An example awaiting an export or a fetch was still running, so a\n failure that arrived later, or one thrown inside a library callback the\n runner's try/catch cannot see, passed. The check now waits for the runner\n to report an outcome and counts an uncaught error or rejection during the\n run as a failure of that example.\n- **181 doc pages had content stranded below \"See also\".** Authoring passes\n appended new sections to the end of the file, which buried them under what\n reads as the page footer: on `help/filtering/number-filter`, for instance, the\n `between` example sat below the link list and a reader would never reach it.\n The footer is back at the bottom on every page, and a test now fails if a\n section lands under it.\n- **The documented way to lock one column's width never worked.**\n `docs/help/columns/column-sizing.md` matched headers on `data-col-id`, which is\n the attribute on *body* cells - the selector matched nothing. Header cells\n carry `data-svgrid-header-col`. (`ColumnDef.resizable` is now the real answer.)\n- **`blockKey` was missing from `tools/lib/md-snippets.d.mts`.** The function was\n exported from the `.mjs` and worked at runtime, so nothing caught it until the\n website was type-checked.\n- **`@svgrid/mcp` reports its real version.** The MCP server's\n `serverInfo.version` was hardcoded to `0.1.0`; it now reads the package version.\n- **`@svgrid/create-studio` scaffold builds out of the box.** The generated app\n shipped an orphaned `/products` route referencing an entity that was never\n defined; the `Product` schema, seed, data source, nav, and home card are now\n complete, so `npm run build` succeeds on a fresh scaffold.\n- **API-docs accuracy pass across all 74 UI-component pages.** Corrected the\n `SvDockLayout`, `SvDockManager`, and `SvDateTimePicker` prop tables (added 13\n missing dock-layout props; removed a `size` prop `SvDateTimePicker` never had),\n and filled scattered gaps (`loading`, `onClosed`, `min` / `max`, `dir`,\n `width` / `height`, `ariaLabel`, and the shared field-contract on\n `SvFileUpload`).\n\n### @svgrid/mcp\n\n#### Changed\n\n- **A cold `npx -y @svgrid/mcp` installs 20 packages instead of 111.** The\n server speaks MCP over stdio itself instead of through\n `@modelcontextprotocol/sdk`, whose HTTP and OAuth dependencies (Express, Hono,\n JOSE, AJV) a stdio server never runs. On an empty npm cache the install took\n 6 to 8 seconds instead of 12, which had left little room under an MCP client's\n 30 second connect timeout. The protocol is unchanged: version negotiation,\n errors and every tool, resource and prompt answer as before.\n\n## [1.0.0] - 2026-06-16\n\nInitial public release - the Svelte 5-native data grid: `@svgrid/grid`\n(MIT core) and `@svgrid/enterprise` (the paid feature pack).\n\n### @svgrid/grid\n\n- **New** `onCellValueChange` callback on `<SvGrid>`. Fires after every\n inline edit commits with `{ rowIndex, columnId, oldValue, newValue,\n row }`. The recommended hook for server-side persistence and cascading\n recomputes; see demo `18-cascade-editing` and\n [Saving values](./help/editing/saving-values.md).\n- **New** `externalSort` + `externalFilter` props. Grid records the UI\n state but does NOT re-order / filter rows - the consumer owns the\n pipeline. Paired with `onSortingChange` / `onFiltersChange` for\n server-side data and tree-data scenarios.\n- **New** `onSortingChange(sorting)` / `onFiltersChange(filters)`\n callbacks. Fire on every change with the consolidated payload shapes\n documented in the [SvGrid reference](./reference/SvGrid.md).\n- **New** `onRowSelectionChange(selection, rows)` and\n `onActiveCellChange(cell)` callbacks.\n- **New** `fitColumns` prop. Scales column widths to fill the viewport,\n with rounding-residue absorbed in the last scalable column. Shrinks\n down to 85 % of natural widths; beyond that a horizontal scrollbar\n appears.\n- **New** `showRowNumbers` prop. Leading 1-based row-number column,\n rendered before any selection column.\n- **New** `pageSize` prop on `<SvGrid>`. Initial page size for the\n built-in pager.\n- **Improved** column virtualizer now detects per-item size changes,\n not just total size. Fixes a regression where resizing a column under\n `fitColumns` left other columns stale.\n- **Improved** the wrapper-managed filter pipeline no longer applies\n filters to the pre-paginated view - the filter UI always sees the\n full dataset, not just the visible page. (Removed\n `paginatedRowModel` from the engine pipeline; pagination is applied\n AFTER filters by the wrapper.)\n\n### @svgrid/enterprise\n\n- **New** package. The paid feature pack adds:\n - **Export** to xlsx / pdf / csv / tsv / html, with theme-matched\n styles, multi-sheet workbooks, header + footer with logo, embedded\n cell images. See [Export](./help/export.md).\n - **Print** with repeat-on-page headers, optional cover page,\n page-size + orientation.\n - **Import** from xlsx / csv / tsv / json with column mapping +\n per-row validation. See [Import](./help/import.md).\n - **AI assistant**: provider-agnostic helpers for natural-language\n filter, smart fill, summarize, and classify. Bring your own model\n adapter. See [AI](./help/ai.md).\n - **Pivot tables** via `createPivotModel(data, config)` and\n `pro.pivot.build(config)`. Row + column axes, 8 built-in\n aggregators or custom, grand-total row + column, subtotals,\n custom axis sort. See [Pivot tables](./help/pivot.md).\n- **New** `installEnterprise(api)` augments a `SvGridApi` with `exportData`,\n `print`, `importData`, `ai.*`, and `pivot.*` methods.\n- **Soft-gated licensing**: features run unlicensed but the grid shows\n a small watermark + a one-time console nudge. `setLicenseKey('SVENTERPRISE-…')`\n at app startup clears both.\n\n### Examples gallery\n\n- **New demos** (56-59): theme-matched export, branded export with\n header + footer + logo, export with cell images, multi-sheet export.\n- **New demos** (52-pivot-table, 51-ai-assistant, 53-excel-import):\n reference implementations for each Enterprise feature.\n- **New demos** for industry verticals: stock market (live ticks),\n HR team, finances ledger, industrial IoT, localization, CSP\n compliant, accessibility, cascade editing, server-side rendering,\n industrial dashboard, healthcare EMR, logistics, compliance queue,\n field service, gantt, scheduler, CRM, admin dashboard, seller panel.\n- **New** \"Source\" button on every demo opens the raw `.svelte` file\n in a modal with a Copy-to-clipboard button. Wired via\n `import.meta.glob('../demos/*.svelte', { query: '?raw' })` so it\n picks up new demos automatically.\n\n### Docs\n\n- **New** [Enterprise feature pack](./enterprise/README.md) landing page.\n- **New** [API reference](./reference/index.md) with hand-curated\n pages for `<SvGrid>`, `SvGridApi`, `ColumnDef`, features, and the\n full Enterprise surface.\n- **New** [Why headless?](./why-headless.md) explains the layered\n architecture and when to drop down to the headless core.\n- **New** [Tailwind integration](./help/tailwind.md) walks the\n `--sg-*` token surface, dark-mode wiring via `data-theme`, and the\n override hooks for the stable `.sv-grid-*` class names.\n- **Split** [Getting started](./getting-started.md) into six short\n pages (install / first-grid / data-and-columns / features /\n theme-and-density / going-to-production); the old single-page\n version is preserved at\n [getting-started-full.md](./getting-started-full.md).\n- **Cleaned** every code sample against the live library surface.\n Removed phantom APIs: `state={...}`, `rowModels={...}`,\n `initialState={...}`, `manualFiltering`, `manualSorting`,\n `manualPagination`, `onColumnFiltersChange`, the wrapper `getRowId`\n prop, per-column `enableSorting` / `enableColumnFilter` /\n `enableGrouping` flags, the never-shipped\n `@svgrid/grid/themes/default.css` import.\n- **Em-dashes globally swept to hyphens** (`-` → `-`) across all\n source-controlled text. Codified as a rule for new content.\n\n## How we version\n\n| Tier | What it means |\n| ----------------- | -------------------------------------------------------------------------------------------------------------- |\n| **Major** (1.x) | Breaking change to a Stable API. Includes prop / method renames, type-narrowing, default-behaviour changes. |\n| **Minor** | New Stable API. New Experimental API. New demo. Doc rewrites that touch the public-facing claims. |\n| **Patch** | Bug fixes. Type-only fixes. Internal refactors with no public surface change. Doc typo fixes. |\n\nThe [API stability page](./help/api-stability.md) annotates each export\nwith its current tier (`Stable`, `Experimental`, `Internal`). As of 1.0,\n`Stable` APIs follow semver; `Experimental` APIs may still change in a\nminor release.\n"
4608
4608
  },
4609
4609
  {
4610
4610
  "slug": "compliance/audit-log",
@@ -5178,28 +5178,28 @@ export const docs = [
5178
5178
  "path": "docs/help/base/automations.md",
5179
5179
  "title": "Automations",
5180
5180
  "section": "SvGrid Base",
5181
- "markdown": "# Automations\n\nAn automation does something on its own when something happens to a\nrecord: a task set to Done ticks its Done box, a new request goes to Todo, a\nblocked task tells its owner, a button starts the work. Each table of a base\nhas its own, under \"Automations\" in the view bar.\n\n<div data-docs-demo=\"504-project-tracker\" data-height=\"680\"></div>\n\nIn the demo, set a task's Status to Done, press Start in the Kick off\ncolumn, or send the Request a task form, then open \"5 automations\" and look\nat Runs.\n\n![An automation has three parts: when, only for records where, and do. What an action changes can set off other automations, up to four deep.](/docs-media/base-automation.svg)\n\n## Add one\n\nOpen \"Automations\" in the view bar and press \"+ New automation\". The dialog\nhas three parts:\n\n- **When**: what sets it off.\n- **Only for records where**: a condition written with the view bar's\n filter, so it compares a select's option, a person, a linked record's name\n or \"within the past week\" the way the filter does. Leave it empty for\n every record.\n- **Do**: the actions, in order.\n\n![The Automations dialog: the list of automations with an on/off switch each, and the one picked: when Status changes, only for records where Status is Done, do update the record and show a notice.](/docs-media/base-automations.png)\n\nEvery change is kept as you make it. \"Run now\" tries the automation on the\nrecord you pick, as if its trigger had happened.\n\n## When\n\n| Trigger | Runs |\n|---|---|\n| A record is created | When a record is added: \"+ Add row\", a form, a board lane, a calendar day, another automation. A record brought back by Restore or an undo of a delete is not created, and no rule runs for it |\n| A record changes | When a record is edited; pick fields to run only when one of them changes. A link made from the other table counts: adding a guest to an event from Guests changes the event's Guests field |\n| A record matches a condition | When a record starts to match the condition. It does not run again while the record keeps matching |\n| A form is sent | When a form view sends a record; pick the form, or any form of the table |\n| A button is pressed | When a Button field is pressed on a record |\n| At a scheduled time | Every day at 9, every Monday, every hour, or a cron expression: once for each record that matches the condition then, or once with no record |\n\nA paste or a fill that changes several cells of one record is one change to\nit. An undo or a redo is no change: it runs no automation.\n\n## Do\n\n| Action | What it does |\n|---|---|\n| Update the record | Sets fields of the record that set it off |\n| Create a record | Adds a record to any table of the base; \"Link it back\" fills a link field of the new record with the record that set it off |\n| Show a notice | A toast; `{Field name}` is that field of the record |\n| Color the record while it matches | Colors the whole row, or some fields, for as long as the record matches the condition |\n| Call a webhook | POSTs the record (or a body you write) to a URL: a team chat, your API. Secrets as `{secret:NAME}`, read on the server only |\n| Send an email | To, subject and text with field names; sent by the [server](#on-the-server) |\n| Your own actions | Anything the app adds with `registerAutomationAction` |\n\nA webhook's URL, headers and body name fields in braces, and secrets as\n`{secret:NAME}`; leave the body empty to send the record as JSON. A secret is\nnamed only by the automation: what a record holds is data, so `{secret:NAME}`\ntyped into a field goes out as that text.\n\nFrom the server a webhook does not call private or local addresses (`localhost`,\n`10.x`, `192.168.x`, `169.254.169.254`, a name that resolves to one, an IPv6 form\nthat wraps one) and does not follow redirects. On Node it connects to the\naddresses it checked, so a name cannot answer one way for the check and another\nfor the call; a name that does not resolve is not sent. A `fetch` of your own\npassed to `createAutomationServer` gets the check but not that pin, and neither\ndoes a runtime with no sockets of its own. For a webhook on your own network, set\n`allowInternalWebhooks: true` on `createAutomationServer`.\n\n\n![A server automation that calls a webhook: POST to https://chat.example.com/hooks/{secret:CHAT_HOOK} with the body {\"text\": \"{Task} is blocked. Assignee: {Assignee}\"}.](/docs-media/base-automation-webhook.png)\n\nValues are written as the field shows them:\n\n| Field | Write |\n|---|---|\n| Single or multiple select | The option's name: `Done`, `Bug, Risk` |\n| Person | The person's name: `Ana Ruiz` |\n| Checkbox | `true` or `false` |\n| Number | `42`; in a percent field, `100%` |\n| Date | `2026-11-01`, `today`, `today+7`, `today-1`, `now` |\n| Link | A record's name, several with commas, or `{record}` for the record that set it off |\n| Any text | Text with fields in braces: `{Task} follow-up` |\n| Any field | Nothing, to clear it |\n\n## Runs\n\nEvery run is listed under Runs, newest first, with what each action did and\nthe record it ran on. A failed action stops the run there and says why.\n\nEdit fields on more than 10,000 records asks first, and that question says how many\nruns the change can start in the page and how long they will likely take: at the rate\nruns went in this tab, or, before any did, from how long the server takes to answer. Untick\n\"Also run automations\" to make the change without them. The choice holds for that change\nonly. An automation that runs on the server is not counted and cannot be skipped from here.\n\nRuns in the page go one at a time, in the tab, after the change is saved. From 20 runs on, the top\nbar says how far they are and about how long is left, with a Stop button. Stop drops the runs\nnot started and says how many, and those do not run later. Runs left when the tab closes do\nnot happen either, so the browser asks before it closes. For a change of hundreds of thousands of records,\nrun the automation on the server (`runOn: 'server'`).\n\n![Runs: Task blocked on Paste URLs as attachments showed a notice; Done closes the task on Lightbox preview set Done to checked and Progress to 100%, then showed a notice; Start a task on Docs page set Status to In progress and Start to today.](/docs-media/base-automation-runs.png)\n\n## In code\n\nA table's automations are its `automations`, plain data you can store with\nthe rest of the table:\n\n```ts\nimport type { BaseAutomation } from '@svgrid/enterprise/base'\n\nconst statusIs = (value: string) => ({\n kind: 'fieldFilter' as const,\n conjunction: 'and' as const,\n conditions: [{ column: 'status', kind: 'select' as const, op: 'is' as const, value }],\n})\n\nexport const taskAutomations: BaseAutomation[] = [\n {\n id: 'done',\n name: 'Done closes the task',\n enabled: true,\n trigger: { type: 'recordUpdated', fields: ['status'] },\n condition: statusIs('Done'),\n actions: [\n { type: 'updateRecord', values: { done: 'true', progress: '100%' } },\n { type: 'notify', message: '{Task} is done', severity: 'success' },\n ],\n },\n {\n id: 'blocked',\n name: 'Open an issue for a blocked task',\n enabled: true,\n trigger: { type: 'recordMatches' },\n condition: statusIs('Blocked'),\n actions: [{ type: 'createRecord', table: 'issues', values: { title: 'Unblock {Task}', owner: '{Assignee}' }, linkField: 'task' }],\n },\n]\n```\n\n```svelte\n<script lang=\"ts\">\n import { SvToaster } from '@svgrid/grid'\n import { SvBase, enableBase, type BaseTable } from '@svgrid/enterprise/base'\n import { taskAutomations } from '$lib/automations'\n\n enableBase()\n let { tasks, issues }: { tasks: BaseTable; issues: BaseTable } = $props()\n let tables = $state.raw([{ ...tasks, automations: taskAutomations }, issues])\n</script>\n\n<SvBase bind:tables editable onTablesChange={(t) => console.log('save', t)} />\n<SvToaster />\n```\n\nAutomations edited in the dialog reach `onTablesChange` with the rest of the\ntable. Automations need `<SvBase tables>`; one table is a list of one.\nNotices show as a base's other notices do: see [Notices](./quick-start.md#notices).\n\n## Your own actions\n\n`registerAutomationAction` adds an action to the dialog's list. It runs in\nthe page, so call your own server for anything that needs a secret:\n\n```ts\nimport { registerAutomationAction } from '@svgrid/enterprise/base'\n\nregisterAutomationAction('post-to-chat', {\n label: 'Post to the team chat',\n run: async ({ record, text }) => {\n if (!record) return 'No record'\n const res = await fetch('/api/chat', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ message: `${text('Task')} needs a hand (${text('Assignee')})` }),\n })\n if (!res.ok) throw new Error(`Chat answered ${res.status}`)\n return 'Posted'\n },\n})\n```\n\nAn automation runs it with `{ type: 'run', action: 'post-to-chat' }`. What it\nreturns is the run's detail; what it throws is the run's error.\n\n## Where they run\n\nEach automation says where it runs, under \"Runs in\":\n\n- **This browser** (the default): in the page of the person who made the\n change, right away and once. A change that arrives from someone else\n through [realtime](./data.md#changes-made-elsewhere) does not run it again\n in your page. Scheduled ones run while the base is open.\n- **The server**: in your server, for every change to the database, whoever\n or whatever made it, with secrets and email. Set it up as below.\n\nEither way, a change an automation makes is an ordinary change: with a\n[source](./data.md), it is written to the database like an edit. What an\naction changes can set off other automations, up to four deep; past that the\nrun stops and says so, so two automations that answer each other cannot loop.\n\n## On the server\n\n![Automations on the server: the base writes through your endpoint to the database; the automation server hears each change through the endpoint's audit hook or a database webhook, writes back through the same sources, and calls webhooks and sends email with secrets that stay on the server.](/docs-media/base-automation-server.svg)\n\n`createAutomationServer` runs the automations set to \"The server\". It reads\nthe same tables, with the same automations, through sources of its own. It remembers\nthe last 10,000 records of a table it heard deleted: on a table with no `softDelete` an\nundo writes the record again, and only that memory tells it from a new one, so such an\nundo after a server restart reads as a creation.\nKeep the tables in one module both sides import, each side handing in its\nown way to reach the data:\n\n```ts\n// src/lib/tables.ts: the tables, for the browser and for the server.\nimport { baseSource, type BaseTable } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\nexport function trackerTables(source: (table: string) => ServerDataSource<Record<string, unknown>>): BaseTable[] {\n return [\n {\n id: 'tasks',\n name: 'Tasks',\n primaryField: 'name',\n data: [],\n columns: [\n { field: 'name', header: 'Task', fieldType: 'text' },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: [{ value: 'Todo' }, { value: 'Blocked' }, { value: 'Done' }] } },\n { field: 'owner_email', header: 'Owner email', fieldType: 'email' },\n ],\n automations: [\n {\n id: 'blocked',\n name: 'Tell the team about a blocked task',\n enabled: true,\n runOn: 'server',\n trigger: { type: 'recordMatches' },\n condition: { kind: 'fieldFilter', conjunction: 'and', conditions: [{ column: 'status', kind: 'select', op: 'is', value: 'Blocked' }] },\n actions: [\n { type: 'webhook', url: 'https://hooks.slack.com/services/{secret:SLACK_HOOK}', body: '{\"text\": \"{Task} is blocked\"}' },\n { type: 'email', to: '{Owner email}', subject: '{Task} is blocked', body: 'Someone marked {Task} as blocked.' },\n ],\n },\n ],\n source: baseSource(source('tasks')),\n },\n ]\n}\n```\n\nThe server side, in a SvelteKit app:\n\nThe code on this page is for SvelteKit 2. On SvelteKit 3,\n[these names differ](../../getting-started/sveltekit.md#on-sveltekit-3).\n\n```ts\n// src/lib/server/automations.ts\nimport pg from 'pg'\nimport { createAutomationServer } from '@svgrid/enterprise/base'\nimport { createSqlDataSource, type EntitySchema } from '@svgrid/enterprise'\nimport { env } from '$env/dynamic/private'\nimport { trackerTables } from '$lib/tables'\n\nconst pool = new pg.Pool({ connectionString: env.DATABASE_URL })\nconst schemas: Record<string, EntitySchema> = {\n tasks: {\n name: 'tasks',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'name', type: 'text' },\n { field: 'status', type: 'text' },\n { field: 'owner_email', type: 'text' },\n ],\n },\n}\n\n// Your mail provider's client.\ndeclare const mailer: { send: (m: { from: string; to: string[]; subject: string; text: string }) => Promise<unknown> }\n\nexport const automations = createAutomationServer({\n tables: trackerTables((table) =>\n createSqlDataSource({\n schema: schemas[table]!,\n table,\n dialect: { placeholders: '$', ilike: true },\n execute: async (sql, params) => (await pool.query(sql, params)).rows,\n }),\n ),\n // {secret:SLACK_HOOK} in a webhook reads this. List what automations may use:\n // whoever writes an automation can send any secret named here to any URL.\n secrets: { SLACK_HOOK: env.SLACK_HOOK },\n sendEmail: (m) => mailer.send({ from: 'tracker@example.com', ...m }),\n onRun: (run) => {\n if (run.status !== 'ok') console.error('automation failed', run)\n },\n})\n```\n\nThen tell it about changes, in whichever way fits your app:\n\n- **Through your endpoint.** Pass its `audit` hook to `createKitHandlers`:\n every create, update and delete through the endpoint reaches it, with the\n record as stored. It compares with what it read last, so it needs a server\n that stays up between requests. A server that has just started reads its\n tables before the first write reaches the database: the hook has a `before`\n that `createKitHandlers` waits for. So each record of a pasted block is\n judged by what it held, and a record the write left as it was runs nothing.\n Columns the row carries that no field names, such as a tenant, are not\n compared.\n\n To keep an audit hook of your own, pass it as the second argument,\n `automations.audit('tasks', (entry) => log(entry))`. It is called first with\n each entry. A hook that wraps this one loses `before`; then the server reads\n its tables after the first write, and each record of that request goes by\n the fields the write named.\n\n ```ts\n // src/routes/api/[table]/+server.ts\n import { createKitHandlers, createSqlDataSource, type EntitySchema } from '@svgrid/enterprise'\n import { automations } from '$lib/server/automations'\n\n declare const schema: EntitySchema\n declare const source: ReturnType<typeof createSqlDataSource>\n\n export const { POST } = createKitHandlers({ schema, source, audit: automations.audit('tasks') })\n ```\n\n- **From the database.** A Supabase database webhook, or a Postgres trigger\n that posts the old and new row, hands over both sides of the change.\n Nothing has to stay in memory, so it suits serverless hosts too:\n\n ```ts\n // src/routes/api/automations/changes/+server.ts\n import { error } from '@sveltejs/kit'\n import { env } from '$env/dynamic/private'\n import { automations } from '$lib/server/automations'\n\n export async function POST({ request }: { request: Request }) {\n if (request.headers.get('x-webhook-secret') !== env.WEBHOOK_SECRET) error(401, 'No')\n // Supabase sends { type, table, record, old_record }.\n const change = await request.json()\n await automations.recordChanged(change.table, { before: change.old_record, after: change.record })\n return new Response('ok')\n }\n ```\n\n- **By reading the table again**, when all you know is that it changed:\n `automations.tableChanged('tasks')`.\n\nScheduled automations run on `tick()`. Call it once a minute: from a cron\njob your host runs against an endpoint, or a timer in a long-running server.\nTwo ticks in one minute run them once.\n\n```ts\n// src/routes/api/automations/tick/+server.ts: called by your host's cron every minute.\nimport { automations } from '$lib/server/automations'\n\nexport async function POST() {\n const runs = await automations.tick()\n return new Response(JSON.stringify({ ran: runs.length }))\n}\n```\n\nThe server's runs show in the dialog next to the page's, tagged Server,\nwhen the base can fetch them:\n\n```ts\n// src/routes/api/automations/runs/+server.ts\nimport { json } from '@sveltejs/kit'\nimport { automations } from '$lib/server/automations'\n\nexport function GET() {\n return json(automations.runs)\n}\n```\n\n```svelte\n<SvBase bind:tables editable serverRuns={() => fetch('/api/automations/runs').then((r) => r.json())} />\n```\n\n![Runs with the server's among them: Tell the team about a blocked task, tagged Server, posted to chat.example.com; Email the lead, tagged Server, sent the email; Done ticks the box ran in the page.](/docs-media/base-automation-server-runs.png)\n\nThe server ignores the echo of its own writes (the webhook the database\nsends for a change it made), so one change runs each automation once.\n\nThe [Postgres demo](./postgres.md) runs a server in the page, standing in\nfor yours: set a task to Blocked and its chat message shows under \"Sent by\nthe server\"; run `update tasks set status = 'Blocked' where id = 't9'` and it\nruns for that change too.\n\n## See also\n\n- [Keep a base in a database](./data.md)\n- [Field types](../fields.md): the view bar's filter, which conditions use\n"
5181
+ "markdown": "# Automations\n\nAn automation does something on its own when something happens to a\nrecord: a task set to Done ticks its Done box, a new request goes to Todo, a\nblocked task tells its owner, a button starts the work. Each table of a base\nhas its own, under \"Automations\" in the view bar.\n\n<div data-docs-demo=\"504-project-tracker\" data-height=\"680\"></div>\n\nIn the demo, set a task's Status to Done, press Start in the Kick off\ncolumn, or send the Request a task form, then open \"5 automations\" and look\nat Runs.\n\n![An automation has three parts: when, only for records where, and do. What an action changes can set off other automations, up to four deep.](/docs-media/base-automation.svg)\n\n## Add one\n\nOpen \"Automations\" in the view bar and press \"+ New automation\". The dialog\nhas three parts:\n\n- **When**: what sets it off.\n- **Only for records where**: a condition written with the view bar's\n filter, so it compares a select's option, a person, a linked record's name\n or \"within the past week\" the way the filter does. Leave it empty for\n every record.\n- **Do**: the actions, in order.\n\n![The Automations dialog: the list of automations with an on/off switch each, and the one picked: when Status changes, only for records where Status is Done, do update the record and show a notice.](/docs-media/base-automations.png)\n\nEvery change is kept as you make it. \"Run now\" tries the automation on the\nrecord you pick, as if its trigger had happened.\n\n## When\n\n| Trigger | Runs |\n|---|---|\n| A record is created | When a record is added: \"+ Add row\", a form, a board lane, a calendar day, another automation. A record brought back by Restore or an undo of a delete is not created, and no rule runs for it |\n| A record changes | When a record is edited; pick fields to run only when one of them changes. A link made from the other table counts: adding a guest to an event from Guests changes the event's Guests field |\n| A record matches a condition | When a record starts to match the condition. It does not run again while the record keeps matching |\n| A form is sent | When a form view sends a record; pick the form, or any form of the table |\n| A button is pressed | When a Button field is pressed on a record |\n| At a scheduled time | Every day at 9, every Monday, every hour, or a cron expression: once for each record that matches the condition then, or once with no record |\n\nA paste or a fill that changes several cells of one record is one change to\nit. An undo or a redo is no change: it runs no automation.\n\n## Do\n\n| Action | What it does |\n|---|---|\n| Update the record | Sets fields of the record that set it off |\n| Create a record | Adds a record to any table of the base; \"Link it back\" fills a link field of the new record with the record that set it off |\n| Show a notice | A toast; `{Field name}` is that field of the record |\n| Color the record while it matches | Colors the whole row, or some fields, for as long as the record matches the condition |\n| Call a webhook | POSTs the record (or a body you write) to a URL: a team chat, your API. Secrets as `{secret:NAME}`, read on the server only |\n| Send an email | To, subject and text with field names; sent by the [server](#on-the-server) |\n| Your own actions | Anything the app adds with `registerAutomationAction` |\n\nA webhook's URL, headers and body name fields in braces, and secrets as\n`{secret:NAME}`; leave the body empty to send the record as JSON. A secret is\nnamed only by the automation: what a record holds is data, so `{secret:NAME}`\ntyped into a field goes out as that text.\n\nFrom the server a webhook does not call private or local addresses (`localhost`,\n`10.x`, `192.168.x`, `169.254.169.254`, a name that resolves to one, an IPv6 form\nthat wraps one) and does not follow redirects. On Node it connects to the\naddresses it checked, so a name cannot answer one way for the check and another\nfor the call; a name that does not resolve is not sent. A `fetch` of your own\npassed to `createAutomationServer` gets the check but not that pin, and neither\ndoes a runtime with no sockets of its own. For a webhook on your own network, set\n`allowInternalWebhooks: true` on `createAutomationServer`.\n\n\n![A server automation that calls a webhook: POST to https://chat.example.com/hooks/{secret:CHAT_HOOK} with the body {\"text\": \"{Task} is blocked. Assignee: {Assignee}\"}.](/docs-media/base-automation-webhook.png)\n\nValues are written as the field shows them:\n\n| Field | Write |\n|---|---|\n| Single or multiple select | The option's name: `Done`, `Bug, Risk` |\n| Person | The person's name: `Ana Ruiz` |\n| Checkbox | `true` or `false` |\n| Number | `42`; in a percent field, `100%` |\n| Date | `2026-11-01`, `today`, `today+7`, `today-1`, `now` |\n| Link | A record's name, several with commas, or `{record}` for the record that set it off |\n| Any text | Text with fields in braces: `{Task} follow-up` |\n| Any field | Nothing, to clear it |\n\n## Runs\n\nEvery run is listed under Runs, newest first, with what each action did and\nthe record it ran on. A failed action stops the run there and says why.\n\nEdit fields on more than 10,000 records asks first, and that question says how many\nruns the change can start in the page and how long they will likely take: at the rate\nruns went in this tab, or, before any did, from how long the server takes to answer. Untick\n\"Also run automations\" to make the change without them. The choice holds for that change\nonly. An automation that runs on the server is not counted and cannot be skipped from here.\n\nRuns in the page go one at a time, in the tab, after the change is saved. From 20 runs on, the top\nbar says how far they are and about how long is left, with a Stop button. Stop drops the runs\nnot started and says how many, and those do not run later. Runs left when the tab closes do\nnot happen either, so the browser asks before it closes. For a change of hundreds of thousands of records,\nrun the automation on the server (`runOn: 'server'`).\n\n![Runs: Task blocked on Paste URLs as attachments showed a notice; Done closes the task on Lightbox preview set Done to checked and Progress to 100%, then showed a notice; Start a task on Docs page set Status to In progress and Start to today.](/docs-media/base-automation-runs.png)\n\n## In code\n\nA table's automations are its `automations`, plain data you can store with\nthe rest of the table:\n\n```ts\nimport type { BaseAutomation } from '@svgrid/enterprise/base'\n\nconst statusIs = (value: string) => ({\n kind: 'fieldFilter' as const,\n conjunction: 'and' as const,\n conditions: [{ column: 'status', kind: 'select' as const, op: 'is' as const, value }],\n})\n\nexport const taskAutomations: BaseAutomation[] = [\n {\n id: 'done',\n name: 'Done closes the task',\n enabled: true,\n trigger: { type: 'recordUpdated', fields: ['status'] },\n condition: statusIs('Done'),\n actions: [\n { type: 'updateRecord', values: { done: 'true', progress: '100%' } },\n { type: 'notify', message: '{Task} is done', severity: 'success' },\n ],\n },\n {\n id: 'blocked',\n name: 'Open an issue for a blocked task',\n enabled: true,\n trigger: { type: 'recordMatches' },\n condition: statusIs('Blocked'),\n actions: [{ type: 'createRecord', table: 'issues', values: { title: 'Unblock {Task}', owner: '{Assignee}' }, linkField: 'task' }],\n },\n]\n```\n\n```svelte\n<script lang=\"ts\">\n import { SvToaster } from '@svgrid/grid'\n import { SvBase, enableBase, type BaseTable } from '@svgrid/enterprise/base'\n import { taskAutomations } from '$lib/automations'\n\n enableBase()\n let { tasks, issues }: { tasks: BaseTable; issues: BaseTable } = $props()\n let tables = $state.raw([{ ...tasks, automations: taskAutomations }, issues])\n</script>\n\n<SvBase bind:tables editable onTablesChange={(t) => console.log('save', t)} />\n<SvToaster />\n```\n\nAutomations edited in the dialog reach `onTablesChange` with the rest of the\ntable. Automations need `<SvBase tables>`; one table is a list of one.\nNotices show as a base's other notices do: see [Notices](./quick-start.md#notices).\n\n## Your own actions\n\n`registerAutomationAction` adds an action to the dialog's list. It runs in\nthe page, so call your own server for anything that needs a secret:\n\n```ts\nimport { registerAutomationAction } from '@svgrid/enterprise/base'\n\nregisterAutomationAction('post-to-chat', {\n label: 'Post to the team chat',\n run: async ({ record, text }) => {\n if (!record) return 'No record'\n const res = await fetch('/api/chat', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ message: `${text('Task')} needs a hand (${text('Assignee')})` }),\n })\n if (!res.ok) throw new Error(`Chat answered ${res.status}`)\n return 'Posted'\n },\n})\n```\n\nAn automation runs it with `{ type: 'run', action: 'post-to-chat' }`. What it\nreturns is the run's detail; what it throws is the run's error.\n\n## Where they run\n\nEach automation says where it runs, under \"Runs in\":\n\n- **This browser** (the default): in the page of the person who made the\n change, right away and once. A change that arrives from someone else\n through [realtime](./data.md#changes-made-elsewhere) does not run it again\n in your page. Scheduled ones run while the base is open.\n- **The server**: in your server, for every change to the database, whoever\n or whatever made it, with secrets and email. Set it up as below.\n\nEither way, a change an automation makes is an ordinary change: with a\n[source](./data.md), it is written to the database like an edit. What an\naction changes can set off other automations, up to four deep; past that the\nrun stops and says so, so two automations that answer each other cannot loop.\n\n## On the server\n\n![Automations on the server: the base writes through your endpoint to the database; the automation server hears each change through the endpoint's audit hook or a database webhook, writes back through the same sources, and calls webhooks and sends email with secrets that stay on the server.](/docs-media/base-automation-server.svg)\n\n`createAutomationServer` runs the automations set to \"The server\". It reads\nthe same tables, with the same automations, through sources of its own. It remembers\nthe last 10,000 records of a table it heard deleted: on a table with no `softDelete` an\nundo writes the record again, and only that memory tells it from a new one, so such an\nundo after a server restart reads as a creation.\nKeep the tables in one module both sides import, each side handing in its\nown way to reach the data:\n\n```ts\n// src/lib/tables.ts: the tables, for the browser and for the server.\nimport { baseSource, type BaseTable } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\nexport function trackerTables(source: (table: string) => ServerDataSource<Record<string, unknown>>): BaseTable[] {\n return [\n {\n id: 'tasks',\n name: 'Tasks',\n primaryField: 'name',\n data: [],\n columns: [\n { field: 'name', header: 'Task', fieldType: 'text' },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: [{ value: 'Todo' }, { value: 'Blocked' }, { value: 'Done' }] } },\n { field: 'owner_email', header: 'Owner email', fieldType: 'email' },\n ],\n automations: [\n {\n id: 'blocked',\n name: 'Tell the team about a blocked task',\n enabled: true,\n runOn: 'server',\n trigger: { type: 'recordMatches' },\n condition: { kind: 'fieldFilter', conjunction: 'and', conditions: [{ column: 'status', kind: 'select', op: 'is', value: 'Blocked' }] },\n actions: [\n { type: 'webhook', url: 'https://hooks.slack.com/services/{secret:SLACK_HOOK}', body: '{\"text\": \"{Task} is blocked\"}' },\n { type: 'email', to: '{Owner email}', subject: '{Task} is blocked', body: 'Someone marked {Task} as blocked.' },\n ],\n },\n ],\n source: baseSource(source('tasks')),\n },\n ]\n}\n```\n\nThe server side, in a SvelteKit app:\n\nThe code on this page is for SvelteKit 2. On SvelteKit 3,\n[these names differ](../../getting-started/sveltekit.md#on-sveltekit-3).\n\n```ts\n// src/lib/server/automations.ts\nimport pg from 'pg'\nimport { createAutomationServer } from '@svgrid/enterprise/base'\nimport { createSqlDataSource, type EntitySchema } from '@svgrid/enterprise'\nimport { env } from '$env/dynamic/private'\nimport { trackerTables } from '$lib/tables'\n\nconst pool = new pg.Pool({ connectionString: env.DATABASE_URL })\nconst schemas: Record<string, EntitySchema> = {\n tasks: {\n name: 'tasks',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'name', type: 'text' },\n { field: 'status', type: 'text' },\n { field: 'owner_email', type: 'text' },\n ],\n },\n}\n\n// Your mail provider's client.\ndeclare const mailer: { send: (m: { from: string; to: string[]; subject: string; text: string }) => Promise<unknown> }\n\nexport const automations = createAutomationServer({\n tables: trackerTables((table) =>\n createSqlDataSource({\n schema: schemas[table]!,\n table,\n dialect: { placeholders: '$', ilike: true },\n execute: async (sql, params) => (await pool.query(sql, params)).rows,\n }),\n ),\n // {secret:SLACK_HOOK} in a webhook reads this. List what automations may use:\n // whoever writes an automation can send any secret named here to any URL.\n secrets: { SLACK_HOOK: env.SLACK_HOOK },\n sendEmail: (m) => mailer.send({ from: 'tracker@example.com', ...m }),\n onRun: (run) => {\n if (run.status !== 'ok') console.error('automation failed', run)\n },\n})\n```\n\nThen tell it about changes, in whichever way fits your app:\n\n- **Through your endpoint.** Pass its `audit` hook to `createKitHandlers`:\n every create, update and delete through the endpoint reaches it, with the\n record as stored. It compares with what it read last, so it needs a server\n that stays up between requests. A server that has just started reads its\n tables before the first write reaches the database: the hook has a `before`\n that `createKitHandlers` waits for. So each record of a pasted block is\n judged by what it held, and a record the write left as it was runs nothing.\n Columns the row carries that no field names, such as a tenant, are not\n compared.\n\n To keep an audit hook of your own, pass it as the second argument,\n `automations.audit('tasks', (entry) => log(entry))`. It is called first with\n each entry. A hook that wraps this one loses `before`; then the server reads\n its tables after the first write, and each record of that request goes by\n the fields the write named.\n\n ```ts\n // src/routes/api/[table]/+server.ts\n import { createKitHandlers, createSqlDataSource, type EntitySchema } from '@svgrid/enterprise'\n import { automations } from '$lib/server/automations'\n\n declare const schema: EntitySchema\n declare const source: ReturnType<typeof createSqlDataSource>\n\n export const { POST } = createKitHandlers({ schema, source, audit: automations.audit('tasks') })\n ```\n\n- **From the database.** A Supabase database webhook, or a Postgres trigger\n that posts the old and new row, hands over both sides of the change.\n Nothing has to stay in memory, so it suits serverless hosts too:\n\n ```ts\n // src/routes/api/automations/changes/+server.ts\n import { error } from '@sveltejs/kit'\n import { env } from '$env/dynamic/private'\n import { automations } from '$lib/server/automations'\n\n export async function POST({ request }: { request: Request }) {\n if (request.headers.get('x-webhook-secret') !== env.WEBHOOK_SECRET) error(401, 'No')\n // Supabase sends { type, table, record, old_record }.\n const change = await request.json()\n await automations.recordChanged(change.table, { before: change.old_record, after: change.record })\n return new Response('ok')\n }\n ```\n\n- **By reading the table again**, when all you know is that it changed:\n `automations.tableChanged('tasks')`.\n\nScheduled automations run on `tick()`. Call it once a minute: from a cron\njob your host runs against an endpoint, or a timer in a long-running server.\nTwo ticks in one minute run them once.\n\n```ts\n// src/routes/api/automations/tick/+server.ts: called by your host's cron every minute.\nimport { automations } from '$lib/server/automations'\n\nexport async function POST() {\n const runs = await automations.tick()\n return new Response(JSON.stringify({ ran: runs.length }))\n}\n```\n\nThe server's runs show in the dialog next to the page's, tagged Server,\nwhen the base can fetch them:\n\n```ts\n// src/routes/api/automations/runs/+server.ts\nimport { json } from '@sveltejs/kit'\nimport { automations } from '$lib/server/automations'\n\nexport function GET() {\n return json(automations.runs)\n}\n```\n\n```svelte\n<SvBase bind:tables editable serverRuns={() => fetch('/api/automations/runs').then((r) => r.json())} />\n```\n\n![Runs with the server's among them: Tell the team about a blocked task, tagged Server, posted to chat.example.com; Email the lead, tagged Server, sent the email; Done ticks the box ran in the page.](/docs-media/base-automation-server-runs.png)\n\nThe server ignores the echo of its own writes (the webhook the database\nsends for a change it made), so one change runs each automation once.\n\nThe [Postgres demo](./postgres.md) runs a server in the page, standing in\nfor yours: set a task to Blocked and its chat message shows under \"Sent by\nthe server\"; run `update tasks set status = 'Blocked' where id = 't9'` and it\nruns for that change too.\n\n## Every automation on the server\n\nAn app whose server hears of every change (people, its API, AI agents)\ncan run all of a base's automations there and none in the page:\n\n```svelte\n<SvBase\n bind:tables\n editable\n automationsOnServer\n serverRuns={() => fetch('/api/automations/runs').then((r) => r.json())}\n runAutomation={(automation, tableId, recordId) =>\n fetch('/api/automations/run', { method: 'POST', body: JSON.stringify({ automation: automation.id, table: tableId, record: recordId }) })}\n/>\n```\n\nWith `automationsOnServer` the dialog offers no \"Runs in\", every\nautomation saved through it is `runOn: 'server'`, and none runs in the page.\n`runAutomation` is \"Run now\" in the dialog and what a Button field does:\nthe page asks the server to run the automation on that record (or on none),\nthen reads the server's runs. A Promise that rejects shows its message in\nthe dialog, or as a notice for a button. Without `runAutomation` the dialog\nsays the automation runs on the server, and a Button field that only server\nautomations listen to does nothing in the page.\n\n## See also\n\n- [Keep a base in a database](./data.md)\n- [Field types](../fields.md): the view bar's filter, which conditions use\n"
5182
5182
  },
5183
5183
  {
5184
5184
  "slug": "help/base/collaboration",
5185
5185
  "path": "docs/help/base/collaboration.md",
5186
5186
  "title": "People, comments and files",
5187
5187
  "section": "SvGrid Base",
5188
- "markdown": "# People, comments and files\n\nA tracker one person fills in needs a status column and little else. Once a\nteam uses it, people want to see who a task is assigned to, who changed the\nstatus and when, a thread of comments on the task, and the mockups attached\nto it, kept somewhere that survives a reload. This page adds those to the\ntask tracker from [Build a project tracker](./start.md).\n\n<div data-docs-demo=\"514-base-hiring-tracker\" data-height=\"680\"></div>\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\nIn the demo, double-click a candidate's card on the pipeline (or switch to\n\"All candidates\" and press Space on a row), change the Stage, leave a\ncomment, and watch the Activity column on the right of the record.\n\nThe base cannot know who your users are or where your files and comments\nlive. You tell it with four calls in the browser; the change log needs\nnothing:\n\n![Your users table goes to setFieldUsers and the session to setCurrentUser, which feed user cells, created by, last modified by and comment authors; setRecordComments connects the record view's Activity to your comments endpoint; setAttachmentUploader sends files to your storage through a presigned URL; the activity log is kept in the page](/docs-media/base-collaboration.svg)\n\nEach step below works on its own.\n\n## 1. List the people\n\nUser fields pick from one list of people that you hand to the base. A person\nis a `FieldUser`:\n\n```ts\nimport { setFieldUsers, type FieldUser } from '@svgrid/enterprise/base'\n\nconst team: FieldUser[] = [\n { id: 'u1', name: 'Ana Ruiz', email: 'ana@example.com', avatarUrl: 'https://cdn.example.com/ana.jpg' },\n { id: 'u2', name: 'Ben Okafor', email: 'ben@example.com' },\n { id: 'u3', name: 'Cleo Park', email: 'cleo@example.com' },\n]\n\n// In the browser only; step 2 shows where in SvelteKit.\nsetFieldUsers(team)\n```\n\nGive the tasks a `user` field for the owner and one with `multiple` for\nreviewers. The cell stores ids, never names, so renaming a person renames\nthem everywhere:\n\n```ts\nconst columns = [\n { field: 'task', header: 'Task', fieldType: 'text' },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: STATUS } },\n { field: 'assignee', header: 'Assignee', fieldType: 'user' },\n { field: 'reviewers', header: 'Reviewers', fieldType: 'user', fieldProps: { multiple: true } },\n]\n\nconst data = [\n { id: 't1', task: 'Write the brief', status: 'Todo', assignee: 'u1', reviewers: ['u2', 'u3'] },\n { id: 't2', task: 'Pick the fonts', status: 'In progress', assignee: 'u2', reviewers: [] },\n]\n```\n\nA cell draws an avatar and the name. With several people it overlaps up to\nfour avatars and writes the first name with a count: \"Ben Okafor +1\".\n`avatarUrl` is used when it is an `http(s):`, `blob:` or `data:image` URL;\nanything else, or no URL, draws the initials on a color worked out from the\nname, so a person keeps the same color on every page. An id that is not in\nthe list shows as the id itself.\n\nEditing a user cell opens a panel with a \"Find a person\" box and the people\nfrom the list, each with their avatar; typing filters them. If the list is\nempty the panel says \"No people (setFieldUsers)\". Text pasted or filled into\na user cell is matched against each person's id, name and email, ignoring\ncase; a multiple field takes them comma-separated and drops repeats. A name\nthat matches nobody is left out and the page says so, and with nobody left the\ncell keeps its value. With no list of people set the field takes what it is\ngiven. Copying a user cell writes the names, so a copy and paste between\nrecords keeps the people, and a copy in the same page keeps the very person when\ntwo share a name.\n`fieldProps.default` takes an id for new rows. The field reference has the\n[user field's properties](../fields.md#declare-columns-by-type).\n\nA Kanban view can stack by a single user field (`stackBy: 'assignee'`): one\nlane per person in the list, plus \"Uncategorized\" while some task has nobody.\nDragging a card to a lane reassigns the task through the grid's edit path,\nso it counts as an edit.\n\n## 2. Say who is signed in\n\n`setCurrentUser` names the person using the page. Several things read it:\n\n- `createdBy` fills new rows with their id, and `lastModifiedBy` writes it\n on every edit it watches.\n- Each comment is saved with their id as `by`.\n- Each change in a record's activity is noted with their id.\n- A filter condition with the value `'@me'` (`ME`) matches them.\n- Your own code can read it with `getCurrentUser()`.\n\nIn SvelteKit the session is on the server, so load the person and the team\nin a layout and hand both to the base in the browser:\n\n```ts\n// src/routes/(app)/+layout.server.ts\nimport { redirect } from '@sveltejs/kit'\nimport type { LayoutServerLoad } from './$types'\nimport { db } from '$lib/server/db'\n\nexport const load: LayoutServerLoad = async ({ locals }) => {\n if (!locals.user) redirect(303, '/login')\n const team = await db.query(\n 'select id, name, email, avatar_url as \"avatarUrl\" from users where team_id = $1 order by name',\n [locals.user.teamId],\n )\n return { me: { id: locals.user.id, name: locals.user.name }, team: team.rows }\n}\n```\n\n```svelte\n<!-- src/routes/(app)/+layout.svelte -->\n<script lang=\"ts\">\n import type { Snippet } from 'svelte'\n import { setCurrentUser, setFieldUsers, type FieldUser } from '@svgrid/enterprise/base'\n\n let { data, children }: { data: { me: FieldUser; team: FieldUser[] }; children: Snippet } = $props()\n\n // Again when the data changes: a sign-in as someone else.\n $effect.pre(() => {\n setFieldUsers(data.team)\n setCurrentUser(data.me)\n })\n</script>\n\n{@render children()}\n```\n\nThe four setters on this page write state shared by the whole module, so\ncall them in the browser only: an `$effect` as here (it does not run during\nserver rendering), `onMount`, or behind `browser` from `$app/environment`.\nCalled at the top of a component's `<script>`, they would also run on the\nserver and hand one request's team and user to the next. Cells and pickers\nread both values reactively, so the base can draw before they arrive. The signed-in person does not have to be in the team list for\ntheir name to show, but they only appear in the picker if they are.\n\nNow add who made each task and when, and give a new task to whoever adds it:\n\n```ts\nimport { getCurrentUser, type BaseTable } from '@svgrid/enterprise/base'\n\nconst tasks: BaseTable = {\n id: 'tasks',\n name: 'Tasks',\n primaryField: 'task',\n data,\n columns: [\n ...columns,\n { field: 'createdBy', header: 'Created by', fieldType: 'createdBy' },\n { field: 'createdAt', header: 'Created', fieldType: 'createdTime', fieldProps: { dateFormat: 'friendly' } },\n ],\n props: {\n rowAdder: { label: 'Add task', defaults: () => ({ status: 'Todo', assignee: getCurrentUser()?.id ?? null }) },\n },\n}\n```\n\nBoth fields are read-only. Every record the base makes gets the current user\nand the time: \"+ Add row\", a form, the record menu, \"+ Add a card\" on a\nKanban, a calendar day. Last modified fields start out the same way. Rows\nyou load keep whatever your database sends.\n\nOn sign-out, call `closeRecord()`, then `setCurrentUser(null)` and\n`setFieldUsers([])`.\n\n## 3. Add a \"My tasks\" view\n\nA filter condition on a user field compares it with a person's id. The value\n`'@me'` (exported as `ME`) stands for whoever is signed in when the filter\nruns, so one view works for everyone:\n\n```ts\nimport { ME, type FieldView } from '@svgrid/enterprise/base'\n\nconst views: FieldView[] = [\n {\n id: 'mine',\n name: 'My tasks',\n type: 'grid',\n state: {\n advancedFilter: {\n kind: 'fieldFilter',\n conjunction: 'or',\n conditions: [\n { column: 'assignee', kind: 'select', op: 'is', value: ME },\n { column: 'reviewers', kind: 'multi', op: 'hasAnyOf', value: [ME] },\n ],\n },\n },\n },\n { id: 'all', name: 'All tasks', type: 'grid' },\n]\n```\n\nIn the Filter panel the person picker lists \"Me\" first, so users can build\nthe same view themselves. The condition reads \"Assignee is me\".\n\nA single user field filters with the select operators (`is`, `isNot`,\n`isAnyOf`, `isNoneOf`), a multiple one with `hasAnyOf`, `hasAllOf`,\n`isExactly` and `hasNoneOf`; both also have `isEmpty` and `isNotEmpty`.\nCreated by and last modified by filter the same way, so \"Created by is me\"\nworks too. The [filter conditions](../fields.md#filter-conditions) table\nlists them all.\n\nThe view stores `'@me'`, not an id, so it can be saved with the other views\n(`onViewsChange`, or the `views` inside `onTablesChange`) and Ben sees his own\ntasks when he opens it.\n\n## 4. Open a record\n\n`expandRows` with `showRowNumbers` turns the row number into a checkbox and an\nexpand button:\n\n```svelte\n<SvBase bind:tables editable showRowNumbers expandRows onTablesChange={save} />\n```\n\nA record opens from the expand button, from Space on a selected cell (not a\ncheckbox cell, and only with `expandRows`), from \"Expand record\" in the\nright-click menu, from a gallery card, from a double click (or F2) on a\nKanban card, from a calendar event and from a Gantt task.\n\n![A candidate open as a record: each field with its type icon on the left, and the Activity column on the right with \"Ana Ruiz changed Stage\" to Offer, her comment below it and the comment box](/docs-media/base-collaboration.png)\n\nThe record shows:\n\n- The primary field as the title.\n- Every field with its type icon and name, drawn as its cell draws it. A click\n opens the cell's own editor; the value is written through the grid's edit\n path, so undo, validation, `onCellValueChange` and last-modified fields work\n the same as in the cell. Read-only fields say \"read-only\".\n- Fields hidden in the view, folded under \"Show N hidden fields\" and marked\n \"hidden in the grid\". They are read-only here.\n- Activity on the right: changes and comments, oldest first, with \"just now\",\n \"5 minutes ago\", \"3 hours ago\" or the date. The box under them takes a\n comment; Ctrl+Enter (Cmd+Enter) or Comment sends it. A base without\n `editable` still takes comments. Text not sent stays with its record until the\n page reloads: stepping to the next record leaves it behind, and coming back finds it.\n Closing the record says so.\n- Previous and next buttons that step through the rows in the view's order,\n and \"3 of 12\".\n\nA record with comments shows how many by its row number (the pointer over the\nrow swaps the count for the checkbox and expand button) and as a badge on its\nKanban card. The card's badge opens the record with the comment box focused,\nas does \"Comment\" in the right-click menu of a row or a card.\n\nOn a narrow screen the activity moves under the fields. If an edit makes the\nrecord leave the view's filter, the record says \"This row is not in the\ngrid's view any more.\" and the footer \"Not in view (12 records)\". Previous\nand next still work: they move to the records that were before and after it.\n\nApp code opens a record with the grid api that `onApiReady` hands it:\n\n```ts\nimport { openRecord, closeRecord } from '@svgrid/enterprise/base'\n\nlet api: unknown\n// <SvBase ... onApiReady={(a) => (api = a)} />\n\nopenRecord({ api, rowId: 't1' })\nopenRecord({ api, rowId: 't1', focus: 'comment' }) // with the comment box focused\ncloseRecord(api)\n```\n\n`closeRecord()` closes whatever record is open, from anywhere: call it on\nsign-out or when the route changes. `closeRecord(api)` closes it only when it\nis open on that grid. A custom field type's `cell` component gets the grid\nas `ctrl` (with `row.id`), and `openRecord({ ctrl, rowId })` takes that too.\nA button field's `fieldProps.onClick` receives `{ row, rowId, columnId, api }`,\nso a button can open its own record.\n\n## 5. Keep comments in your backend\n\nWithout a store, comments live in memory and are gone on reload. In\n`<SvBase tables>` they are kept by table id and record id for the whole\npage, so two bases that show the same table, or a base mounted again, show\nthe same comments. Give unrelated bases on one page different table ids.\nWith one table (`data` and `columns`) they are kept per grid. A comment is\na `RecordComment`, and a store is two functions:\n\n```ts {nocheck}\nexport type RecordComment = { id: string; rowId: string; tableId?: string; text: string; at: number; by?: string; mentions?: string[] }\n\nexport type RecordCommentContext = { tableId?: string; api: unknown }\n\nexport type RecordCommentStore = {\n list: (rowId: string, api: unknown, context: RecordCommentContext) => ReadonlyArray<RecordComment>\n add: (rowId: string, comment: RecordComment, api: unknown, context: RecordCommentContext) => void\n update?: (rowId: string, comment: RecordComment, api: unknown, context: RecordCommentContext) => void\n remove?: (rowId: string, commentId: string, api: unknown, context: RecordCommentContext) => void\n onMention?: (event: MentionEvent) => void\n}\n```\n\n`rowId` is the record's id (its `id`, or the table's `getRowId`). In\n`<SvBase tables>`, `context.tableId` (and the comment's `tableId`) names the\ntable, so Tasks `1` and Milestones `1` are two records; with one table it is\nundefined. `at` is milliseconds since 1970, `by` the id of the person who was\nsigned in. The record view makes `id`, `at` and `by` itself and calls `add`.\nA comment's author sees Edit and Delete under it, with no one signed in that is\nwhoever is at the page. An edited comment shows \"(edited)\" and carries `editedAt`;\nsomeone newly mentioned in it hears of it once. A store without `update` or\n`remove` offers neither.\nA store written for `list(rowId, api)` and `add(rowId, comment, api)` still\nworks.\n\nBoth functions are synchronous, and `list` runs often: for the open record,\nand for every row the grid draws (with `expandRows`) and every Kanban card,\nto show the comment count. So fetch a whole table's comments once into reactive state and answer\nfrom it. Because `list` runs inside the base's reactive code, the counts and\nthe record draw again when the fetch lands:\n\n```ts\n// src/lib/comments.svelte.ts\nimport { browser } from '$app/environment'\nimport { toast } from '@svgrid/grid'\nimport { setRecordComments, type RecordComment, type RecordCommentStore } from '@svgrid/enterprise/base'\n\ntype ByRecord = Record<string, ReadonlyArray<RecordComment>>\n\n// Each table's comments by record id, fetched once per table.\nlet byTable = $state.raw<Record<string, ByRecord>>({})\nconst asked = new Set<string>()\n\nfunction put(table: string, rowId: string, list: ReadonlyArray<RecordComment>) {\n byTable = { ...byTable, [table]: { ...byTable[table], [rowId]: list } }\n}\n\nasync function load(table: string) {\n asked.add(table)\n const res = await fetch(`/api/comments?table=${encodeURIComponent(table)}`)\n if (!res.ok) return void asked.delete(table)\n const grouped: Record<string, RecordComment[]> = {}\n for (const c of (await res.json()) as RecordComment[]) (grouped[c.rowId] ??= []).push(c)\n byTable = { ...byTable, [table]: grouped }\n}\n\n// `update` and `remove` are optional: without them a comment stays as it was written.\nconst store: RecordCommentStore = {\n list(rowId, _api, { tableId = '' }) {\n if (!asked.has(tableId)) void load(tableId)\n return byTable[tableId]?.[rowId] ?? []\n },\n add(rowId, comment, _api, { tableId = '' }) {\n const mine = () => byTable[tableId]?.[rowId] ?? []\n // Shown at once; the server's copy replaces it, or it goes away again.\n put(tableId, rowId, [...mine(), comment])\n fetch('/api/comments', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ id: comment.id, tableId, rowId, text: comment.text }),\n })\n .then(async (res) => {\n if (!res.ok) throw new Error(`HTTP ${res.status}`)\n const saved = (await res.json()) as RecordComment\n put(tableId, rowId, mine().map((c) => (c.id === comment.id ? saved : c)))\n })\n .catch(() => {\n put(tableId, rowId, mine().filter((c) => c.id !== comment.id))\n toast.error('Your comment was not saved')\n })\n },\n}\n\n// On the server this cache would be shared by every request.\nif (browser) setRecordComments(store)\n```\n\nImport the file once, in the page that shows the base. The failure notice shows\nas a base's other notices do: see [Notices](./quick-start.md#notices).\n`setRecordComments(null)` goes back to memory. For a table with a very large\ndiscussion, have the endpoint return counts per record and fetch a record's\ncomments when it opens.\n\nThe endpoint takes the author and the time from the server, never from the\nrequest body:\n\n```ts\n// src/routes/api/comments/+server.ts\nimport { error, json } from '@sveltejs/kit'\nimport type { RequestHandler } from './$types'\nimport { db } from '$lib/server/db'\n\nexport const GET: RequestHandler = async ({ url, locals }) => {\n if (!locals.user) error(401, 'Sign in first')\n const tableId = url.searchParams.get('table') ?? ''\n // `at` is a bigint column; the cast hands it back as a number, not a string.\n const { rows } = await db.query(\n 'select id, table_id as \"tableId\", row_id as \"rowId\", text, at::float8 as at, author as \"by\" from comments where team_id = $1 and table_id = $2 order by at',\n [locals.user.teamId, tableId],\n )\n return json(rows)\n}\n\nexport const POST: RequestHandler = async ({ request, locals }) => {\n if (!locals.user) error(401, 'Sign in first')\n const body = (await request.json()) as { id?: unknown; tableId?: unknown; rowId?: unknown; text?: unknown }\n const text = String(body.text ?? '').trim().slice(0, 4000)\n if (typeof body.id !== 'string' || typeof body.rowId !== 'string' || !text) error(400, 'Bad comment')\n const tableId = typeof body.tableId === 'string' ? body.tableId : ''\n const comment = { id: body.id, tableId, rowId: body.rowId, text, at: Date.now(), by: locals.user.id }\n await db.query(\n 'insert into comments (id, team_id, table_id, row_id, text, at, author) values ($1, $2, $3, $4, $5, $6, $7)',\n [comment.id, locals.user.teamId, comment.tableId, comment.rowId, comment.text, comment.at, comment.by],\n )\n return json(comment, { status: 201 })\n}\n```\n\nThe activity names the author from the people list: a `by` that is not in\n`setFieldUsers` (and is not the current user) shows as \"Someone\", and a\ncomment with no `by` shows as \"You\". Keep people who left the team in the\nlist, or return their name from your own data.\n\nComments other people add show once the cache changes. To see them while the\nbase is open, call `put` from your realtime channel or a poll.\n\n### Mentions\n\nTyping `@` in the comment box lists the people from `setFieldUsers`, narrowed\nby the letters after it (the start of any word of a name, or of an email).\nUp and Down move through the list, Enter or Tab picks, Escape closes the\nlist and leaves the record open. An `@` inside a word, as in an email\naddress, does not open it.\n\nEscape in the comment box leaves the box and keeps the text. On a comment\nbeing changed it is Cancel. While a comment has text not sent, or is being\nchanged, Escape does not close the record and says why, a click outside does\nnothing, and the Close button still closes it. An empty box holds nothing open.\n\nA mention is kept in the comment's `text` as a token, and its id in\n`mentions`:\n\n```ts {nocheck}\n{ text: 'Numbers match. @[Ana Ruiz](u1) can you get legal to sign off?', mentions: ['u1'] }\n```\n\nThe record view draws the token as a chip with the person's avatar and their\nname from the people list today, so renaming someone renames their mentions.\nA person no longer in the list shows with the name written in the token. The\nrest of the text is drawn as text, never as HTML. `mentionText(text)` turns\nthe tokens back into `@Ana Ruiz` for an email or a chat message, and\n`mentionIds(text)` reads the ids, for a server that does not trust the\n`mentions` the browser sent.\n\nTo send notifications, pass `onMention` to `<SvBase>`, give the store an\n`onMention` hook, or call `setMentionHandler` (they work with the in-memory\ndefault too). Each runs once\nper comment that mentions someone, after `add`:\n\n```ts\nimport { mentionText, setMentionHandler, type MentionEvent } from '@svgrid/enterprise/base'\n\nsetMentionHandler((e: MentionEvent) => {\n // e.comment, e.mentioned (FieldUser[]), e.rowId, e.tableId, e.record, e.api\n const to = e.mentioned.filter((p) => p.id !== e.comment.by && p.email).map((p) => p.email)\n if (!to.length) return\n void fetch('/api/notify', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ to, record: e.record?.task, text: mentionText(e.comment.text) }),\n })\n})\n```\n\n`mentioned` lists each person once, in the order they first appear,\nincluding the author when they mention themselves. Someone not in\n`setFieldUsers` comes with the name written in the token. A hook that throws\nis logged, and the comment stays. When your comments endpoint already saves\n`mentions`, sending from the server, where the author is known, is the safer\nplace for the email.\n\n<div data-docs-demo=\"712-base-mentions\" data-height=\"640\"></div>\n\n## 6. Upload files\n\nAn attachment cell holds an array of `Attachment`: `id`, `url`, `name`,\n`type` (the MIME type), and optional `size`, `width`, `height` and\n`thumbnailUrl` (a smaller image the cell uses instead of `url`). The\n[attachments section](../fields.md#attachments) of the field reference has\nan example value.\n\nWithout an uploader, added files stay in the page as `blob:` URLs and are\ngone on reload. `setAttachmentUploader` sets the function every attachment\ncolumn uses. It gets the `File` and an `AttachmentUploadContext`: `row` (the\nrecord), `columnId`, a `signal` that aborts when the user removes the file\nor closes the dialog, and `onProgress(fraction)` with a value from 0 to 1.\nIt returns the attachment.\n\nThis one asks your server for a presigned URL and sends the bytes straight\nto object storage. It uses `XMLHttpRequest` because `fetch` reports no upload\nprogress:\n\n```ts\n// src/lib/uploads.ts\nimport { browser } from '$app/environment'\nimport { setAttachmentUploader, type AttachmentUploader } from '@svgrid/enterprise/base'\n\nconst upload: AttachmentUploader = async (file, { columnId, signal, onProgress }) => {\n const res = await fetch('/api/uploads', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ name: file.name, type: file.type, size: file.size, columnId }),\n signal,\n })\n if (!res.ok) throw new Error(res.status === 413 ? 'File is too large' : `Upload refused (${res.status})`)\n const { id, uploadUrl, fileUrl } = (await res.json()) as { id: string; uploadUrl: string; fileUrl: string }\n\n await new Promise<void>((resolve, reject) => {\n const xhr = new XMLHttpRequest()\n xhr.open('PUT', uploadUrl)\n xhr.setRequestHeader('content-type', file.type || 'application/octet-stream')\n xhr.upload.onprogress = (e) => e.lengthComputable && onProgress(e.loaded / e.total)\n xhr.onload = () => (xhr.status < 300 ? resolve() : reject(new Error(`Storage said ${xhr.status}`)))\n xhr.onerror = () => reject(new Error('Network error'))\n xhr.onabort = () => reject(new Error('Cancelled'))\n signal.addEventListener('abort', () => xhr.abort(), { once: true })\n xhr.send(file)\n })\n\n return { id, url: fileUrl, name: file.name, type: file.type, size: file.size }\n}\n\nif (browser) setAttachmentUploader(upload)\n```\n\nThe message of the error you throw is what the user reads on the failed tile,\nnext to Retry and Remove, so write it for them.\n\nThe server checks the user and the size, signs a URL for one key, and keeps\na row per file:\n\n```ts\n// src/routes/api/uploads/+server.ts\nimport { error, json } from '@sveltejs/kit'\nimport type { RequestHandler } from './$types'\nimport { db } from '$lib/server/db'\nimport { presignPut } from '$lib/server/storage'\n\nconst MAX = 25 * 1024 * 1024\n\nexport const POST: RequestHandler = async ({ request, locals, url }) => {\n if (!locals.user) error(401, 'Sign in first')\n const { name, type, size } = (await request.json()) as { name: string; type: string; size: number }\n if (!(size > 0 && size <= MAX)) error(413, 'File is too large')\n const id = crypto.randomUUID()\n const safe = String(name).replace(/[^\\w.-]+/g, '_')\n const key = `${locals.user.teamId}/${id}/${safe}`\n await db.query('insert into files (id, team_id, key, type) values ($1, $2, $3, $4)', [id, locals.user.teamId, key, type])\n return json({ id, uploadUrl: await presignPut(key, type), fileUrl: `${url.origin}/api/files/${id}/${safe}` })\n}\n```\n\n`fileUrl` points at your app, not at the bucket. A presigned read URL\nexpires, and a cell keeps its URL for as long as the record exists. Serve\n`/api/files/[id]/[name]` with a `GET` that checks the user and the team,\nlooks up the key by `id` and answers `redirect(303, signedReadUrl)`:\nthumbnails, the preview and Download all follow the redirect, and old records\nkeep working.\n\nThe URL is absolute and ends in the file name for the sake of copy and\npaste. Copying an attachment cell writes the files' URLs, and a paste\nattaches each `http(s)` URL it finds, taking the name from the path and the\ntype from its extension. A relative URL would show fine in the cell but be\ndropped on paste, and a URL without the name would come back as a file named\nby its id, an image drawn as a chip.\n\nThe column limits what the dialog takes with\n[`AttachmentFieldProps`](../fields.md#attachments): `accept` (as for\n`<input accept>`, such as `'image/*,.pdf'`), `maxFiles` (default 20) and\n`maxSize` in bytes (default 25 MB). A file over a limit is listed under the\ntiles with the reason, such as \"At most 1 file\", \"Larger than 2 MB\" or \"Not\nan accepted type\". A URL added in the dialog, and text pasted into a cell,\nare checked the same way, their type read from the extension: what fails is\ndropped, the rest is kept. The field editor offers `accept` and `maxFiles`;\n`maxSize` is set in code:\n\n```ts\nconst photo = {\n field: 'photo',\n header: 'Photo',\n fieldType: 'attachment',\n fieldProps: { accept: 'image/*', maxFiles: 1, maxSize: 2 * 1024 * 1024 },\n}\n```\n\n`fieldProps.upload` sets an uploader for one column instead of the app's.\n\n### What the user sees\n\nThe cell shows images as thumbnails and other files as chips with the\nextension and the name, as many as the column width holds, then \"+N\". Enter\nor a double click opens the dialog: add files with \"Add files\", by dropping,\nby pasting files, or by pasting a URL into \"or paste a URL\". Each upload is a\ntile with its percent and a bar; Done reads \"Uploading 2...\" and stays\ndisabled until they finish. Tiles can be dragged to reorder, or moved with\nAlt+Left and Alt+Right; Delete removes one. Cancel or Escape leaves the cell\nas it was and aborts the uploads still running.\n\nFiles can also be dropped straight on a cell. Each one uploads on its own,\nwith a thin progress bar along the bottom of the cell, and the ones that\nupload are written to the cell in one edit. A file over a limit shows in a\n\"Not added\" toast and an upload that throws in an \"Upload failed\" toast, each\nwith the file name and the reason; the others still land. The toasts need\nthe same `<SvToaster />` as the comment store. Uploads still running when the\ncell goes away (scrolled out, or the table switched) are aborted through\n`signal`.\n\nEnter on a tile, or a click on its picture, opens the preview: images inline,\nPDFs in an iframe, anything else as its name with a Download link. Left and\nRight step through the files. A PDF is framed only when the page could read its\nbytes and they are a PDF, so your storage has to allow the page's origin with CORS\n(`Access-Control-Allow-Origin`). A file it cannot read, or one that is not a PDF\nwhatever it calls itself, offers its Download link instead.\n\n## 7. Track who changed what\n\n`lastModifiedTime` and `lastModifiedBy` change on every edit to the record by\ndefault. With `watch: 'specific'` they change only when one of `watchFields`\ndoes, which turns them into \"status changed at\" and \"status changed by\":\n\n```ts\nconst stamps = [\n { field: 'updatedAt', header: 'Last modified', fieldType: 'lastModifiedTime', fieldProps: { dateFormat: 'friendly' } },\n { field: 'statusAt', header: 'Status changed', fieldType: 'lastModifiedTime', fieldProps: { watch: 'specific', watchFields: ['status'] } },\n { field: 'statusBy', header: 'Status changed by', fieldType: 'lastModifiedBy', fieldProps: { watch: 'specific', watchFields: ['status'] } },\n]\n```\n\n`watchFields` holds column ids: the column's `id`, or its `field` when it has\nnone. The time is an ISO string; the person is the current user's id. With\nnobody signed in, the time still moves and the person stays as it was. A\ncommit that leaves the value as it was (an editor opened and closed with\nEnter) is not an edit: nothing is stamped and `onCellValueChange` does not\nfire. An automation's `updateRecord` moves the time but not the person.\n\nThe stamp is written into the row before `onCellValueChange` fires, so the\nevent's `row` already holds it. If you save cell by cell, the event names\nonly the edited column: send the stamps along, or have the server write\n`updated_at` and `updated_by` from its own clock and session, which a\nbrowser cannot fake. With `onTablesChange` or a table `source` ([Keep a base\nin a database](./data.md)) the stamps travel with the rest of the row.\n\n## Protected fields\n\nA protected field takes edits from the people it lists. Everyone else sees\nit read-only, with a lock after its name that says who can edit it:\n\n```ts\nconst columns = [\n { field: 'priority', header: 'Priority', fieldType: 'singleSelect', fieldProps: { options: PRIORITY, editors: ['u1', 'u2'] } },\n { field: 'approved', header: 'Approved', fieldType: 'checkbox', fieldProps: { editors: [] } },\n]\n```\n\n`editors` holds user ids, matched against `setCurrentUser`. An empty list\nmeans nobody can edit the values, and with nobody signed in a protected\nfield is read-only. In the base, Edit field > Who can edit its values sets\nthe list: \"Only these people\" starts with the person setting it, so they do\nnot lock themselves out. `protectFields={false}` on `<SvBase>` takes that\nchoice away from someone your app does not let change it.\n\nWhat the protection covers, for the person in front of the page:\n\n- typing, pasting, the fill handle, Delete, and the record view;\n- a board stacked by the field (its cards do not move) and a calendar or\n timeline whose dates it holds (its events do not drag);\n- an import into the table, which does not offer the field as a target.\n\nA field's name, type and options stay editable: protection is about the\nvalues. A new record still takes its group's value, so \"+\" on a group still\nadds a record to it, and a form fills the fields its author put on it.\n\nWith roles of your own, pass a check in place of the lists. It decides for\nevery field of every base on the page; `defaultCanEditField` is the\n`editors` check, to fall back on:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, defaultCanEditField, type CanEditField } from '@svgrid/enterprise/base'\n let { tables, role } = $props()\n // Admins edit everything; everyone else follows each field's editors.\n const canEditField: CanEditField = (field, user) => role === 'admin' || defaultCanEditField(field, user)\n</script>\n\n<SvBase bind:tables {canEditField} protectFields={role === 'admin'} editable />\n```\n\nThe browser enforces this. A base whose records live in a database still\nneeds the server to refuse a write to a protected field, as it would any\nwrite the person is not allowed.\n\n<div data-docs-demo=\"710-base-personal-locked-views\" data-height=\"620\"></div>\n\n## Share a base read-only\n\nFor people who may look but not change anything, such as a client opening a\nshare link your app serves, pass `readOnly`:\n\n```svelte\n<SvBase {tables} readOnly sortable expandRows fieldMenu />\n```\n\nWhat it closes:\n\n- cells and the record view: no typing, pasting, fill handle, Delete or\n checkbox clicks, and buttons do not run automations;\n- the record menu keeps Expand record and Copy; the field menu keeps Filter,\n Group, Hide and Color rows by this field;\n- no \"+ Add row\", no \"+\" column, no \"+\" on a group, and the selection bar\n drops Edit fields and Delete;\n- a board's cards do not move and its lanes add none; a calendar's or\n timeline's events do not drag, and a day picked adds no record;\n- the tab bar has no table menu, no renaming, no Add table, no import, no\n trash, no undo, and no Download as SQL; it shows a \"View only\" badge;\n- no Automations button, and scheduled automations do not run in the page;\n- comments are read, and the box to write one is gone;\n- form views are not shared.\n\nViewers still switch tables and views, search, filter, sort, group, and open\nrecords with their activity. What they change in a view (a sort, a filter,\na new view) stays in the page while it is open, also across table switches,\nand never reaches `onTablesChange` or a table's `source`.\n\nA share on the same page as the editing base (a preview, or the demo below)\nshows the comments written in the editor, with their counts on rows and\ncards: comments are kept by table id and record id, not by base. The\nchanges in a record's activity and the unseen marks stay with the base that\nnoted them. On a separate page, the share sees the comments your\n`setRecordComments` store returns; in-page comments do not travel between\nbrowsers.\n\nAn object picks what the share shows. Every list is optional:\n\n```svelte\n<SvBase\n {tables}\n readOnly={{\n tables: ['projects', 'clients'],\n views: ['all', 'board'],\n hideFields: ['budget', 'clients.notes'],\n }}\n/>\n```\n\n`tables` and `views` are ids; a table none of whose listed views exists\nshows its first view. `hideFields` takes a column id, which hides it in\nevery table, or `tableId.columnId`. The primary field always shows. A hidden\nfield takes along every field that would show what it holds: the formulas\nthat read it, a rollup or lookup of it in another table, and the lookups,\nrollups and counts of a link to a table the share leaves out.\n\nHiding is for display. A page handed the whole base still holds the hidden\nvalues in memory, so for anything secret build the shared copy on the server\nwith `sharedTables(tables, share)`, which also takes the hidden fields'\nvalues out of the records, and send only that.\n\n<div data-docs-demo=\"717-base-read-only-share\" data-height=\"620\"></div>\n\n## The activity log\n\nEvery edit through the grid is noted for the record: a typed value, a paste,\na fill, a clear, a checkbox click, a card dragged to another lane, an edit in\nthe record view, files added in the dialog or dropped on a cell. Each note\nholds the time, the current user's id, the column, the new value and the one\nbefore it, and the record view writes it as \"Ana Ruiz changed Status\" over the\nold value and the new one as the cell shows them, or \"(cleared)\". Undo and redo\nare noted too, as \"You undid a change to Status\" and \"You redid a change to\nStatus\", so the log ends on what the record holds. A field an automation's `updateRecord` sets is\nnoted under the automation's name: \"Automation: Done ticks the box changed\nDone\".\n\nA change someone else made in the database, read in through a table's\n`realtime` source, is noted too, under the name in the record's\n`lastModifiedBy` field, with the value before it: \"Triage agent changed\nPriority\", Low → Urgent. Until the record is opened, its changed cells are\ntinted and its row number has a dot. See [Changes made\nelsewhere](./data.md#changes-made-elsewhere).\n\nBy default the log lives in memory and keeps the latest 50 changes of each\nrecord, for the 20,000 records of a table changed longest ago last: a bulk edit of a\nmillion records does not stay in the tab. In `<SvBase tables>` it is kept by base, table and record id, so it\nsurvives switching tabs; with one table it is kept per grid. Changes made by\nyour code replacing `tables` or `data` are not noted.\n\n### A history that lasts\n\nYour audit log already has every change. `setRecordHistory` lets the record\nview read it, and hands it each change made in the page to keep:\n\n```ts\nimport { setRecordHistory, type RecordChange } from '@svgrid/enterprise/base'\n\nsetRecordHistory({\n // Each time a record is opened. A promise is fine: the activity says\n // \"Loading the history...\" until it settles.\n list: async (rowId, { tableId }) => {\n const res = await fetch(`/api/history/${tableId}/${rowId}`)\n return (await res.json()) as RecordChange[]\n },\n // Each change made in this page, as it is made.\n add: (rowId, change, { tableId }) => {\n void fetch(`/api/history/${tableId}/${rowId}`, { method: 'POST', body: JSON.stringify(change) })\n },\n})\n```\n\nA change is `{ at, by, columnId, value, from? }` (`at` in milliseconds,\n`by` a user id from `setFieldUsers`), plus `automation` for a change an\nautomation made and `via` for a proposal someone accepted. The record view\nshows what `list` gave with this session's changes it does not have yet;\ngive back the `at` that `add` received, and a change shows once. A change\nmade elsewhere and read in through `realtime` is not passed to `add`:\nwhoever made it keeps it.\n\n`historyOf(ctrl, rowId)` gives a record's history as the record view shows\nit, for a panel of your own. `setRecordHistory(null)` goes back to memory.\n\n## The record menu\n\nRight-click a record in a base's grid:\n\n| Item | What it does |\n| --- | --- |\n| Expand record | Opens the record, with or without `expandRows` |\n| Comment | Opens the record with the comment box focused |\n| Insert record above, Insert record below | A new record beside this one with the values of the fields the view groups by, so it stays in the group; editing starts on its first field |\n| Duplicate record | A copy below, with lists copied; autonumber, created and last modified fields get fresh values |\n| Copy, Paste | The grid's own cell copy and paste |\n| Delete record | Asks \"Delete record?\" (\"Write the brief is deleted, and its links from other records go with it.\"), then removes it and fires `onRowsDelete` |\n\nWithout `editable`, only Expand record, Comment and Copy show. A new record\nfrom the menu is made the way \"+ Add row\" makes one: the `rowAdder` defaults,\na new id, the next autonumber, the current user in created by, and `onRowAdd`\nfired, so a table with a database writes it. The grid scrolls to it. A Kanban\ncard's right-click menu has Expand record and Comment, and with `editable`\n\"Move to\" a lane, Duplicate record and Delete record. Passing your own\n`contextMenu` to `SvBase` replaces the grid's menu.\n\n## Pitfalls\n\n- **No current user, no author.** Without `setCurrentUser`, new rows get\n `null` in created by, and an edit leaves last modified by as it was, so\n the field names whoever edited last while signed in.\n- **People are page-wide.** `setFieldUsers` and `setCurrentUser` set one list\n and one person for every base on the page. Two bases for two teams on one\n page share them.\n- **Records need real ids.** Comments and the activity are keyed by row id:\n each record's `id`, or `getRowId`. With one table a record without an `id`\n falls back to its index, so its comments move to another record when a row\n above it is deleted. In `<SvBase tables>` every record without an `id`\n shares the id `''`, and with it one thread. Give every record an id.\n- **Your store tells tables apart by id only.** In memory, each `<SvBase>`\n keeps its own comments and activity, even when two bases on one page use\n the same table ids. Your comment store gets the table id and nothing about\n the base: if two bases share one store and their table ids, key the store\n by something of your own as well. With `serverIds: true` a new record\n keeps the page's id until the table loads again; re-key its comments when\n the insert returns.\n- **`'@me'` is whoever runs the filter.** Before `setCurrentUser` it\n matches no one; the grid filters again once a user is set. In an\n automation's condition it is the person whose edit set it off, and on the\n server runner, where nobody is signed in, it matches no one.\n- **A URL's type is its extension.** Under `accept: 'image/*'` a URL with no\n extension is refused, and a URL's size is checked only when it is known.\n The browser checks are for the user; check types and sizes on the server\n too.\n- **Storage keeps what the cell drops.** Removing a file, Cancel after an\n upload finished, or deleting the record leaves the object in the bucket.\n Clean up on the server, for example files no record has pointed at for a\n day.\n- **`fieldProps.upload` does not survive JSON.** A function in a column is\n lost when you store columns in a database through `onTablesChange`. Use\n `setAttachmentUploader` and branch on `columnId` instead.\n\n## Next\n\n- [Linked records](./linked-records.md): tasks that point at milestones,\n people that point at teams, and the records that link back.\n- [Formulas](./formulas.md): fields worked out from other fields.\n- [Views](./views.md): grids, boards, calendars and galleries over the same\n records.\n- [Forms](./forms.md): let people outside the team add records.\n- [Your own field type](./custom-field-type.md): a cell of your own, with the\n grid's `ctrl` to open records from it.\n- [Automations](./automations.md): tell the assignee when a task is blocked.\n"
5188
+ "markdown": "# People, comments and files\n\nA tracker one person fills in needs a status column and little else. Once a\nteam uses it, people want to see who a task is assigned to, who changed the\nstatus and when, a thread of comments on the task, and the mockups attached\nto it, kept somewhere that survives a reload. This page adds those to the\ntask tracker from [Build a project tracker](./start.md).\n\n<div data-docs-demo=\"514-base-hiring-tracker\" data-height=\"680\"></div>\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\nIn the demo, double-click a candidate's card on the pipeline (or switch to\n\"All candidates\" and press Space on a row), change the Stage, leave a\ncomment, and watch the Activity column on the right of the record.\n\nThe base cannot know who your users are or where your files and comments\nlive. You tell it with four calls in the browser; the change log needs\nnothing:\n\n![Your users table goes to setFieldUsers and the session to setCurrentUser, which feed user cells, created by, last modified by and comment authors; setRecordComments connects the record view's Activity to your comments endpoint; setAttachmentUploader sends files to your storage through a presigned URL; the activity log is kept in the page](/docs-media/base-collaboration.svg)\n\nEach step below works on its own.\n\n## 1. List the people\n\nUser fields pick from one list of people that you hand to the base. A person\nis a `FieldUser`:\n\n```ts\nimport { setFieldUsers, type FieldUser } from '@svgrid/enterprise/base'\n\nconst team: FieldUser[] = [\n { id: 'u1', name: 'Ana Ruiz', email: 'ana@example.com', avatarUrl: 'https://cdn.example.com/ana.jpg' },\n { id: 'u2', name: 'Ben Okafor', email: 'ben@example.com' },\n { id: 'u3', name: 'Cleo Park', email: 'cleo@example.com' },\n]\n\n// In the browser only; step 2 shows where in SvelteKit.\nsetFieldUsers(team)\n```\n\nGive the tasks a `user` field for the owner and one with `multiple` for\nreviewers. The cell stores ids, never names, so renaming a person renames\nthem everywhere:\n\n```ts\nconst columns = [\n { field: 'task', header: 'Task', fieldType: 'text' },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: STATUS } },\n { field: 'assignee', header: 'Assignee', fieldType: 'user' },\n { field: 'reviewers', header: 'Reviewers', fieldType: 'user', fieldProps: { multiple: true } },\n]\n\nconst data = [\n { id: 't1', task: 'Write the brief', status: 'Todo', assignee: 'u1', reviewers: ['u2', 'u3'] },\n { id: 't2', task: 'Pick the fonts', status: 'In progress', assignee: 'u2', reviewers: [] },\n]\n```\n\nA cell draws an avatar and the name. With several people it overlaps up to\nfour avatars and writes the first name with a count: \"Ben Okafor +1\".\n`avatarUrl` is used when it is an `http(s):`, `blob:` or `data:image` URL;\nanything else, or no URL, draws the initials on a color worked out from the\nname, so a person keeps the same color on every page. An id that is not in\nthe list shows as the id itself.\n\nEditing a user cell opens a panel with a \"Find a person\" box and the people\nfrom the list, each with their avatar; typing filters them. If the list is\nempty the panel says \"No people (setFieldUsers)\". Text pasted or filled into\na user cell is matched against each person's id, name and email, ignoring\ncase; a multiple field takes them comma-separated and drops repeats. A name\nthat matches nobody is left out and the page says so, and with nobody left the\ncell keeps its value. With no list of people set the field takes what it is\ngiven. Copying a user cell writes the names, so a copy and paste between\nrecords keeps the people, and a copy in the same page keeps the very person when\ntwo share a name.\n`fieldProps.default` takes an id for new rows. The field reference has the\n[user field's properties](../fields.md#declare-columns-by-type).\n\nA Kanban view can stack by a single user field (`stackBy: 'assignee'`): one\nlane per person in the list, plus \"Uncategorized\" while some task has nobody.\nDragging a card to a lane reassigns the task through the grid's edit path,\nso it counts as an edit.\n\n## 2. Say who is signed in\n\n`setCurrentUser` names the person using the page. Several things read it:\n\n- `createdBy` fills new rows with their id, and `lastModifiedBy` writes it\n on every edit it watches.\n- Each comment is saved with their id as `by`.\n- Each change in a record's activity is noted with their id.\n- A filter condition with the value `'@me'` (`ME`) matches them.\n- Your own code can read it with `getCurrentUser()`.\n\nIn SvelteKit the session is on the server, so load the person and the team\nin a layout and hand both to the base in the browser:\n\n```ts\n// src/routes/(app)/+layout.server.ts\nimport { redirect } from '@sveltejs/kit'\nimport type { LayoutServerLoad } from './$types'\nimport { db } from '$lib/server/db'\n\nexport const load: LayoutServerLoad = async ({ locals }) => {\n if (!locals.user) redirect(303, '/login')\n const team = await db.query(\n 'select id, name, email, avatar_url as \"avatarUrl\" from users where team_id = $1 order by name',\n [locals.user.teamId],\n )\n return { me: { id: locals.user.id, name: locals.user.name }, team: team.rows }\n}\n```\n\n```svelte\n<!-- src/routes/(app)/+layout.svelte -->\n<script lang=\"ts\">\n import type { Snippet } from 'svelte'\n import { setCurrentUser, setFieldUsers, type FieldUser } from '@svgrid/enterprise/base'\n\n let { data, children }: { data: { me: FieldUser; team: FieldUser[] }; children: Snippet } = $props()\n\n // Again when the data changes: a sign-in as someone else.\n $effect.pre(() => {\n setFieldUsers(data.team)\n setCurrentUser(data.me)\n })\n</script>\n\n{@render children()}\n```\n\nThe four setters on this page write state shared by the whole module, so\ncall them in the browser only: an `$effect` as here (it does not run during\nserver rendering), `onMount`, or behind `browser` from `$app/environment`.\nCalled at the top of a component's `<script>`, they would also run on the\nserver and hand one request's team and user to the next. Cells and pickers\nread both values reactively, so the base can draw before they arrive. The signed-in person does not have to be in the team list for\ntheir name to show, but they only appear in the picker if they are.\n\nNow add who made each task and when, and give a new task to whoever adds it:\n\n```ts\nimport { getCurrentUser, type BaseTable } from '@svgrid/enterprise/base'\n\nconst tasks: BaseTable = {\n id: 'tasks',\n name: 'Tasks',\n primaryField: 'task',\n data,\n columns: [\n ...columns,\n { field: 'createdBy', header: 'Created by', fieldType: 'createdBy' },\n { field: 'createdAt', header: 'Created', fieldType: 'createdTime', fieldProps: { dateFormat: 'friendly' } },\n ],\n props: {\n rowAdder: { label: 'Add task', defaults: () => ({ status: 'Todo', assignee: getCurrentUser()?.id ?? null }) },\n },\n}\n```\n\nBoth fields are read-only. Every record the base makes gets the current user\nand the time: \"+ Add row\", a form, the record menu, \"+ Add a card\" on a\nKanban, a calendar day. Last modified fields start out the same way. Rows\nyou load keep whatever your database sends.\n\nOn sign-out, call `closeRecord()`, then `setCurrentUser(null)` and\n`setFieldUsers([])`.\n\n## 3. Add a \"My tasks\" view\n\nA filter condition on a user field compares it with a person's id. The value\n`'@me'` (exported as `ME`) stands for whoever is signed in when the filter\nruns, so one view works for everyone:\n\n```ts\nimport { ME, type FieldView } from '@svgrid/enterprise/base'\n\nconst views: FieldView[] = [\n {\n id: 'mine',\n name: 'My tasks',\n type: 'grid',\n state: {\n advancedFilter: {\n kind: 'fieldFilter',\n conjunction: 'or',\n conditions: [\n { column: 'assignee', kind: 'select', op: 'is', value: ME },\n { column: 'reviewers', kind: 'multi', op: 'hasAnyOf', value: [ME] },\n ],\n },\n },\n },\n { id: 'all', name: 'All tasks', type: 'grid' },\n]\n```\n\nIn the Filter panel the person picker lists \"Me\" first, so users can build\nthe same view themselves. The condition reads \"Assignee is me\".\n\nA single user field filters with the select operators (`is`, `isNot`,\n`isAnyOf`, `isNoneOf`), a multiple one with `hasAnyOf`, `hasAllOf`,\n`isExactly` and `hasNoneOf`; both also have `isEmpty` and `isNotEmpty`.\nCreated by and last modified by filter the same way, so \"Created by is me\"\nworks too. The [filter conditions](../fields.md#filter-conditions) table\nlists them all.\n\nThe view stores `'@me'`, not an id, so it can be saved with the other views\n(`onViewsChange`, or the `views` inside `onTablesChange`) and Ben sees his own\ntasks when he opens it.\n\n## 4. Open a record\n\n`expandRows` with `showRowNumbers` turns the row number into a checkbox and an\nexpand button:\n\n```svelte\n<SvBase bind:tables editable showRowNumbers expandRows onTablesChange={save} />\n```\n\nA record opens from the expand button, from Space on a selected cell (not a\ncheckbox cell, and only with `expandRows`), from \"Expand record\" in the\nright-click menu, from a gallery card, from a double click (or F2) on a\nKanban card, from a calendar event and from a Gantt task.\n\n![A candidate open as a record: each field with its type icon on the left, and the Activity column on the right with \"Ana Ruiz changed Stage\" to Offer, her comment below it and the comment box](/docs-media/base-collaboration.png)\n\nThe record shows:\n\n- The primary field as the title.\n- Every field with its type icon and name, drawn as its cell draws it. A click\n opens the cell's own editor; the value is written through the grid's edit\n path, so undo, validation, `onCellValueChange` and last-modified fields work\n the same as in the cell. Read-only fields say \"read-only\".\n- Fields hidden in the view, folded under \"Show N hidden fields\" and marked\n \"hidden in the grid\". They are read-only here.\n- Activity on the right: changes and comments, oldest first, with \"just now\",\n \"5 minutes ago\", \"3 hours ago\" or the date. The box under them takes a\n comment; Ctrl+Enter (Cmd+Enter) or Comment sends it. A base without\n `editable` still takes comments. Text not sent stays with its record until the\n page reloads: stepping to the next record leaves it behind, and coming back finds it.\n Closing the record says so.\n- Previous and next buttons that step through the rows in the view's order,\n and \"3 of 12\".\n\nA record with comments shows how many by its row number (the pointer over the\nrow swaps the count for the checkbox and expand button) and as a badge on its\nKanban card. The card's badge opens the record with the comment box focused,\nas does \"Comment\" in the right-click menu of a row or a card.\n\nOn a narrow screen the activity moves under the fields. If an edit makes the\nrecord leave the view's filter, the record says \"This row is not in the\ngrid's view any more.\" and the footer \"Not in view (12 records)\". Previous\nand next still work: they move to the records that were before and after it.\n\nApp code opens a record with the grid api that `onApiReady` hands it:\n\n```ts\nimport { openRecord, closeRecord } from '@svgrid/enterprise/base'\n\nlet api: unknown\n// <SvBase ... onApiReady={(a) => (api = a)} />\n\nopenRecord({ api, rowId: 't1' })\nopenRecord({ api, rowId: 't1', focus: 'comment' }) // with the comment box focused\ncloseRecord(api)\n```\n\n`closeRecord()` closes whatever record is open, from anywhere: call it on\nsign-out or when the route changes. `closeRecord(api)` closes it only when it\nis open on that grid. A custom field type's `cell` component gets the grid\nas `ctrl` (with `row.id`), and `openRecord({ ctrl, rowId })` takes that too.\nA button field's `fieldProps.onClick` receives `{ row, rowId, columnId, api }`,\nso a button can open its own record.\n\n## 5. Keep comments in your backend\n\nWithout a store, comments live in memory and are gone on reload. In\n`<SvBase tables>` they are kept by table id and record id for the whole\npage, so two bases that show the same table, or a base mounted again, show\nthe same comments. Give unrelated bases on one page different table ids.\nWith one table (`data` and `columns`) they are kept per grid. A comment is\na `RecordComment`, and a store is two functions:\n\n```ts {nocheck}\nexport type RecordComment = { id: string; rowId: string; tableId?: string; text: string; at: number; by?: string; mentions?: string[] }\n\nexport type RecordCommentContext = { tableId?: string; api: unknown }\n\nexport type RecordCommentStore = {\n list: (rowId: string, api: unknown, context: RecordCommentContext) => ReadonlyArray<RecordComment>\n add: (rowId: string, comment: RecordComment, api: unknown, context: RecordCommentContext) => void\n update?: (rowId: string, comment: RecordComment, api: unknown, context: RecordCommentContext) => void\n remove?: (rowId: string, commentId: string, api: unknown, context: RecordCommentContext) => void\n onMention?: (event: MentionEvent) => void\n}\n```\n\n`rowId` is the record's id (its `id`, or the table's `getRowId`). In\n`<SvBase tables>`, `context.tableId` (and the comment's `tableId`) names the\ntable, so Tasks `1` and Milestones `1` are two records; with one table it is\nundefined. `at` is milliseconds since 1970, `by` the id of the person who was\nsigned in. The record view makes `id`, `at` and `by` itself and calls `add`.\nA comment's author sees Edit and Delete under it, with no one signed in that is\nwhoever is at the page. An edited comment shows \"(edited)\" and carries `editedAt`;\nsomeone newly mentioned in it hears of it once. A store without `update` or\n`remove` offers neither.\nA store written for `list(rowId, api)` and `add(rowId, comment, api)` still\nworks.\n\nBoth functions are synchronous, and `list` runs often: for the open record,\nand for every row the grid draws (with `expandRows`) and every Kanban card,\nto show the comment count. So fetch a whole table's comments once into reactive state and answer\nfrom it. Because `list` runs inside the base's reactive code, the counts and\nthe record draw again when the fetch lands:\n\n```ts\n// src/lib/comments.svelte.ts\nimport { browser } from '$app/environment'\nimport { toast } from '@svgrid/grid'\nimport { setRecordComments, type RecordComment, type RecordCommentStore } from '@svgrid/enterprise/base'\n\ntype ByRecord = Record<string, ReadonlyArray<RecordComment>>\n\n// Each table's comments by record id, fetched once per table.\nlet byTable = $state.raw<Record<string, ByRecord>>({})\nconst asked = new Set<string>()\n\nfunction put(table: string, rowId: string, list: ReadonlyArray<RecordComment>) {\n byTable = { ...byTable, [table]: { ...byTable[table], [rowId]: list } }\n}\n\nasync function load(table: string) {\n asked.add(table)\n const res = await fetch(`/api/comments?table=${encodeURIComponent(table)}`)\n if (!res.ok) return void asked.delete(table)\n const grouped: Record<string, RecordComment[]> = {}\n for (const c of (await res.json()) as RecordComment[]) (grouped[c.rowId] ??= []).push(c)\n byTable = { ...byTable, [table]: grouped }\n}\n\n// `update` and `remove` are optional: without them a comment stays as it was written.\nconst store: RecordCommentStore = {\n list(rowId, _api, { tableId = '' }) {\n if (!asked.has(tableId)) void load(tableId)\n return byTable[tableId]?.[rowId] ?? []\n },\n add(rowId, comment, _api, { tableId = '' }) {\n const mine = () => byTable[tableId]?.[rowId] ?? []\n // Shown at once; the server's copy replaces it, or it goes away again.\n put(tableId, rowId, [...mine(), comment])\n fetch('/api/comments', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ id: comment.id, tableId, rowId, text: comment.text }),\n })\n .then(async (res) => {\n if (!res.ok) throw new Error(`HTTP ${res.status}`)\n const saved = (await res.json()) as RecordComment\n put(tableId, rowId, mine().map((c) => (c.id === comment.id ? saved : c)))\n })\n .catch(() => {\n put(tableId, rowId, mine().filter((c) => c.id !== comment.id))\n toast.error('Your comment was not saved')\n })\n },\n}\n\n// On the server this cache would be shared by every request.\nif (browser) setRecordComments(store)\n```\n\nImport the file once, in the page that shows the base. The failure notice shows\nas a base's other notices do: see [Notices](./quick-start.md#notices).\n`setRecordComments(null)` goes back to memory. For a table with a very large\ndiscussion, have the endpoint return counts per record and fetch a record's\ncomments when it opens.\n\nThe endpoint takes the author and the time from the server, never from the\nrequest body:\n\n```ts\n// src/routes/api/comments/+server.ts\nimport { error, json } from '@sveltejs/kit'\nimport type { RequestHandler } from './$types'\nimport { db } from '$lib/server/db'\n\nexport const GET: RequestHandler = async ({ url, locals }) => {\n if (!locals.user) error(401, 'Sign in first')\n const tableId = url.searchParams.get('table') ?? ''\n // `at` is a bigint column; the cast hands it back as a number, not a string.\n const { rows } = await db.query(\n 'select id, table_id as \"tableId\", row_id as \"rowId\", text, at::float8 as at, author as \"by\" from comments where team_id = $1 and table_id = $2 order by at',\n [locals.user.teamId, tableId],\n )\n return json(rows)\n}\n\nexport const POST: RequestHandler = async ({ request, locals }) => {\n if (!locals.user) error(401, 'Sign in first')\n const body = (await request.json()) as { id?: unknown; tableId?: unknown; rowId?: unknown; text?: unknown }\n const text = String(body.text ?? '').trim().slice(0, 4000)\n if (typeof body.id !== 'string' || typeof body.rowId !== 'string' || !text) error(400, 'Bad comment')\n const tableId = typeof body.tableId === 'string' ? body.tableId : ''\n const comment = { id: body.id, tableId, rowId: body.rowId, text, at: Date.now(), by: locals.user.id }\n await db.query(\n 'insert into comments (id, team_id, table_id, row_id, text, at, author) values ($1, $2, $3, $4, $5, $6, $7)',\n [comment.id, locals.user.teamId, comment.tableId, comment.rowId, comment.text, comment.at, comment.by],\n )\n return json(comment, { status: 201 })\n}\n```\n\nThe activity names the author from the people list: a `by` that is not in\n`setFieldUsers` (and is not the current user) shows as \"Someone\", and a\ncomment with no `by` shows as \"You\". Keep people who left the team in the\nlist, or return their name from your own data.\n\nComments other people add show once the cache changes. To see them while the\nbase is open, call `put` from your realtime channel or a poll.\n\n### Mentions\n\nTyping `@` in the comment box lists the people from `setFieldUsers`, narrowed\nby the letters after it (the start of any word of a name, or of an email).\nUp and Down move through the list, Enter or Tab picks, Escape closes the\nlist and leaves the record open. An `@` inside a word, as in an email\naddress, does not open it.\n\nEscape in the comment box leaves the box and keeps the text. On a comment\nbeing changed it is Cancel. While a comment has text not sent, or is being\nchanged, Escape does not close the record and says why, a click outside does\nnothing, and the Close button still closes it. An empty box holds nothing open.\n\nA mention is kept in the comment's `text` as a token, and its id in\n`mentions`:\n\n```ts {nocheck}\n{ text: 'Numbers match. @[Ana Ruiz](u1) can you get legal to sign off?', mentions: ['u1'] }\n```\n\nThe record view draws the token as a chip with the person's avatar and their\nname from the people list today, so renaming someone renames their mentions.\nA person no longer in the list shows with the name written in the token. The\nrest of the text is drawn as text, never as HTML. `mentionText(text)` turns\nthe tokens back into `@Ana Ruiz` for an email or a chat message, and\n`mentionIds(text)` reads the ids, for a server that does not trust the\n`mentions` the browser sent.\n\nTo send notifications, pass `onMention` to `<SvBase>`, give the store an\n`onMention` hook, or call `setMentionHandler` (they work with the in-memory\ndefault too). Each runs once\nper comment that mentions someone, after `add`:\n\n```ts\nimport { mentionText, setMentionHandler, type MentionEvent } from '@svgrid/enterprise/base'\n\nsetMentionHandler((e: MentionEvent) => {\n // e.comment, e.mentioned (FieldUser[]), e.rowId, e.tableId, e.record, e.api\n const to = e.mentioned.filter((p) => p.id !== e.comment.by && p.email).map((p) => p.email)\n if (!to.length) return\n void fetch('/api/notify', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ to, record: e.record?.task, text: mentionText(e.comment.text) }),\n })\n})\n```\n\n`mentioned` lists each person once, in the order they first appear,\nincluding the author when they mention themselves. Someone not in\n`setFieldUsers` comes with the name written in the token. A hook that throws\nis logged, and the comment stays. When your comments endpoint already saves\n`mentions`, sending from the server, where the author is known, is the safer\nplace for the email.\n\n<div data-docs-demo=\"712-base-mentions\" data-height=\"640\"></div>\n\n## 6. Upload files\n\nAn attachment cell holds an array of `Attachment`: `id`, `url`, `name`,\n`type` (the MIME type), and optional `size`, `width`, `height` and\n`thumbnailUrl` (a smaller image the cell uses instead of `url`). The\n[attachments section](../fields.md#attachments) of the field reference has\nan example value.\n\nWithout an uploader, added files stay in the page as `blob:` URLs and are\ngone on reload. `setAttachmentUploader` sets the function every attachment\ncolumn uses. It gets the `File` and an `AttachmentUploadContext`: `row` (the\nrecord), `columnId`, a `signal` that aborts when the user removes the file\nor closes the dialog, and `onProgress(fraction)` with a value from 0 to 1.\nIt returns the attachment.\n\nThis one asks your server for a presigned URL and sends the bytes straight\nto object storage. It uses `XMLHttpRequest` because `fetch` reports no upload\nprogress:\n\n```ts\n// src/lib/uploads.ts\nimport { browser } from '$app/environment'\nimport { setAttachmentUploader, type AttachmentUploader } from '@svgrid/enterprise/base'\n\nconst upload: AttachmentUploader = async (file, { columnId, signal, onProgress }) => {\n const res = await fetch('/api/uploads', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify({ name: file.name, type: file.type, size: file.size, columnId }),\n signal,\n })\n if (!res.ok) throw new Error(res.status === 413 ? 'File is too large' : `Upload refused (${res.status})`)\n const { id, uploadUrl, fileUrl } = (await res.json()) as { id: string; uploadUrl: string; fileUrl: string }\n\n await new Promise<void>((resolve, reject) => {\n const xhr = new XMLHttpRequest()\n xhr.open('PUT', uploadUrl)\n xhr.setRequestHeader('content-type', file.type || 'application/octet-stream')\n xhr.upload.onprogress = (e) => e.lengthComputable && onProgress(e.loaded / e.total)\n xhr.onload = () => (xhr.status < 300 ? resolve() : reject(new Error(`Storage said ${xhr.status}`)))\n xhr.onerror = () => reject(new Error('Network error'))\n xhr.onabort = () => reject(new Error('Cancelled'))\n signal.addEventListener('abort', () => xhr.abort(), { once: true })\n xhr.send(file)\n })\n\n return { id, url: fileUrl, name: file.name, type: file.type, size: file.size }\n}\n\nif (browser) setAttachmentUploader(upload)\n```\n\nThe message of the error you throw is what the user reads on the failed tile,\nnext to Retry and Remove, so write it for them.\n\nThe server checks the user and the size, signs a URL for one key, and keeps\na row per file:\n\n```ts\n// src/routes/api/uploads/+server.ts\nimport { error, json } from '@sveltejs/kit'\nimport type { RequestHandler } from './$types'\nimport { db } from '$lib/server/db'\nimport { presignPut } from '$lib/server/storage'\n\nconst MAX = 25 * 1024 * 1024\n\nexport const POST: RequestHandler = async ({ request, locals, url }) => {\n if (!locals.user) error(401, 'Sign in first')\n const { name, type, size } = (await request.json()) as { name: string; type: string; size: number }\n if (!(size > 0 && size <= MAX)) error(413, 'File is too large')\n const id = crypto.randomUUID()\n const safe = String(name).replace(/[^\\w.-]+/g, '_')\n const key = `${locals.user.teamId}/${id}/${safe}`\n await db.query('insert into files (id, team_id, key, type) values ($1, $2, $3, $4)', [id, locals.user.teamId, key, type])\n return json({ id, uploadUrl: await presignPut(key, type), fileUrl: `${url.origin}/api/files/${id}/${safe}` })\n}\n```\n\n`fileUrl` points at your app, not at the bucket. A presigned read URL\nexpires, and a cell keeps its URL for as long as the record exists. Serve\n`/api/files/[id]/[name]` with a `GET` that checks the user and the team,\nlooks up the key by `id` and answers `redirect(303, signedReadUrl)`:\nthumbnails, the preview and Download all follow the redirect, and old records\nkeep working.\n\nThe URL is absolute and ends in the file name for the sake of copy and\npaste. Copying an attachment cell writes the files' URLs, and a paste\nattaches each `http(s)` URL it finds, taking the name from the path and the\ntype from its extension. A relative URL would show fine in the cell but be\ndropped on paste, and a URL without the name would come back as a file named\nby its id, an image drawn as a chip.\n\nThe column limits what the dialog takes with\n[`AttachmentFieldProps`](../fields.md#attachments): `accept` (as for\n`<input accept>`, such as `'image/*,.pdf'`), `maxFiles` (default 20) and\n`maxSize` in bytes (default 25 MB). A file over a limit is listed under the\ntiles with the reason, such as \"At most 1 file\", \"Larger than 2 MB\" or \"Not\nan accepted type\". A URL added in the dialog, and text pasted into a cell,\nare checked the same way, their type read from the extension: what fails is\ndropped, the rest is kept. The field editor offers `accept` and `maxFiles`;\n`maxSize` is set in code:\n\n```ts\nconst photo = {\n field: 'photo',\n header: 'Photo',\n fieldType: 'attachment',\n fieldProps: { accept: 'image/*', maxFiles: 1, maxSize: 2 * 1024 * 1024 },\n}\n```\n\n`fieldProps.upload` sets an uploader for one column instead of the app's.\n\n### What the user sees\n\nThe cell shows images as thumbnails and other files as chips with the\nextension and the name, as many as the column width holds, then \"+N\". Enter\nor a double click opens the dialog: add files with \"Add files\", by dropping,\nby pasting files, or by pasting a URL into \"or paste a URL\". Each upload is a\ntile with its percent and a bar; Done reads \"Uploading 2...\" and stays\ndisabled until they finish. Tiles can be dragged to reorder, or moved with\nAlt+Left and Alt+Right; Delete removes one. Cancel or Escape leaves the cell\nas it was and aborts the uploads still running.\n\nFiles can also be dropped straight on a cell. Each one uploads on its own,\nwith a thin progress bar along the bottom of the cell, and the ones that\nupload are written to the cell in one edit. A file over a limit shows in a\n\"Not added\" toast and an upload that throws in an \"Upload failed\" toast, each\nwith the file name and the reason; the others still land. The toasts need\nthe same `<SvToaster />` as the comment store. Uploads still running when the\ncell goes away (scrolled out, or the table switched) are aborted through\n`signal`.\n\nEnter on a tile, or a click on its picture, opens the preview: images inline,\nPDFs in an iframe, anything else as its name with a Download link. Left and\nRight step through the files. A PDF is framed only when the page could read its\nbytes and they are a PDF, so your storage has to allow the page's origin with CORS\n(`Access-Control-Allow-Origin`). A file it cannot read, or one that is not a PDF\nwhatever it calls itself, offers its Download link instead.\n\n## 7. Track who changed what\n\n`lastModifiedTime` and `lastModifiedBy` change on every edit to the record by\ndefault. With `watch: 'specific'` they change only when one of `watchFields`\ndoes, which turns them into \"status changed at\" and \"status changed by\":\n\n```ts\nconst stamps = [\n { field: 'updatedAt', header: 'Last modified', fieldType: 'lastModifiedTime', fieldProps: { dateFormat: 'friendly' } },\n { field: 'statusAt', header: 'Status changed', fieldType: 'lastModifiedTime', fieldProps: { watch: 'specific', watchFields: ['status'] } },\n { field: 'statusBy', header: 'Status changed by', fieldType: 'lastModifiedBy', fieldProps: { watch: 'specific', watchFields: ['status'] } },\n]\n```\n\n`watchFields` holds column ids: the column's `id`, or its `field` when it has\nnone. The time is an ISO string; the person is the current user's id. With\nnobody signed in, the time still moves and the person stays as it was. A\ncommit that leaves the value as it was (an editor opened and closed with\nEnter) is not an edit: nothing is stamped and `onCellValueChange` does not\nfire. An automation's `updateRecord` moves the time but not the person.\n\nThe stamp is written into the row before `onCellValueChange` fires, so the\nevent's `row` already holds it. If you save cell by cell, the event names\nonly the edited column: send the stamps along, or have the server write\n`updated_at` and `updated_by` from its own clock and session, which a\nbrowser cannot fake. With `onTablesChange` or a table `source` ([Keep a base\nin a database](./data.md)) the stamps travel with the rest of the row.\n\n## Protected fields\n\nA protected field takes edits from the people it lists. Everyone else sees\nit read-only, with a lock after its name that says who can edit it:\n\n```ts\nconst columns = [\n { field: 'priority', header: 'Priority', fieldType: 'singleSelect', fieldProps: { options: PRIORITY, editors: ['u1', 'u2'] } },\n { field: 'approved', header: 'Approved', fieldType: 'checkbox', fieldProps: { editors: [] } },\n]\n```\n\n`editors` holds user ids, matched against `setCurrentUser`. An empty list\nmeans nobody can edit the values, and with nobody signed in a protected\nfield is read-only. In the base, Edit field > Who can edit its values sets\nthe list: \"Only these people\" starts with the person setting it, so they do\nnot lock themselves out. `protectFields={false}` on `<SvBase>` takes that\nchoice away from someone your app does not let change it.\n\nWhat the protection covers, for the person in front of the page:\n\n- typing, pasting, the fill handle, Delete, and the record view;\n- a board stacked by the field (its cards do not move) and a calendar or\n timeline whose dates it holds (its events do not drag);\n- an import into the table, which does not offer the field as a target.\n\nA field's name, type and options stay editable: protection is about the\nvalues. A new record still takes its group's value, so \"+\" on a group still\nadds a record to it, and a form fills the fields its author put on it.\n\nWith roles of your own, pass a check in place of the lists. It decides for\nevery field of every base on the page; `defaultCanEditField` is the\n`editors` check, to fall back on:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, defaultCanEditField, type CanEditField } from '@svgrid/enterprise/base'\n let { tables, role } = $props()\n // Admins edit everything; everyone else follows each field's editors.\n const canEditField: CanEditField = (field, user) => role === 'admin' || defaultCanEditField(field, user)\n</script>\n\n<SvBase bind:tables {canEditField} protectFields={role === 'admin'} editable />\n```\n\nThe browser enforces this. A base whose records live in a database still\nneeds the server to refuse a write to a protected field, as it would any\nwrite the person is not allowed.\n\n<div data-docs-demo=\"710-base-personal-locked-views\" data-height=\"620\"></div>\n\n## Share a base read-only\n\nFor people who may look but not change anything, such as a client opening a\nshare link your app serves, pass `readOnly`:\n\n```svelte\n<SvBase {tables} readOnly sortable expandRows fieldMenu />\n```\n\nWhat it closes:\n\n- cells and the record view: no typing, pasting, fill handle, Delete or\n checkbox clicks, and buttons do not run automations;\n- the record menu keeps Expand record and Copy; the field menu keeps Filter,\n Group, Hide and Color rows by this field;\n- no \"+ Add row\", no \"+\" column, no \"+\" on a group, and the selection bar\n drops Edit fields and Delete;\n- a board's cards do not move and its lanes add none; a calendar's or\n timeline's events do not drag, and a day picked adds no record;\n- the tab bar has no table menu, no renaming, no Add table, no import, no\n trash, no undo, and no Download as SQL; it shows a \"View only\" badge;\n- no Automations button, and scheduled automations do not run in the page;\n- comments are read, and the box to write one is gone;\n- form views are not shared.\n\nViewers still switch tables and views, search, filter, sort, group, and open\nrecords with their activity. What they change in a view (a sort, a filter,\na new view) stays in the page while it is open, also across table switches,\nand never reaches `onTablesChange` or a table's `source`.\n\nA share on the same page as the editing base (a preview, or the demo below)\nshows the comments written in the editor, with their counts on rows and\ncards: comments are kept by table id and record id, not by base. The\nchanges in a record's activity and the unseen marks stay with the base that\nnoted them. On a separate page, the share sees the comments your\n`setRecordComments` store returns; in-page comments do not travel between\nbrowsers.\n\nAn object picks what the share shows. Every list is optional:\n\n```svelte\n<SvBase\n {tables}\n readOnly={{\n tables: ['projects', 'clients'],\n views: ['all', 'board'],\n hideFields: ['budget', 'clients.notes'],\n }}\n/>\n```\n\n`tables` and `views` are ids; a table none of whose listed views exists\nshows its first view. `hideFields` takes a column id, which hides it in\nevery table, or `tableId.columnId`. The primary field always shows. A hidden\nfield takes along every field that would show what it holds: the formulas\nthat read it, a rollup or lookup of it in another table, and the lookups,\nrollups and counts of a link to a table the share leaves out.\n\nHiding is for display. A page handed the whole base still holds the hidden\nvalues in memory, so for anything secret build the shared copy on the server\nwith `sharedTables(tables, share)`, which also takes the hidden fields'\nvalues out of the records, and send only that.\n\n<div data-docs-demo=\"717-base-read-only-share\" data-height=\"620\"></div>\n\n## The activity log\n\nEvery edit through the grid is noted for the record: a typed value, a paste,\na fill, a clear, a checkbox click, a card dragged to another lane, an edit in\nthe record view, files added in the dialog or dropped on a cell. Each note\nholds the time, the current user's id, the column, the new value and the one\nbefore it, and the record view writes it as \"Ana Ruiz changed Status\" over the\nold value and the new one as the cell shows them, or \"(cleared)\". Undo and redo\nare noted too, as \"You undid a change to Status\" and \"You redid a change to\nStatus\", so the log ends on what the record holds. A field an automation's `updateRecord` sets is\nnoted under the automation's name: \"Automation: Done ticks the box changed\nDone\".\n\nA change someone else made in the database, read in through a table's\n`realtime` source, is noted too, under the name in the record's\n`lastModifiedBy` field, with the value before it: \"Triage agent changed\nPriority\", Low → Urgent. Until the record is opened, its changed cells are\ntinted and its row number has a dot. See [Changes made\nelsewhere](./data.md#changes-made-elsewhere).\n\nBy default the log lives in memory and keeps the latest 50 changes of each\nrecord, for the 20,000 records of a table changed longest ago last: a bulk edit of a\nmillion records does not stay in the tab. In `<SvBase tables>` it is kept by base, table and record id, so it\nsurvives switching tabs; with one table it is kept per grid. Changes made by\nyour code replacing `tables` or `data` are not noted.\n\n### A history that lasts\n\nYour audit log already has every change. `setRecordHistory` lets the record\nview read it, and hands it each change made in the page to keep:\n\n```ts\nimport { setRecordHistory, type RecordChange } from '@svgrid/enterprise/base'\n\nsetRecordHistory({\n // Each time a record is opened. A promise is fine: the activity says\n // \"Loading the history...\" until it settles.\n list: async (rowId, { tableId }) => {\n const res = await fetch(`/api/history/${tableId}/${rowId}`)\n return (await res.json()) as RecordChange[]\n },\n // Each change made in this page, as it is made.\n add: (rowId, change, { tableId }) => {\n void fetch(`/api/history/${tableId}/${rowId}`, { method: 'POST', body: JSON.stringify(change) })\n },\n})\n```\n\nA change is `{ at, by, columnId, value, from? }` (`at` in milliseconds,\n`by` a user id from `setFieldUsers`), plus `automation` for a change an\nautomation made, `via` for a proposal someone accepted (\"CRM agent (accepted\nby Ana)\"), and `through` for a change a person made through an app as\nthemselves, an AI chat client say (\"Ana via Claude\"). The record view\nshows what `list` gave with this session's changes it does not have yet;\ngive back the `at` that `add` received, and a change shows once. A change\nmade elsewhere and read in through `realtime` is not passed to `add`:\nwhoever made it keeps it.\n\n`historyOf(ctrl, rowId)` gives a record's history as the record view shows\nit, for a panel of your own. `setRecordHistory(null)` goes back to memory.\n\n## The record menu\n\nRight-click a record in a base's grid:\n\n| Item | What it does |\n| --- | --- |\n| Expand record | Opens the record, with or without `expandRows` |\n| Comment | Opens the record with the comment box focused |\n| Insert record above, Insert record below | A new record beside this one with the values of the fields the view groups by, so it stays in the group; editing starts on its first field |\n| Duplicate record | A copy below, with lists copied; autonumber, created and last modified fields get fresh values |\n| Copy, Paste | The grid's own cell copy and paste |\n| Delete record | Asks \"Delete record?\" (\"Write the brief is deleted, and its links from other records go with it.\"), then removes it and fires `onRowsDelete` |\n\nWithout `editable`, only Expand record, Comment and Copy show. A new record\nfrom the menu is made the way \"+ Add row\" makes one: the `rowAdder` defaults,\na new id, the next autonumber, the current user in created by, and `onRowAdd`\nfired, so a table with a database writes it. The grid scrolls to it. A Kanban\ncard's right-click menu has Expand record and Comment, and with `editable`\n\"Move to\" a lane, Duplicate record and Delete record. Passing your own\n`contextMenu` to `SvBase` replaces the grid's menu.\n\n## Pitfalls\n\n- **No current user, no author.** Without `setCurrentUser`, new rows get\n `null` in created by, and an edit leaves last modified by as it was, so\n the field names whoever edited last while signed in.\n- **People are page-wide.** `setFieldUsers` and `setCurrentUser` set one list\n and one person for every base on the page. Two bases for two teams on one\n page share them.\n- **Records need real ids.** Comments and the activity are keyed by row id:\n each record's `id`, or `getRowId`. With one table a record without an `id`\n falls back to its index, so its comments move to another record when a row\n above it is deleted. In `<SvBase tables>` every record without an `id`\n shares the id `''`, and with it one thread. Give every record an id.\n- **Your store tells tables apart by id only.** In memory, each `<SvBase>`\n keeps its own comments and activity, even when two bases on one page use\n the same table ids. Your comment store gets the table id and nothing about\n the base: if two bases share one store and their table ids, key the store\n by something of your own as well. With `serverIds: true` a new record\n keeps the page's id until the table loads again; re-key its comments when\n the insert returns.\n- **`'@me'` is whoever runs the filter.** Before `setCurrentUser` it\n matches no one; the grid filters again once a user is set. In an\n automation's condition it is the person whose edit set it off, and on the\n server runner, where nobody is signed in, it matches no one.\n- **A URL's type is its extension.** Under `accept: 'image/*'` a URL with no\n extension is refused, and a URL's size is checked only when it is known.\n The browser checks are for the user; check types and sizes on the server\n too.\n- **Storage keeps what the cell drops.** Removing a file, Cancel after an\n upload finished, or deleting the record leaves the object in the bucket.\n Clean up on the server, for example files no record has pointed at for a\n day.\n- **`fieldProps.upload` does not survive JSON.** A function in a column is\n lost when you store columns in a database through `onTablesChange`. Use\n `setAttachmentUploader` and branch on `columnId` instead.\n\n## Next\n\n- [Linked records](./linked-records.md): tasks that point at milestones,\n people that point at teams, and the records that link back.\n- [Formulas](./formulas.md): fields worked out from other fields.\n- [Views](./views.md): grids, boards, calendars and galleries over the same\n records.\n- [Forms](./forms.md): let people outside the team add records.\n- [Your own field type](./custom-field-type.md): a cell of your own, with the\n grid's `ctrl` to open records from it.\n- [Automations](./automations.md): tell the assignee when a task is blocked.\n"
5189
5189
  },
5190
5190
  {
5191
5191
  "slug": "help/base/custom-field-type",
5192
5192
  "path": "docs/help/base/custom-field-type.md",
5193
5193
  "title": "Build your own field type",
5194
5194
  "section": "SvGrid Base",
5195
- "markdown": "# Build your own field type\n\nA field type decides what a column holds and how it behaves: what the cell\nshows, which editor opens, how pasted text is read, what the footer can sum.\nThe built-in types are registered with `registerFieldType`, the same call\nyou will use, so your type gets the field editor, the views and the\nexpanded record the way they do.\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\nThis page builds two types. **Phone (international)** stores numbers as\nE.164 text (`+14155550123`), shows them the local way, dials on click and\nreads whatever people paste. **Traffic light** stores a score from 0 to 100\nand draws a green, amber or red light with its own cell and its own editor.\nThe [field types page](../fields.md#your-own-field-type) has the short\nversion.\n\n![A FieldTypeDef on the left, where each member shows up on the right: id, label, icon and group in the type picker and the header; fill's formatter in cell text, footer summaries, group headers and cards; valueParser for typed edits, paste, forms and type changes; editorType for the editor; sortValue for sort order; cellDataType for filter operators, summary choices, sort labels and the SQL column type; cell and hasCell for grid cells, group headers, cards and the record view; properties and defaults in the Edit field dialog; newRowValue for new records; onCreate for existing rows when the field is created; copy: false for a duplicated record, which gets a new value from newRowValue](/docs-media/base-field-type.svg)\n\n## 1. Decide what the field stores\n\nSettle the stored value before you write the type. Filters, sorting, copy,\nthe database and your API all see the stored value; only the cell and the\nformatter see the text on screen. A phone number stored as `+14155550123`\ndials from any app, sorts by country and compares exactly.\n\n```ts\n// phone-field.ts\nexport type PhoneProps = { country: string; display: 'national' | 'international' }\n\nconst E164 = /^\\+[1-9]\\d{6,14}$/\n\n/** Typed or pasted text as E.164 (\"+14155550123\"), or null when it is no phone number. */\nexport function toE164(text: string, country: string): string | null {\n let s = text.trim().replace(/^tel:/i, '').replace(/[\\s().\\-\\/]/g, '')\n if (s.startsWith('00')) s = '+' + s.slice(2)\n if (!s.startsWith('+')) {\n // A national number: drop the trunk prefix (1 in North America, 0 elsewhere).\n s = country === '1' ? s.replace(/^1(?=\\d{10}$)/, '') : s.replace(/^0/, '')\n // North American numbers need all ten digits: no guessing an area code.\n if (country === '1' && !/^\\d{10}$/.test(s)) return null\n s = `+${country}${s}`\n }\n return E164.test(s) ? s : null\n}\n\n/** A stored value as the field shows it. A number from another country shows as stored. */\nexport function showPhone(value: unknown, p: PhoneProps): string {\n const s = value == null ? '' : String(value)\n const home = `+${p.country}`\n if (!E164.test(s) || !s.startsWith(home)) return s\n const rest = s.slice(home.length)\n if (p.country === '1' && rest.length === 10) {\n const local = `(${rest.slice(0, 3)}) ${rest.slice(3, 6)}-${rest.slice(6)}`\n return p.display === 'national' ? local : `+1 ${local}`\n }\n const grouped = rest.replace(/\\B(?=(\\d{4})+$)/g, ' ')\n return p.display === 'national' ? `0${grouped}` : `${home} ${grouped}`\n}\n```\n\n| Typed, home country | Stored | Shown (national) |\n|---|---|---|\n| `(415) 555-0123`, +1 | `+14155550123` | `(415) 555-0123` |\n| `1 415 555 0123`, +1 | `+14155550123` | `(415) 555-0123` |\n| `020 7946 0958`, +44 | `+442079460958` | `020 7946 0958` |\n| `0044 20 7946 0958`, +1 | `+442079460958` | `+442079460958` |\n| `555-0123`, +1 | refused | |\n| `n/a` | refused | |\n\nBoth are plain functions, so you can unit-test them without a grid.\n\n## 2. Register the type\n\n```ts\n// phone-field.ts, continued\nimport { registerFieldType, type FieldTypeDef } from '@svgrid/enterprise/base'\n\nexport const phoneType: FieldTypeDef<PhoneProps> = {\n id: 'intlPhone',\n label: 'Phone (international)',\n description: 'Stored as +14155550123, shown in the local format, dialed with a click.',\n icon: 'M5 4h4l2 5-2.5 1.5a11 11 0 0 0 5 5L15 13l5 2v4a2 2 0 0 1-2 2A16 16 0 0 1 3 6a2 2 0 0 1 2-2',\n group: 'basic',\n width: 170,\n defaults: { country: '1', display: 'national' },\n fill: (def, props) => ({\n ...def,\n editorType: def.editorType ?? 'text',\n formatter: def.formatter ?? ((ctx: { value: unknown }) => showPhone(ctx.value, props)),\n // Typed and pasted text is stored as E.164; text that is no number keeps the old value.\n valueParser: def.valueParser ?? ((x: { newValue: unknown; oldValue: unknown; rawInput?: unknown }) => {\n const text = String((typeof x.rawInput === 'string' ? x.rawInput : x.newValue) ?? '').trim()\n if (!text) return null\n return toE164(text, props.country) ?? x.oldValue\n }),\n // Values that never went through the parser (loaded data, an API write) are flagged.\n validate: def.validate ?? ((ctx: { value: unknown }) =>\n ctx.value == null || ctx.value === '' || E164.test(String(ctx.value)) ? null : 'Not a +country phone number'),\n }),\n}\n\nregisterFieldType(phoneType)\n```\n\nImport the file once at app start, next to `enableBase()`, and name the type\nin a column:\n\n```ts\nimport { enableBase } from '@svgrid/enterprise/base'\nimport './phone-field'\n\nenableBase()\n\nconst columns = [\n { field: 'name', header: 'Name', fieldType: 'text' },\n { field: 'phone', header: 'Phone', fieldType: 'intlPhone', fieldProps: { country: '44' } },\n]\n```\n\n`fill` is where a type turns into grid behavior. It receives the column as\nthe app declared it, the type's `defaults` with the column's `fieldProps` on\ntop (`props`), and a context with every column. It returns the column the\ngrid sees: `editorType`, `formatter`, `valueParser`, `validate`, `align`,\n`cellDataType`, `sortValue`, `editable`. Write each key as\n`def.key ?? yours`, so a column that sets `formatter` itself keeps it. The\nbuilt-ins in `builtin-types.ts` follow that rule.\n\nWith this registered, the header shows the phone icon before \"Phone\", cells\nread `020 7946 0958`, and \"Phone (international)\" appears in the field\neditor's type picker under \"Fields\", with its description under the picker\nonce it is chosen. `width` is the width a new field of this type starts\nwith when someone adds it there.\n\n![The Add field dialog in a base, its type list scrolled to show the end of the Fields section (Phone number, Email, URL, Barcode, Button), then Numbers and time (Number, Duration, Rating) and Computed and automatic (Formula, Created time, Last modified time, Created by), with the chosen type's description and its Default value property below](/docs-media/base-field-type.png)\n\nThe picker draws a section per `group`, in a fixed order, and skips a\nsection no type uses. The built-ins fill only Fields, Numbers and time,\nComputed and automatic, and Other tables (Date and User sit under Fields),\nso \"Dates\" and \"People\" appear only once a type of yours names them. A\n`group` the list does not know puts the type under Fields. Once more than\neight types are on offer, a \"Find a type\" box sits above the list.\n\n## 3. Read what people type and paste\n\n`valueParser` gets `newValue` (after the grid's own coercion for the\neditor type), `oldValue`, `rawInput` (the text as typed or pasted), the\nrow as `data` and the `columnId`. It runs on a cell edit, a paste (one\nvalue pasted over a range included), a full-row edit, an edit in the\nexpanded record and a form field. Returning `oldValue` refuses the input:\npaste `n/a` over a number and the number stays, and the page says that\nvalue did not fit. The text editor\nopens on the stored `+442079460958`, and whatever the user leaves in it\ngoes back through the parser.\n\n`validate` runs for every rendered cell, so a bad value gets the red mark\nand the message as its tooltip even when it was loaded that way. While the\ncell's editor is open, it checks the draft before `valueParser` has read\nit, so `(415) 555-0123` shows red until Enter stores it as E.164. Add\n`rejectInvalid: def.rejectInvalid ?? true` to refuse a flagged edit\noutright; a paste is not checked against it, which is why the parser does\nthe refusing here.\n\nCheck the filled column without mounting anything:\n\n```ts\nimport { fillFieldColumn } from '@svgrid/enterprise/base'\n\nconst col = fillFieldColumn({ field: 'phone', fieldType: 'intlPhone', fieldProps: { country: '44' } })\nconst show = col.formatter as (ctx: { value: unknown }) => string\nconst parse = col.valueParser as (x: { newValue: unknown; oldValue: unknown; rawInput: string }) => unknown\n\nshow({ value: '+442079460958' }) // '020 7946 0958'\nparse({ newValue: '020 7946 0958', oldValue: null, rawInput: '020 7946 0958' }) // '+442079460958'\nparse({ newValue: 'n/a', oldValue: '+442079460958', rawInput: 'n/a' }) // '+442079460958'\n```\n\n## 4. Draw the cell\n\nWithout `cell`, the grid draws the formatter's text. A cell component can\ndraw anything; this one is a `tel:` link, so a click dials.\n\n```svelte\n<!-- PhoneCell.svelte -->\n<script lang=\"ts\">\n import { linkHref, type FieldCellProps } from '@svgrid/enterprise/base'\n import { showPhone, type PhoneProps } from './phone-field'\n\n let { value, props }: FieldCellProps<PhoneProps> = $props()\n // 'tel:+14155550123'; null unless the value is 3 to 20 digits with an optional +.\n const href = $derived(linkHref('phone', value))\n const text = $derived(showPhone(value, props))\n</script>\n\n{#if href}\n <!-- Stop pointerdown: a click on the number dials, a click beside it selects the cell. -->\n <a class=\"phone\" {href} onpointerdown={(e) => e.stopPropagation()}>{text}</a>\n{:else}\n <span class=\"phone is-bad\">{text}</span>\n{/if}\n\n<style>\n .phone { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; color: inherit; }\n a.phone { text-decoration: underline; text-decoration-color: var(--sg-muted, #94a3b8); text-underline-offset: 2px; }\n .is-bad { color: var(--sg-muted, #64748b); }\n</style>\n```\n\n```ts\nimport { registerFieldType } from '@svgrid/enterprise/base'\nimport { phoneType } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nregisterFieldType({ ...phoneType, cell: PhoneCell })\n```\n\nRegistering an id again replaces the type and keeps its place in the\npicker; in your own file, add `cell: PhoneCell` to `phoneType` instead.\nThe cell receives `FieldCellProps`: `value`, `props` (defaults merged with\n`fieldProps`), `row` (`{ id, original }`), `column` (`{ id, columnDef }`)\nand `ctrl`, the grid. `linkHref` is the helper the built-in URL, email and\nphone cells use; it returns an href only for the scheme the kind allows,\nso a value like `javascript:alert(1)` never becomes a link.\n\n## 5. Add properties to the field editor\n\n`properties` is what \"Edit field\" shows for the type. Each property is\nkept under its `key` in the column's `fieldProps`.\n\n```ts\nimport { registerFieldType, type FieldPropertyDef } from '@svgrid/enterprise/base'\nimport { phoneType } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nexport const PHONE_PROPERTIES: FieldPropertyDef[] = [\n {\n key: 'country', label: 'Home country', kind: 'choice',\n choices: [\n { value: '1', label: 'United States, Canada (+1)' },\n { value: '44', label: 'United Kingdom (+44)' },\n { value: '49', label: 'Germany (+49)' },\n { value: '359', label: 'Bulgaria (+359)' },\n ],\n hint: 'Numbers typed without a + are read as this country\\'s.',\n },\n {\n key: 'display', label: 'Show numbers of that country', kind: 'segmented',\n choices: [{ value: 'national', label: '(415) 555-0123' }, { value: 'international', label: '+1 (415) 555-0123' }],\n },\n]\n\nregisterFieldType({ ...phoneType, cell: PhoneCell, properties: PHONE_PROPERTIES })\n```\n\nThe dialog draws a dropdown with the hint under it, then a row of two\nbuttons. A property the column has not set shows the type's default.\nPicking your type in the dialog starts the draft from your `defaults` and\nkeeps only the values of properties your type also lists, so a field made\nthere saves your defaults into its `fieldProps`; changing `defaults` later\ndoes not move it. On save, empty strings, empty lists (other than\n`options`) and `undefined` are dropped. A `choice` hands back the option's\n`value` string; a `number` hands back a number, or `undefined` when\ncleared.\n\n## 6. A type with its own cell: Traffic light\n\nThe second type stores a number and draws a light. Its properties set the\nthresholds, the direction and the green color.\n\n```ts\n// traffic-light.ts\nimport { registerFieldType, formatNumber, BRIGHT, type FieldTypeDef } from '@svgrid/enterprise/base'\nimport TrafficLightCell from './TrafficLightCell.svelte'\n\nexport type LightProps = {\n green: number\n amber: number\n higherIsBetter: boolean\n goodColor: string\n show: 'light' | 'number'\n showNumber: boolean\n default?: number | string\n}\n\nexport const LIGHT_DEFAULTS: LightProps = {\n green: 80, amber: 50, higherIsBetter: true, goodColor: '#20c933', show: 'light', showNumber: true,\n}\n\nexport type Light = { band: 'green' | 'amber' | 'red'; color: string; label: string }\n\n/** The light a score gets under the field's thresholds. */\nexport function lightOf(n: number, p: LightProps): Light {\n const reaches = (limit: number) => (p.higherIsBetter ? n >= limit : n <= limit)\n if (reaches(Number(p.green))) return { band: 'green', color: p.goodColor, label: 'On track' }\n if (reaches(Number(p.amber))) return { band: 'amber', color: '#fcb400', label: 'At risk' }\n return { band: 'red', color: '#f82b60', label: 'Off track' }\n}\n\n/** A stored value as a score, or null. Old data may hold \"72\" or \"72%\" as text. */\nexport function scoreOf(value: unknown): number | null {\n if (value == null || value === '') return null\n const n = typeof value === 'number' ? value : Number(String(value).replace(/[%\\s]/g, ''))\n return Number.isFinite(n) ? Math.max(0, Math.min(100, n)) : null\n}\n\nexport const trafficLightType: FieldTypeDef<LightProps> = {\n id: 'trafficLight',\n label: 'Traffic light',\n description: 'A score from 0 to 100, shown as a green, amber or red light.',\n icon: 'M9 2h6v20H9zM12 6h.01M12 12h.01M12 18h.01',\n group: 'number',\n width: 120,\n defaults: LIGHT_DEFAULTS,\n properties: [\n { key: 'green', label: 'Green at', kind: 'number', min: 0, max: 100, hint: 'With \"Higher is better\" off, green is at or below this value.' },\n { key: 'amber', label: 'Amber at', kind: 'number', min: 0, max: 100, hint: 'Past amber the light is red.' },\n {\n key: 'higherIsBetter', label: 'Higher is better', kind: 'checkbox',\n // Flipping the direction swaps the thresholds. `p` is the draft with\n // the type's defaults filled in, as `visible` sees it.\n alsoSets: (_v, p) => ({ green: p.amber, amber: p.green }),\n },\n { key: 'goodColor', label: 'Green color', kind: 'color', palette: BRIGHT },\n { key: 'show', label: 'Show as', kind: 'segmented', choices: [{ value: 'light', label: 'Light' }, { value: 'number', label: 'Number' }] },\n { key: 'showNumber', label: 'Number next to the light', kind: 'checkbox', visible: (p) => p.show !== 'number' },\n { key: 'default', label: 'Default value', kind: 'default', hint: 'New rows start with this score.' },\n ],\n fill: (def) => ({\n ...def,\n cellDataType: def.cellDataType ?? 'number',\n editorType: def.editorType ?? 'trafficLight',\n align: def.align ?? 'left',\n formatter: def.formatter ?? ((ctx: { value: unknown }) => {\n const n = scoreOf(ctx.value)\n return n == null ? '' : formatNumber(n, { precision: 0 })\n }),\n valueParser: def.valueParser ?? ((x: { newValue: unknown; oldValue: unknown; rawInput?: unknown }) => {\n if (typeof x.rawInput !== 'string') return scoreOf(x.newValue)\n if (!x.rawInput.trim()) return null\n return scoreOf(x.rawInput) ?? x.oldValue\n }),\n }),\n cell: TrafficLightCell,\n // \"Show as: Number\" hands the cell back to the grid, which draws the formatter's text.\n hasCell: (p) => p.show !== 'number',\n newRowValue: ({ props }) => (props.default == null ? undefined : scoreOf(props.default)),\n}\n\nregisterFieldType(trafficLightType)\n```\n\n```svelte\n<!-- TrafficLightCell.svelte -->\n<script lang=\"ts\">\n import type { FieldCellProps } from '@svgrid/enterprise/base'\n import { lightOf, scoreOf, type LightProps } from './traffic-light'\n\n let { value, props }: FieldCellProps<LightProps> = $props()\n const n = $derived(scoreOf(value))\n const light = $derived(n == null ? null : lightOf(n, props))\n const number = $derived(n == null ? '' : String(Math.round(n)))\n</script>\n\n{#if light}\n <span class=\"tl\" title={`${light.label}: ${number}`}>\n <span class=\"tl-dot\" style:background={light.color} role=\"img\" aria-label={light.label}></span>\n {#if props.showNumber}<span class=\"tl-num\">{number}</span>{/if}\n </span>\n{/if}\n\n<style>\n .tl { display: inline-flex; align-items: center; gap: 6px; min-width: 0; }\n .tl-dot { flex: none; width: 10px; height: 10px; border-radius: 50%; }\n .tl-num { font-variant-numeric: tabular-nums; color: var(--sg-fg, #0f172a); }\n</style>\n```\n\n`cellDataType: 'number'` makes it a number column to the rest of the base:\nthe footer offers Sum, Average, Median, Min, Max and Range, each formatted\nby your `formatter`; the column menu labels its sorts \"1 → 9\" and\n\"9 → 1\"; the view bar filters it with =, <, > and the other number\noperators. The light has an `aria-label`, so a screen reader user gets the\nband as a word, not only a color.\n\nIn \"Edit field\" the type shows two number boxes, a checkbox, ten color\nswatches, a Light / Number switch, a second checkbox and the default box.\n\"Number next to the light\" disappears while the switch is on Number,\nbecause `visible` reads the draft. Group by this field and each score is a\ngroup of its own: the grid groups by the stored value, and the group\nheader draws its light.\n\nUntil step 7 registers an editor named `trafficLight`, the cell edits in a\nplain text box and the parser reads the text: type `72%` and 72 is stored.\n\n## 7. Give it its own editor\n\nAn editor is a Svelte component registered with `registerCellEditor` from\n`@svgrid/grid` under the name your `fill` puts in `editorType`.\n\n```svelte\n<!-- TrafficLightEditor.svelte -->\n<script lang=\"ts\">\n import { fieldPropsOf } from '@svgrid/enterprise/base'\n import { lightOf, scoreOf, type LightProps } from './traffic-light'\n\n let {\n value,\n columnDef,\n onChange,\n onCommit,\n onCommitAndMove,\n onCancel,\n }: {\n value: unknown\n columnDef?: Record<string, unknown>\n onChange: (v: unknown) => void\n onCommit: (v?: unknown) => void\n onCommitAndMove: (v: unknown, direction: 1 | -1) => void\n onCancel: () => void\n } = $props()\n\n // The column's properties: the type's defaults under its fieldProps.\n const props = $derived(fieldPropsOf(columnDef ?? {}) as LightProps)\n // svelte-ignore state_referenced_locally\n let n = $state(scoreOf(value) ?? 0)\n const light = $derived(lightOf(n, props))\n\n function focusOnMount(el: HTMLElement) {\n queueMicrotask(() => el.focus())\n }\n function slide(e: Event & { currentTarget: HTMLInputElement }) {\n n = Number(e.currentTarget.value)\n // Keep the grid's draft in step, so a click elsewhere saves what is shown.\n onChange(n)\n }\n function onKey(e: KeyboardEvent) {\n if (e.key.startsWith('Arrow') || e.key === 'Home' || e.key === 'End') {\n // The slider moves; the grid's active cell stays put.\n e.stopPropagation()\n return\n }\n if (e.key === 'Enter') onCommit(n)\n else if (e.key === 'Tab') onCommitAndMove(n, e.shiftKey ? -1 : 1)\n else if (e.key === 'Escape') onCancel()\n else return\n e.preventDefault()\n e.stopPropagation()\n }\n</script>\n\n<span class=\"tle\">\n <input type=\"range\" min=\"0\" max=\"100\" step=\"1\" value={n} aria-label=\"Score\"\n style:accent-color={light.color} use:focusOnMount oninput={slide} onkeydown={onKey} />\n <output>{n}</output>\n</span>\n\n<style>\n .tle { display: flex; align-items: center; gap: 8px; height: 100%; padding: 0 8px; background: var(--sg-bg, #fff); }\n .tle input { flex: 1; min-width: 0; }\n .tle output { width: 3ch; text-align: end; font-variant-numeric: tabular-nums; }\n</style>\n```\n\n```ts\nimport { registerCellEditor } from '@svgrid/grid'\nimport TrafficLightEditor from './TrafficLightEditor.svelte'\n\nregisterCellEditor('trafficLight', {\n component: TrafficLightEditor,\n props: (ctx) => ({\n value: ctx.value,\n columnDef: ctx.columnDef,\n onChange: ctx.onChange,\n onCommit: ctx.onCommit,\n onCommitAndMove: ctx.onCommitAndMove,\n onCancel: ctx.onCancel,\n }),\n typeToEdit: false,\n})\n```\n\nDouble-click or F2 opens the slider in the cell. Arrows move it, Enter\nsaves, Tab saves and moves to the next cell, Escape leaves the cell as it\nwas. `typeToEdit: false` keeps typed keys from opening it; a function such\nas `(c) => /^[0-9]$/.test(c)` would open it on a digit, with `typed: true`\nand the digit as `value` on the context. Without `props`, the grid passes\n`value`, `onChange`, `onCommit`, `onCancel`, `onCommitAndMove`,\n`onRequestClose` and `inCell`. The context also carries `rowId`,\n`columnId`, `columnDef`, `row`, `grid`, `typed` and `onCommitAndMoveDown`.\n\nThe expanded record and a form open the same registered editor, with\n`inCell: false`. A value from a registered editor reaches `valueParser` as\nit is, with `rawInput` set to that same value rather than to text.\n\n## 8. Values the type writes itself\n\n`newRowValue` gives each new row a value: from the row adder, a group's\n\"+\", a form, a board lane. Return `undefined` to fall back to the field's\n`default` property. `onCreate` fills every existing row once, when someone\ncreates the field in the field editor. Both get a `FieldRowContext`: the\nfield's `props`, its `field` key and the table's `rows`.\n\n```ts\nimport { registerFieldType } from '@svgrid/enterprise/base'\n\nconst ticket = (prefix: string, n: number) => `${prefix}${String(n).padStart(4, '0')}`\n/** The number in \"T-0042\", or 0. */\nconst numberIn = (v: unknown) => Number(/(\\d+)$/.exec(String(v ?? ''))?.[1] ?? 0)\n\nregisterFieldType({\n id: 'ticketId',\n label: 'Ticket ID',\n description: 'T-0001, T-0002 and on. Set automatically.',\n icon: 'M4 7h16v10H4zM9 7v10',\n group: 'computed',\n width: 100,\n defaults: { prefix: 'T-' },\n properties: [{ key: 'prefix', label: 'Prefix', kind: 'text', placeholder: 'T-' }],\n fill: (def) => ({ ...def, editable: def.editable ?? false }),\n onCreate: ({ rows, props }) => rows.map((_, i) => ticket(String(props.prefix), i + 1)),\n newRowValue: ({ rows, field, props }) =>\n ticket(String(props.prefix), rows.reduce((max, r) => Math.max(max, numberIn(r?.[field])), 0) + 1),\n copy: false,\n})\n```\n\nWhen `onCreate` returns values, `onColumnAdd` hands your app the new rows\nas `data` next to the columns, so store both. `editable: false` also keeps\nthe field out of a form. \"Duplicate record\" copies every stored field into\nthe copy. `copy: false` opts a type out: the copy gets a new value from\n`newRowValue` (the next ticket number here), or the field's `default` when\nthe type has no `newRowValue`. Leave it out and two records share a ticket\nnumber. The built-in autonumber, created and last-modified fields always\nget fresh values. The traffic light in step\n6 reads its `default` property in `newRowValue` because a `default`\nproperty on a custom type is drawn as a text box and saved as a string.\n\n## Replace a built-in type\n\nRegister a type with a built-in id before calling `enableBase()`:\n`enableBase()` registers each built-in only when no type has its id yet.\n\n```ts\nimport { enableBase, registerFieldType } from '@svgrid/enterprise/base'\nimport { phoneType, PHONE_PROPERTIES } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nregisterFieldType({ ...phoneType, id: 'phone', label: 'Phone number', cell: PhoneCell, properties: PHONE_PROPERTIES })\nenableBase()\n```\n\nEvery column with `fieldType: 'phone'` now uses yours, and it is listed\nfirst in its picker section, since the picker keeps registration order.\nSome behavior stays keyed to the id rather than to your definition: changing a\nfield to `phone` converts its values as text without your parser, the\nfilter treats it as text, and a form checks it as a phone number.\n`baseToSql` follows your `fill`'s `cellDataType`, as for any registered\ntype. Override a number type such as `rating` and the type change, the\nfilter and `baseToSql` keep treating it as a number.\n\n## Keep an old type id\n\nRename a type without breaking saved columns: register the old id with\n`editAs`, pointing at the new one.\n\n```ts\nimport { registerFieldType } from '@svgrid/enterprise/base'\nimport { phoneType } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nregisterFieldType({ ...phoneType, id: 'mobile', label: 'Mobile', cell: PhoneCell, editAs: { type: 'intlPhone', props: { display: 'international' } } })\n```\n\nColumns that say `fieldType: 'mobile'` still draw, and the type picker\nleaves `mobile` out. \"Edit field\" opens such a column as Phone\n(international) with `display: 'international'` on top of its properties.\nSaving writes `fieldType: 'intlPhone'` and converts the stored values as\nany type change does, through the `intlPhone` parser, which keeps every\nvalid E.164 number. The conversion starts from the stored type, `mobile`,\nso the editor warns before you save when it would clear cells: \"Mobile to\nPhone (international) converts the values. 2 cells will be cleared or cut\nshort.\" When\nevery value converts it says nothing, since you did not pick a new type.\nThe built-in Currency and Percent types work this way: both open as Number.\n\n## Where your type shows up\n\n| Place | What it uses |\n|---|---|\n| Grid cell | `cell` when `hasCell` allows it, else the filled column's `formatter` and `align` |\n| Header, type picker, record view and form labels, view bar panels, gallery labels | `icon`, drawn as a 24x24 stroke path |\n| Group header | `cell`, given the group's first record as `row`; else `formatter` |\n| Gallery and Kanban cards | `cell`, with `ctrl` the grid at the card's width and editing off; else `formatter` |\n| Expanded record | `cell` for a non-empty value; a click opens the editor `editorType` names (a registered one, the date picker or a text box) and saves through the grid, so `valueParser` and `rejectInvalid` apply |\n| Form view | The same cells and editors; the editor's value goes through the grid's coercion and `valueParser`, as in a cell edit. On submit the form runs `validate` and refuses the record while it fails, with or without `rejectInvalid` |\n| Footer summary | Number summaries for `cellDataType: 'number'`, date ones for `'date'` and `'dateString'`, checked, unchecked and percent checked for `'boolean'`, as a checkbox gets; counts for the rest; see [Summaries at the bottom](../fields.md#summaries-at-the-bottom) |\n| View bar filter | By the filled column's `cellDataType`: the number operators for `'number'`, the date ones for `'date'` and `'dateString'`, \"is\" checked / unchecked for `'boolean'`; otherwise the text operators (contains, does not contain, is, is not, is empty, is not empty) on the stored value |\n| Sort | The stored value; a `sortValue` in `fill` sorts by a number of your own, as a single select sorts by option order |\n| Copy | What the cell shows when your `valueParser` reads that text back to the same value. Otherwise `String(value)`: an array copies as `a,b`, an object as `[object Object]`, so keep stored values to strings, numbers and arrays of them. A paste of a copy made on the same page into the same type gets the stored value back, whatever the text says |\n| Fill handle | One source cell copies. Two or more repeat: a series is for number and date fields |\n| Paste | `valueParser` with the pasted text as `rawInput`; `rejectInvalid` does not refuse it |\n| Changing a field to your type | Each value as text through your `valueParser`, as a paste with `oldValue: null`; what it refuses or throws on is cleared and counted in the editor's warning. Without a parser, a list becomes its text and a plain value stays |\n| Changing your type to a built-in | The built-in's conversion, reading your value as text (`convertValue`) |\n| `baseToSql` | By the filled column's `cellDataType`, as the built-ins map: `double precision` for `'number'`, `boolean` for `'boolean'`, `timestamptz` for `'date'`, `date` for `'dateString'`; otherwise `text` |\n\n`convertValue(value, from, to, toProps)` and\n`countCleared(values, from, to, toProps)` are exported, so a migration\nscript can preview a type change the way the field editor's warning does.\n`toProps` is the new field's `fieldProps`, which your `fill` sees when the\nparser is made. `filterKindOf(def)` returns `'number'`, `'date'` or\n`'checkbox'` from the filled column's `cellDataType` for an id it does not\nknow, and `'text'` when `cellDataType` says nothing; formula, lookup and\nrollup fields follow their `cellDataType` the same way. `operatorsFor(kind)`\nlists the operators a kind offers.\n\n## FieldTypeDef\n\n| Member | Type | What it does |\n|---|---|---|\n| `id` | `string` | The name columns use in `fieldType` |\n| `label` | `string` | Name in the type picker; a new unnamed field is named after it |\n| `description` | `string` | One line under the type list once the type is chosen |\n| `icon` | `string` | SVG path data, 24x24 viewBox, drawn as a stroke |\n| `group` | `'basic' \\| 'number' \\| 'date' \\| 'people' \\| 'computed' \\| 'relation'` | Picker section, in this order: Fields, Numbers and time, Dates, People, Computed and automatic, Other tables; default `'basic'` |\n| `defaults` | `P` | Property values a column gets when its `fieldProps` leave them out |\n| `fill` | `(def, props: P, ctx: FieldFillContext) => Record<string, unknown>` | Turns the declared column into the column the grid sees. Optional: without it the column stays as declared and still gets its footer summary |\n| `cell` | `Component<FieldCellProps<P>>` | Draws the cell |\n| `hasCell` | `(props: P) => boolean` | Whether `cell` draws; default always |\n| `properties` | `ReadonlyArray<FieldPropertyDef>` | What the field editor lets a user set |\n| `width` | `number` | Width of a new field made in the field editor (else 150) |\n| `newRowValue` | `(ctx: FieldRowContext) => unknown` | The value a new row gets |\n| `onCreate` | `(ctx: FieldRowContext) => unknown[] \\| null` | Values for every existing row when the field is created |\n| `copy` | `false` | A duplicated record gets a new value from `newRowValue` (else the field's `default`) instead of a copy |\n| `editAs` | `{ type: string; props: Record<string, unknown> }` | Keeps an old id drawing; edited as `type` and left out of the picker |\n\n## Property kinds\n\nEvery `FieldPropertyDef` has `key`, `label` and `kind`, and may have\n`visible(props)` (shown only when it returns true), `hint` (a line under the\ncontrol, for every kind) and `alsoSets(value, props)` (other keys to change\nwith this one). `visible` and `alsoSets` get the same `props`: the draft\nwith the type's defaults filled in, before this change.\n\n| `kind` | Extra keys | Draws | Hands back |\n|---|---|---|---|\n| `number` | `min`, `max` | Number box | `number`, or `undefined` when cleared |\n| `text` | `placeholder` | Text box | `string` |\n| `checkbox` | | Checkbox with the label beside it | `boolean` |\n| `choice` | `choices`: a list, or `(ctx: FieldEditorContext) => list` | Dropdown, with \"Choose...\" while the value is not in the list | The option's `value` |\n| `segmented` | `choices` | Row of buttons | The option's `value` |\n| `color` | `palette` | Swatches; `BRIGHT` is the ten-color set the built-ins use | Hex string |\n| `icon` | `icons` (`value`, `label`, `path`) | Icon buttons, the picked one in the `color` property's color | The icon's `value` |\n| `options` | | The select options editor | Option list |\n| `default` | | A checkbox, select, person or number input for those built-in types; a text box for every other type, yours included | Depends on the type; a `string` for yours |\n| `formula` | | Formula box with field buttons, a functions list and error checking; a formula that does not compile blocks Save | `string` |\n| `columns` | | Chips for the other fields | Column ids |\n\nA `choice` whose options depend on the table passes a function, as the\nbuilt-in Lookup and Rollup do to list the table's link fields:\n\n```ts\nimport type { FieldEditorContext, FieldPropertyDef } from '@svgrid/enterprise/base'\n\nconst numberFields = (ctx: FieldEditorContext) =>\n ctx.columns.filter((c) => c.fieldType === 'number').map((c) => ({ value: String(c.id ?? c.field), label: String(c.header ?? c.field) }))\n\nexport const TARGET: FieldPropertyDef = { key: 'target', label: 'Compare with', kind: 'choice', choices: numberFields }\n```\n\n## The contexts\n\n| Type | Members | Given to |\n|---|---|---|\n| `FieldCellProps<P>` | `ctrl`, `row: { id, original }`, `column: { id, columnDef }`, `value`, `props: P` | `cell` |\n| `FieldEditorContext` | `ctrl`, `props` (draft with defaults), `columns` (every other field, as declared), `self` (the field's own key, id, saved header and draft name, which its formula may not read) | `choices` functions |\n| `FieldFillContext` | `columns` (every column, as declared, this one included) | `fill` |\n| `FieldRowContext` | `props`, `field`, `rows` | `newRowValue`, `onCreate` |\n\n## Look up registered types\n\n`getFieldType(id)` returns a registered type or `undefined`;\n`listFieldTypes()` returns them all in registration order. To offer only\nsome types in the \"+\" column, pass their ids; the picker lists them in\nthat order within each section:\n\n```svelte\n<SvBase data={rows} {columns} columnAdder={{ types: ['text', 'intlPhone', 'trafficLight', 'date'] }} fieldMenu editable />\n```\n\n## Pitfalls\n\n- **Always spread `def` in `fill`**, or the column loses its `field` and\n `fieldProps`.\n- **A formatter should read only `value`.** A grid cell passes\n `{ value, row, column, table }` with `row` the grid's row, and a pinned\n row passes `row: null`. The group header (its first record), the\n expanded record, a form, and gallery and Kanban cards pass\n `{ value, row, column }` with `row` shaped as the grid's,\n `{ id, original }`; in a form `id` is `''`. A linked record's preview,\n automation messages and alert text pass the plain record as `row`. The\n footer summary and the\n chart view's axis and tooltips pass `{ value }` alone.\n- **On a card, editing is off.** In a base's gallery and Kanban views,\n `ctrl` is the grid with the card's width as the column's (220 px on a\n Kanban card) and `editingEnabled` false: a cell that edits in place\n should check it, since the card opens its record for editing. A plain\n `<SvGrid>` gallery or board without a base passes a stand-in with only\n `getColumnWidth`. In a form, `row.original` is the record being filled in.\n- **Converting into your type is only as good as your parser.** Changing a\n text field to Phone stores `(415) 555-0123` as `+14155550123`; a multiple\n select changed to Traffic light reaches the parser as `\"Design, Backend\"`,\n which is no score, so the cell is cleared. Data loaded from elsewhere\n never goes through the parser: read stored values defensively, as\n `scoreOf` does.\n- **`onCreate` runs only when the field is created**, not when an existing\n field changes to your type.\n- **A numeric type needs `cellDataType`.** The filter, the footer\n summaries, the sort labels and `baseToSql` read it from the filled\n column. Without it, Traffic light filters as text, \"contains 7\" is what\n a user can write, and its Postgres column is `text`.\n\n## Next\n\n- [Field types](../fields.md): every built-in type and its properties.\n- [Linked records](./linked-records.md): fields that point at records of another table.\n- [Formulas](./formulas.md): computed fields written as formulas.\n- [Views](./views.md): grid, Kanban, calendar, gallery and chart views of the same records.\n- [Forms](./forms.md): a form view that adds records.\n- [Keep a base in a database](./data.md): load the tables and write every change back.\n"
5195
+ "markdown": "# Build your own field type\n\nA field type decides what a column holds and how it behaves: what the cell\nshows, which editor opens, how pasted text is read, what the footer can sum.\nThe built-in types are registered with `registerFieldType`, the same call\nyou will use, so your type gets the field editor, the views and the\nexpanded record the way they do.\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\nThis page builds two types. **Phone (international)** stores numbers as\nE.164 text (`+14155550123`), shows them the local way, dials on click and\nreads whatever people paste. **Traffic light** stores a score from 0 to 100\nand draws a green, amber or red light with its own cell and its own editor.\nThe [field types page](../fields.md#your-own-field-type) has the short\nversion.\n\n![A FieldTypeDef on the left, where each member shows up on the right: id, label, icon and group in the type picker and the header; fill's formatter in cell text, footer summaries, group headers and cards; valueParser for typed edits, paste, forms and type changes; editorType for the editor; sortValue for sort order; cellDataType for filter operators, summary choices, sort labels and the SQL column type; cell and hasCell for grid cells, group headers, cards and the record view; properties and defaults in the Edit field dialog; newRowValue for new records; onCreate for existing rows when the field is created; copy: false for a duplicated record, which gets a new value from newRowValue](/docs-media/base-field-type.svg)\n\n## 1. Decide what the field stores\n\nSettle the stored value before you write the type. Filters, sorting, copy,\nthe database and your API all see the stored value; only the cell and the\nformatter see the text on screen. A phone number stored as `+14155550123`\ndials from any app, sorts by country and compares exactly.\n\n```ts\n// phone-field.ts\nexport type PhoneProps = { country: string; display: 'national' | 'international' }\n\nconst E164 = /^\\+[1-9]\\d{6,14}$/\n\n/** Typed or pasted text as E.164 (\"+14155550123\"), or null when it is no phone number. */\nexport function toE164(text: string, country: string): string | null {\n let s = text.trim().replace(/^tel:/i, '').replace(/[\\s().\\-\\/]/g, '')\n if (s.startsWith('00')) s = '+' + s.slice(2)\n if (!s.startsWith('+')) {\n // A national number: drop the trunk prefix (1 in North America, 0 elsewhere).\n s = country === '1' ? s.replace(/^1(?=\\d{10}$)/, '') : s.replace(/^0/, '')\n // North American numbers need all ten digits: no guessing an area code.\n if (country === '1' && !/^\\d{10}$/.test(s)) return null\n s = `+${country}${s}`\n }\n return E164.test(s) ? s : null\n}\n\n/** A stored value as the field shows it. A number from another country shows as stored. */\nexport function showPhone(value: unknown, p: PhoneProps): string {\n const s = value == null ? '' : String(value)\n const home = `+${p.country}`\n if (!E164.test(s) || !s.startsWith(home)) return s\n const rest = s.slice(home.length)\n if (p.country === '1' && rest.length === 10) {\n const local = `(${rest.slice(0, 3)}) ${rest.slice(3, 6)}-${rest.slice(6)}`\n return p.display === 'national' ? local : `+1 ${local}`\n }\n const grouped = rest.replace(/\\B(?=(\\d{4})+$)/g, ' ')\n return p.display === 'national' ? `0${grouped}` : `${home} ${grouped}`\n}\n```\n\n| Typed, home country | Stored | Shown (national) |\n|---|---|---|\n| `(415) 555-0123`, +1 | `+14155550123` | `(415) 555-0123` |\n| `1 415 555 0123`, +1 | `+14155550123` | `(415) 555-0123` |\n| `020 7946 0958`, +44 | `+442079460958` | `020 7946 0958` |\n| `0044 20 7946 0958`, +1 | `+442079460958` | `+442079460958` |\n| `555-0123`, +1 | refused | |\n| `n/a` | refused | |\n\nBoth are plain functions, so you can unit-test them without a grid.\n\n## 2. Register the type\n\n```ts\n// phone-field.ts, continued\nimport { registerFieldType, type FieldTypeDef } from '@svgrid/enterprise/base'\n\nexport const phoneType: FieldTypeDef<PhoneProps> = {\n id: 'intlPhone',\n label: 'Phone (international)',\n description: 'Stored as +14155550123, shown in the local format, dialed with a click.',\n icon: 'M5 4h4l2 5-2.5 1.5a11 11 0 0 0 5 5L15 13l5 2v4a2 2 0 0 1-2 2A16 16 0 0 1 3 6a2 2 0 0 1 2-2',\n group: 'basic',\n width: 170,\n defaults: { country: '1', display: 'national' },\n fill: (def, props) => ({\n ...def,\n editorType: def.editorType ?? 'text',\n formatter: def.formatter ?? ((ctx: { value: unknown }) => showPhone(ctx.value, props)),\n // Typed and pasted text is stored as E.164; text that is no number keeps the old value.\n valueParser: def.valueParser ?? ((x: { newValue: unknown; oldValue: unknown; rawInput?: unknown }) => {\n const text = String((typeof x.rawInput === 'string' ? x.rawInput : x.newValue) ?? '').trim()\n if (!text) return null\n return toE164(text, props.country) ?? x.oldValue\n }),\n // Values that never went through the parser (loaded data, an API write) are flagged.\n validate: def.validate ?? ((ctx: { value: unknown }) =>\n ctx.value == null || ctx.value === '' || E164.test(String(ctx.value)) ? null : 'Not a +country phone number'),\n }),\n}\n\nregisterFieldType(phoneType)\n```\n\nImport the file once at app start, next to `enableBase()`, and name the type\nin a column:\n\n```ts\nimport { enableBase } from '@svgrid/enterprise/base'\nimport './phone-field'\n\nenableBase()\n\nconst columns = [\n { field: 'name', header: 'Name', fieldType: 'text' },\n { field: 'phone', header: 'Phone', fieldType: 'intlPhone', fieldProps: { country: '44' } },\n]\n```\n\n`fill` is where a type turns into grid behavior. It receives the column as\nthe app declared it, the type's `defaults` with the column's `fieldProps` on\ntop (`props`), and a context with every column. It returns the column the\ngrid sees: `editorType`, `formatter`, `valueParser`, `validate`, `align`,\n`cellDataType`, `sortValue`, `editable`. Write each key as\n`def.key ?? yours`, so a column that sets `formatter` itself keeps it. The\nbuilt-ins in `builtin-types.ts` follow that rule.\n\nWith this registered, the header shows the phone icon before \"Phone\", cells\nread `020 7946 0958`, and \"Phone (international)\" appears in the field\neditor's type picker under \"Fields\", with its description under the picker\nonce it is chosen. `width` is the width a new field of this type starts\nwith when someone adds it there.\n\n![The Add field dialog in a base, its type list scrolled to show the end of the Fields section (Phone number, Email, URL, Barcode, Button), then Numbers and time (Number, Duration, Rating) and Computed and automatic (Formula, Created time, Last modified time, Created by), with the chosen type's description and its Default value property below](/docs-media/base-field-type.png)\n\nThe picker draws a section per `group`, in a fixed order, and skips a\nsection no type uses. The built-ins fill only Fields, Numbers and time,\nComputed and automatic, and Other tables (Date and User sit under Fields),\nso \"Dates\" and \"People\" appear only once a type of yours names them. A\n`group` the list does not know puts the type under Fields. Once more than\neight types are on offer, a \"Find a type\" box sits above the list.\n\n## 3. Read what people type and paste\n\n`valueParser` gets `newValue` (after the grid's own coercion for the\neditor type), `oldValue`, `rawInput` (the text as typed or pasted), the\nrow as `data` and the `columnId`. It runs on a cell edit, a paste (one\nvalue pasted over a range included), a full-row edit, an edit in the\nexpanded record and a form field. Returning `oldValue` refuses the input:\npaste `n/a` over a number and the number stays, and the page says that\nvalue did not fit. The text editor\nopens on the stored `+442079460958`, and whatever the user leaves in it\ngoes back through the parser.\n\n`validate` runs for every rendered cell, so a bad value gets the red mark\nand the message as its tooltip even when it was loaded that way. While the\ncell's editor is open, it checks the draft before `valueParser` has read\nit, so `(415) 555-0123` shows red until Enter stores it as E.164. Add\n`rejectInvalid: def.rejectInvalid ?? true` to refuse a flagged edit\noutright; a paste is not checked against it, which is why the parser does\nthe refusing here.\n\nCheck the filled column without mounting anything:\n\n```ts\nimport { fillFieldColumn } from '@svgrid/enterprise/base'\n\nconst col = fillFieldColumn({ field: 'phone', fieldType: 'intlPhone', fieldProps: { country: '44' } })\nconst show = col.formatter as (ctx: { value: unknown }) => string\nconst parse = col.valueParser as (x: { newValue: unknown; oldValue: unknown; rawInput: string }) => unknown\n\nshow({ value: '+442079460958' }) // '020 7946 0958'\nparse({ newValue: '020 7946 0958', oldValue: null, rawInput: '020 7946 0958' }) // '+442079460958'\nparse({ newValue: 'n/a', oldValue: '+442079460958', rawInput: 'n/a' }) // '+442079460958'\n```\n\n## 4. Draw the cell\n\nWithout `cell`, the grid draws the formatter's text. A cell component can\ndraw anything; this one is a `tel:` link, so a click dials.\n\n```svelte\n<!-- PhoneCell.svelte -->\n<script lang=\"ts\">\n import { linkHref, type FieldCellProps } from '@svgrid/enterprise/base'\n import { showPhone, type PhoneProps } from './phone-field'\n\n let { value, props }: FieldCellProps<PhoneProps> = $props()\n // 'tel:+14155550123'; null unless the value is 3 to 20 digits with an optional +.\n const href = $derived(linkHref('phone', value))\n const text = $derived(showPhone(value, props))\n</script>\n\n{#if href}\n <!-- Stop pointerdown: a click on the number dials, a click beside it selects the cell. -->\n <a class=\"phone\" {href} onpointerdown={(e) => e.stopPropagation()}>{text}</a>\n{:else}\n <span class=\"phone is-bad\">{text}</span>\n{/if}\n\n<style>\n .phone { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; color: inherit; }\n a.phone { text-decoration: underline; text-decoration-color: var(--sg-muted, #94a3b8); text-underline-offset: 2px; }\n .is-bad { color: var(--sg-muted, #64748b); }\n</style>\n```\n\n```ts\nimport { registerFieldType } from '@svgrid/enterprise/base'\nimport { phoneType } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nregisterFieldType({ ...phoneType, cell: PhoneCell })\n```\n\nRegistering an id again replaces the type and keeps its place in the\npicker; in your own file, add `cell: PhoneCell` to `phoneType` instead.\nThe cell receives `FieldCellProps`: `value`, `props` (defaults merged with\n`fieldProps`), `row` (`{ id, original }`), `column` (`{ id, columnDef }`)\nand `ctrl`, the grid. `linkHref` is the helper the built-in URL, email and\nphone cells use; it returns an href only for the scheme the kind allows,\nso a value like `javascript:alert(1)` never becomes a link.\n\n## 5. Add properties to the field editor\n\n`properties` is what \"Edit field\" shows for the type. Each property is\nkept under its `key` in the column's `fieldProps`.\n\n```ts\nimport { registerFieldType, type FieldPropertyDef } from '@svgrid/enterprise/base'\nimport { phoneType } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nexport const PHONE_PROPERTIES: FieldPropertyDef[] = [\n {\n key: 'country', label: 'Home country', kind: 'choice',\n choices: [\n { value: '1', label: 'United States, Canada (+1)' },\n { value: '44', label: 'United Kingdom (+44)' },\n { value: '49', label: 'Germany (+49)' },\n { value: '359', label: 'Bulgaria (+359)' },\n ],\n hint: 'Numbers typed without a + are read as this country\\'s.',\n },\n {\n key: 'display', label: 'Show numbers of that country', kind: 'segmented',\n choices: [{ value: 'national', label: '(415) 555-0123' }, { value: 'international', label: '+1 (415) 555-0123' }],\n },\n]\n\nregisterFieldType({ ...phoneType, cell: PhoneCell, properties: PHONE_PROPERTIES })\n```\n\nThe dialog draws a dropdown with the hint under it, then a row of two\nbuttons. A property the column has not set shows the type's default.\nPicking your type in the dialog starts the draft from your `defaults` and\nkeeps only the values of properties your type also lists, so a field made\nthere saves your defaults into its `fieldProps`; changing `defaults` later\ndoes not move it. On save, empty strings, empty lists (other than\n`options`) and `undefined` are dropped. A `choice` hands back the option's\n`value` string; a `number` hands back a number, or `undefined` when\ncleared.\n\n## 6. A type with its own cell: Traffic light\n\nThe second type stores a number and draws a light. Its properties set the\nthresholds, the direction and the green color.\n\n```ts\n// traffic-light.ts\nimport { registerFieldType, formatNumber, BRIGHT, type FieldTypeDef } from '@svgrid/enterprise/base'\nimport TrafficLightCell from './TrafficLightCell.svelte'\n\nexport type LightProps = {\n green: number\n amber: number\n higherIsBetter: boolean\n goodColor: string\n show: 'light' | 'number'\n showNumber: boolean\n default?: number | string\n}\n\nexport const LIGHT_DEFAULTS: LightProps = {\n green: 80, amber: 50, higherIsBetter: true, goodColor: '#20c933', show: 'light', showNumber: true,\n}\n\nexport type Light = { band: 'green' | 'amber' | 'red'; color: string; label: string }\n\n/** The light a score gets under the field's thresholds. */\nexport function lightOf(n: number, p: LightProps): Light {\n const reaches = (limit: number) => (p.higherIsBetter ? n >= limit : n <= limit)\n if (reaches(Number(p.green))) return { band: 'green', color: p.goodColor, label: 'On track' }\n if (reaches(Number(p.amber))) return { band: 'amber', color: '#fcb400', label: 'At risk' }\n return { band: 'red', color: '#f82b60', label: 'Off track' }\n}\n\n/** A stored value as a score, or null. Old data may hold \"72\" or \"72%\" as text. */\nexport function scoreOf(value: unknown): number | null {\n if (value == null || value === '') return null\n const n = typeof value === 'number' ? value : Number(String(value).replace(/[%\\s]/g, ''))\n return Number.isFinite(n) ? Math.max(0, Math.min(100, n)) : null\n}\n\nexport const trafficLightType: FieldTypeDef<LightProps> = {\n id: 'trafficLight',\n label: 'Traffic light',\n description: 'A score from 0 to 100, shown as a green, amber or red light.',\n icon: 'M9 2h6v20H9zM12 6h.01M12 12h.01M12 18h.01',\n group: 'number',\n width: 120,\n defaults: LIGHT_DEFAULTS,\n properties: [\n { key: 'green', label: 'Green at', kind: 'number', min: 0, max: 100, hint: 'With \"Higher is better\" off, green is at or below this value.' },\n { key: 'amber', label: 'Amber at', kind: 'number', min: 0, max: 100, hint: 'Past amber the light is red.' },\n {\n key: 'higherIsBetter', label: 'Higher is better', kind: 'checkbox',\n // Flipping the direction swaps the thresholds. `p` is the draft with\n // the type's defaults filled in, as `visible` sees it.\n alsoSets: (_v, p) => ({ green: p.amber, amber: p.green }),\n },\n { key: 'goodColor', label: 'Green color', kind: 'color', palette: BRIGHT },\n { key: 'show', label: 'Show as', kind: 'segmented', choices: [{ value: 'light', label: 'Light' }, { value: 'number', label: 'Number' }] },\n { key: 'showNumber', label: 'Number next to the light', kind: 'checkbox', visible: (p) => p.show !== 'number' },\n { key: 'default', label: 'Default value', kind: 'default', hint: 'New rows start with this score.' },\n ],\n fill: (def) => ({\n ...def,\n cellDataType: def.cellDataType ?? 'number',\n editorType: def.editorType ?? 'trafficLight',\n align: def.align ?? 'left',\n formatter: def.formatter ?? ((ctx: { value: unknown }) => {\n const n = scoreOf(ctx.value)\n return n == null ? '' : formatNumber(n, { precision: 0 })\n }),\n valueParser: def.valueParser ?? ((x: { newValue: unknown; oldValue: unknown; rawInput?: unknown }) => {\n if (typeof x.rawInput !== 'string') return scoreOf(x.newValue)\n if (!x.rawInput.trim()) return null\n return scoreOf(x.rawInput) ?? x.oldValue\n }),\n }),\n cell: TrafficLightCell,\n // \"Show as: Number\" hands the cell back to the grid, which draws the formatter's text.\n hasCell: (p) => p.show !== 'number',\n newRowValue: ({ props }) => (props.default == null ? undefined : scoreOf(props.default)),\n}\n\nregisterFieldType(trafficLightType)\n```\n\n```svelte\n<!-- TrafficLightCell.svelte -->\n<script lang=\"ts\">\n import type { FieldCellProps } from '@svgrid/enterprise/base'\n import { lightOf, scoreOf, type LightProps } from './traffic-light'\n\n let { value, props }: FieldCellProps<LightProps> = $props()\n const n = $derived(scoreOf(value))\n const light = $derived(n == null ? null : lightOf(n, props))\n const number = $derived(n == null ? '' : String(Math.round(n)))\n</script>\n\n{#if light}\n <span class=\"tl\" title={`${light.label}: ${number}`}>\n <span class=\"tl-dot\" style:background={light.color} role=\"img\" aria-label={light.label}></span>\n {#if props.showNumber}<span class=\"tl-num\">{number}</span>{/if}\n </span>\n{/if}\n\n<style>\n .tl { display: inline-flex; align-items: center; gap: 6px; min-width: 0; }\n .tl-dot { flex: none; width: 10px; height: 10px; border-radius: 50%; }\n .tl-num { font-variant-numeric: tabular-nums; color: var(--sg-fg, #0f172a); }\n</style>\n```\n\n`cellDataType: 'number'` makes it a number column to the rest of the base:\nthe footer offers Sum, Average, Median, Min, Max and Range, each formatted\nby your `formatter`; the column menu labels its sorts \"1 → 9\" and\n\"9 → 1\"; the view bar filters it with =, <, > and the other number\noperators. The light has an `aria-label`, so a screen reader user gets the\nband as a word, not only a color.\n\nIn \"Edit field\" the type shows two number boxes, a checkbox, ten color\nswatches, a Light / Number switch, a second checkbox and the default box.\n\"Number next to the light\" disappears while the switch is on Number,\nbecause `visible` reads the draft. Group by this field and each score is a\ngroup of its own: the grid groups by the stored value, and the group\nheader draws its light.\n\nUntil step 7 registers an editor named `trafficLight`, the cell edits in a\nplain text box and the parser reads the text: type `72%` and 72 is stored.\n\n## 7. Give it its own editor\n\nAn editor is a Svelte component registered with `registerCellEditor` from\n`@svgrid/grid` under the name your `fill` puts in `editorType`.\n\n```svelte\n<!-- TrafficLightEditor.svelte -->\n<script lang=\"ts\">\n import { fieldPropsOf } from '@svgrid/enterprise/base'\n import { lightOf, scoreOf, type LightProps } from './traffic-light'\n\n let {\n value,\n columnDef,\n onChange,\n onCommit,\n onCommitAndMove,\n onCancel,\n }: {\n value: unknown\n columnDef?: Record<string, unknown>\n onChange: (v: unknown) => void\n onCommit: (v?: unknown) => void\n onCommitAndMove: (v: unknown, direction: 1 | -1) => void\n onCancel: () => void\n } = $props()\n\n // The column's properties: the type's defaults under its fieldProps.\n const props = $derived(fieldPropsOf(columnDef ?? {}) as LightProps)\n // svelte-ignore state_referenced_locally\n let n = $state(scoreOf(value) ?? 0)\n const light = $derived(lightOf(n, props))\n\n function focusOnMount(el: HTMLElement) {\n queueMicrotask(() => el.focus())\n }\n function slide(e: Event & { currentTarget: HTMLInputElement }) {\n n = Number(e.currentTarget.value)\n // Keep the grid's draft in step, so a click elsewhere saves what is shown.\n onChange(n)\n }\n function onKey(e: KeyboardEvent) {\n if (e.key.startsWith('Arrow') || e.key === 'Home' || e.key === 'End') {\n // The slider moves; the grid's active cell stays put.\n e.stopPropagation()\n return\n }\n if (e.key === 'Enter') onCommit(n)\n else if (e.key === 'Tab') onCommitAndMove(n, e.shiftKey ? -1 : 1)\n else if (e.key === 'Escape') onCancel()\n else return\n e.preventDefault()\n e.stopPropagation()\n }\n</script>\n\n<span class=\"tle\">\n <input type=\"range\" min=\"0\" max=\"100\" step=\"1\" value={n} aria-label=\"Score\"\n style:accent-color={light.color} use:focusOnMount oninput={slide} onkeydown={onKey} />\n <output>{n}</output>\n</span>\n\n<style>\n .tle { display: flex; align-items: center; gap: 8px; height: 100%; padding: 0 8px; background: var(--sg-bg, #fff); }\n .tle input { flex: 1; min-width: 0; }\n .tle output { width: 3ch; text-align: end; font-variant-numeric: tabular-nums; }\n</style>\n```\n\n```ts\nimport { registerCellEditor } from '@svgrid/grid'\nimport TrafficLightEditor from './TrafficLightEditor.svelte'\n\nregisterCellEditor('trafficLight', {\n component: TrafficLightEditor,\n props: (ctx) => ({\n value: ctx.value,\n columnDef: ctx.columnDef,\n onChange: ctx.onChange,\n onCommit: ctx.onCommit,\n onCommitAndMove: ctx.onCommitAndMove,\n onCancel: ctx.onCancel,\n }),\n typeToEdit: false,\n})\n```\n\nDouble-click or F2 opens the slider in the cell. Arrows move it, Enter\nsaves, Tab saves and moves to the next cell, Escape leaves the cell as it\nwas. `typeToEdit: false` keeps typed keys from opening it; a function such\nas `(c) => /^[0-9]$/.test(c)` would open it on a digit, with `typed: true`\nand the digit as `value` on the context. Without `props`, the grid passes\n`value`, `onChange`, `onCommit`, `onCancel`, `onCommitAndMove`,\n`onRequestClose` and `inCell`. The context also carries `rowId`,\n`columnId`, `columnDef`, `row`, `grid`, `typed` and `onCommitAndMoveDown`.\n\nThe expanded record and a form open the same registered editor, with\n`inCell: false`. A value from a registered editor reaches `valueParser` as\nit is, with `rawInput` set to that same value rather than to text.\n\n## 8. Values the type writes itself\n\n`newRowValue` gives each new row a value: from the row adder, a group's\n\"+\", a form, a board lane. Return `undefined` to fall back to the field's\n`default` property. `onCreate` fills every existing row once, when someone\ncreates the field in the field editor. Without `onCreate`, a field created\nwith a default fills every existing row with it: a new field holds nothing\nyet, so nothing is overwritten. A default set on a field later fills only\nnew records. Both get a `FieldRowContext`: the\nfield's `props`, its `field` key and the table's `rows`.\n\n```ts\nimport { registerFieldType } from '@svgrid/enterprise/base'\n\nconst ticket = (prefix: string, n: number) => `${prefix}${String(n).padStart(4, '0')}`\n/** The number in \"T-0042\", or 0. */\nconst numberIn = (v: unknown) => Number(/(\\d+)$/.exec(String(v ?? ''))?.[1] ?? 0)\n\nregisterFieldType({\n id: 'ticketId',\n label: 'Ticket ID',\n description: 'T-0001, T-0002 and on. Set automatically.',\n icon: 'M4 7h16v10H4zM9 7v10',\n group: 'computed',\n width: 100,\n defaults: { prefix: 'T-' },\n properties: [{ key: 'prefix', label: 'Prefix', kind: 'text', placeholder: 'T-' }],\n fill: (def) => ({ ...def, editable: def.editable ?? false }),\n onCreate: ({ rows, props }) => rows.map((_, i) => ticket(String(props.prefix), i + 1)),\n newRowValue: ({ rows, field, props }) =>\n ticket(String(props.prefix), rows.reduce((max, r) => Math.max(max, numberIn(r?.[field])), 0) + 1),\n copy: false,\n})\n```\n\nWhen `onCreate` returns values, or the new field has a default,\n`onColumnAdd` hands your app the new rows as `data` next to the columns, so\nstore both. `editable: false` also keeps\nthe field out of a form. \"Duplicate record\" copies every stored field into\nthe copy. `copy: false` opts a type out: the copy gets a new value from\n`newRowValue` (the next ticket number here), or the field's `default` when\nthe type has no `newRowValue`. Leave it out and two records share a ticket\nnumber. The built-in autonumber, created and last-modified fields always\nget fresh values. The traffic light in step\n6 reads its `default` property in `newRowValue` because a `default`\nproperty on a custom type is drawn as a text box and saved as a string.\n\n## Replace a built-in type\n\nRegister a type with a built-in id before calling `enableBase()`:\n`enableBase()` registers each built-in only when no type has its id yet.\n\n```ts\nimport { enableBase, registerFieldType } from '@svgrid/enterprise/base'\nimport { phoneType, PHONE_PROPERTIES } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nregisterFieldType({ ...phoneType, id: 'phone', label: 'Phone number', cell: PhoneCell, properties: PHONE_PROPERTIES })\nenableBase()\n```\n\nEvery column with `fieldType: 'phone'` now uses yours, and it is listed\nfirst in its picker section, since the picker keeps registration order.\nSome behavior stays keyed to the id rather than to your definition: changing a\nfield to `phone` converts its values as text without your parser, the\nfilter treats it as text, and a form checks it as a phone number.\n`baseToSql` follows your `fill`'s `cellDataType`, as for any registered\ntype. Override a number type such as `rating` and the type change, the\nfilter and `baseToSql` keep treating it as a number.\n\n## Keep an old type id\n\nRename a type without breaking saved columns: register the old id with\n`editAs`, pointing at the new one.\n\n```ts\nimport { registerFieldType } from '@svgrid/enterprise/base'\nimport { phoneType } from './phone-field'\nimport PhoneCell from './PhoneCell.svelte'\n\nregisterFieldType({ ...phoneType, id: 'mobile', label: 'Mobile', cell: PhoneCell, editAs: { type: 'intlPhone', props: { display: 'international' } } })\n```\n\nColumns that say `fieldType: 'mobile'` still draw, and the type picker\nleaves `mobile` out. \"Edit field\" opens such a column as Phone\n(international) with `display: 'international'` on top of its properties.\nSaving writes `fieldType: 'intlPhone'` and converts the stored values as\nany type change does, through the `intlPhone` parser, which keeps every\nvalid E.164 number. The conversion starts from the stored type, `mobile`,\nso the editor warns before you save when it would clear cells: \"Mobile to\nPhone (international) converts the values. 2 cells will be cleared or cut\nshort.\" When\nevery value converts it says nothing, since you did not pick a new type.\nThe built-in Currency and Percent types work this way: both open as Number.\n\n## Where your type shows up\n\n| Place | What it uses |\n|---|---|\n| Grid cell | `cell` when `hasCell` allows it, else the filled column's `formatter` and `align` |\n| Header, type picker, record view and form labels, view bar panels, gallery labels | `icon`, drawn as a 24x24 stroke path |\n| Group header | `cell`, given the group's first record as `row`; else `formatter` |\n| Gallery and Kanban cards | `cell`, with `ctrl` the grid at the card's width and editing off; else `formatter` |\n| Expanded record | `cell` for a non-empty value; a click opens the editor `editorType` names (a registered one, the date picker or a text box) and saves through the grid, so `valueParser` and `rejectInvalid` apply |\n| Form view | The same cells and editors; the editor's value goes through the grid's coercion and `valueParser`, as in a cell edit. On submit the form runs `validate` and refuses the record while it fails, with or without `rejectInvalid` |\n| Footer summary | Number summaries for `cellDataType: 'number'`, date ones for `'date'` and `'dateString'`, checked, unchecked and percent checked for `'boolean'`, as a checkbox gets; counts for the rest; see [Summaries at the bottom](../fields.md#summaries-at-the-bottom) |\n| View bar filter | By the filled column's `cellDataType`: the number operators for `'number'`, the date ones for `'date'` and `'dateString'`, \"is\" checked / unchecked for `'boolean'`; otherwise the text operators (contains, does not contain, is, is not, is empty, is not empty) on the stored value |\n| Sort | The stored value; a `sortValue` in `fill` sorts by a number of your own, as a single select sorts by option order |\n| Copy | What the cell shows when your `valueParser` reads that text back to the same value. Otherwise `String(value)`: an array copies as `a,b`, an object as `[object Object]`, so keep stored values to strings, numbers and arrays of them. A paste of a copy made on the same page into the same type gets the stored value back, whatever the text says |\n| Fill handle | One source cell copies. Two or more repeat: a series is for number and date fields |\n| Paste | `valueParser` with the pasted text as `rawInput`; `rejectInvalid` does not refuse it |\n| Changing a field to your type | Each value as text through your `valueParser`, as a paste with `oldValue: null`; what it refuses or throws on is cleared and counted in the editor's warning. Without a parser, a list becomes its text and a plain value stays |\n| Changing your type to a built-in | The built-in's conversion, reading your value as text (`convertValue`) |\n| `baseToSql` | By the filled column's `cellDataType`, as the built-ins map: `double precision` for `'number'`, `boolean` for `'boolean'`, `timestamptz` for `'date'`, `date` for `'dateString'`; otherwise `text` |\n\n`convertValue(value, from, to, toProps)` and\n`countCleared(values, from, to, toProps)` are exported, so a migration\nscript can preview a type change the way the field editor's warning does.\n`toProps` is the new field's `fieldProps`, which your `fill` sees when the\nparser is made. `filterKindOf(def)` returns `'number'`, `'date'` or\n`'checkbox'` from the filled column's `cellDataType` for an id it does not\nknow, and `'text'` when `cellDataType` says nothing; formula, lookup and\nrollup fields follow their `cellDataType` the same way. `operatorsFor(kind)`\nlists the operators a kind offers.\n\n## FieldTypeDef\n\n| Member | Type | What it does |\n|---|---|---|\n| `id` | `string` | The name columns use in `fieldType` |\n| `label` | `string` | Name in the type picker; a new unnamed field is named after it |\n| `description` | `string` | One line under the type list once the type is chosen |\n| `icon` | `string` | SVG path data, 24x24 viewBox, drawn as a stroke |\n| `group` | `'basic' \\| 'number' \\| 'date' \\| 'people' \\| 'computed' \\| 'relation'` | Picker section, in this order: Fields, Numbers and time, Dates, People, Computed and automatic, Other tables; default `'basic'` |\n| `defaults` | `P` | Property values a column gets when its `fieldProps` leave them out |\n| `fill` | `(def, props: P, ctx: FieldFillContext) => Record<string, unknown>` | Turns the declared column into the column the grid sees. Optional: without it the column stays as declared and still gets its footer summary |\n| `cell` | `Component<FieldCellProps<P>>` | Draws the cell |\n| `hasCell` | `(props: P) => boolean` | Whether `cell` draws; default always |\n| `properties` | `ReadonlyArray<FieldPropertyDef>` | What the field editor lets a user set |\n| `width` | `number` | Width of a new field made in the field editor (else 150) |\n| `newRowValue` | `(ctx: FieldRowContext) => unknown` | The value a new row gets |\n| `onCreate` | `(ctx: FieldRowContext) => unknown[] \\| null` | Values for every existing row when the field is created |\n| `copy` | `false` | A duplicated record gets a new value from `newRowValue` (else the field's `default`) instead of a copy |\n| `editAs` | `{ type: string; props: Record<string, unknown> }` | Keeps an old id drawing; edited as `type` and left out of the picker |\n\n## Property kinds\n\nEvery `FieldPropertyDef` has `key`, `label` and `kind`, and may have\n`visible(props)` (shown only when it returns true), `hint` (a line under the\ncontrol, for every kind) and `alsoSets(value, props)` (other keys to change\nwith this one). `visible` and `alsoSets` get the same `props`: the draft\nwith the type's defaults filled in, before this change.\n\n| `kind` | Extra keys | Draws | Hands back |\n|---|---|---|---|\n| `number` | `min`, `max` | Number box | `number`, or `undefined` when cleared |\n| `text` | `placeholder` | Text box | `string` |\n| `checkbox` | | Checkbox with the label beside it | `boolean` |\n| `choice` | `choices`: a list, or `(ctx: FieldEditorContext) => list` | Dropdown, with \"Choose...\" while the value is not in the list | The option's `value` |\n| `segmented` | `choices` | Row of buttons | The option's `value` |\n| `color` | `palette` | Swatches; `BRIGHT` is the ten-color set the built-ins use | Hex string |\n| `icon` | `icons` (`value`, `label`, `path`) | Icon buttons, the picked one in the `color` property's color | The icon's `value` |\n| `options` | | The select options editor | Option list |\n| `default` | | A checkbox, select, person or number input for those built-in types; a text box for every other type, yours included | Depends on the type; a `string` for yours |\n| `formula` | | Formula box with field buttons, a functions list and error checking; a formula that does not compile blocks Save | `string` |\n| `columns` | | Chips for the other fields | Column ids |\n\nA `choice` whose options depend on the table passes a function, as the\nbuilt-in Lookup and Rollup do to list the table's link fields:\n\n```ts\nimport type { FieldEditorContext, FieldPropertyDef } from '@svgrid/enterprise/base'\n\nconst numberFields = (ctx: FieldEditorContext) =>\n ctx.columns.filter((c) => c.fieldType === 'number').map((c) => ({ value: String(c.id ?? c.field), label: String(c.header ?? c.field) }))\n\nexport const TARGET: FieldPropertyDef = { key: 'target', label: 'Compare with', kind: 'choice', choices: numberFields }\n```\n\n## The contexts\n\n| Type | Members | Given to |\n|---|---|---|\n| `FieldCellProps<P>` | `ctrl`, `row: { id, original }`, `column: { id, columnDef }`, `value`, `props: P` | `cell` |\n| `FieldEditorContext` | `ctrl`, `props` (draft with defaults), `columns` (every other field, as declared), `self` (the field's own key, id, saved header and draft name, which its formula may not read) | `choices` functions |\n| `FieldFillContext` | `columns` (every column, as declared, this one included) | `fill` |\n| `FieldRowContext` | `props`, `field`, `rows` | `newRowValue`, `onCreate` |\n\n## Look up registered types\n\n`getFieldType(id)` returns a registered type or `undefined`;\n`listFieldTypes()` returns them all in registration order. To offer only\nsome types in the \"+\" column, pass their ids; the picker lists them in\nthat order within each section:\n\n```svelte\n<SvBase data={rows} {columns} columnAdder={{ types: ['text', 'intlPhone', 'trafficLight', 'date'] }} fieldMenu editable />\n```\n\n## Pitfalls\n\n- **Always spread `def` in `fill`**, or the column loses its `field` and\n `fieldProps`.\n- **A formatter should read only `value`.** A grid cell passes\n `{ value, row, column, table }` with `row` the grid's row, and a pinned\n row passes `row: null`. The group header (its first record), the\n expanded record, a form, and gallery and Kanban cards pass\n `{ value, row, column }` with `row` shaped as the grid's,\n `{ id, original }`; in a form `id` is `''`. A linked record's preview,\n automation messages and alert text pass the plain record as `row`. The\n footer summary and the\n chart view's axis and tooltips pass `{ value }` alone.\n- **On a card, editing is off.** In a base's gallery and Kanban views,\n `ctrl` is the grid with the card's width as the column's (220 px on a\n Kanban card) and `editingEnabled` false: a cell that edits in place\n should check it, since the card opens its record for editing. A plain\n `<SvGrid>` gallery or board without a base passes a stand-in with only\n `getColumnWidth`. In a form, `row.original` is the record being filled in.\n- **Converting into your type is only as good as your parser.** Changing a\n text field to Phone stores `(415) 555-0123` as `+14155550123`; a multiple\n select changed to Traffic light reaches the parser as `\"Design, Backend\"`,\n which is no score, so the cell is cleared. Data loaded from elsewhere\n never goes through the parser: read stored values defensively, as\n `scoreOf` does.\n- **`onCreate` runs only when the field is created**, not when an existing\n field changes to your type.\n- **A numeric type needs `cellDataType`.** The filter, the footer\n summaries, the sort labels and `baseToSql` read it from the filled\n column. Without it, Traffic light filters as text, \"contains 7\" is what\n a user can write, and its Postgres column is `text`.\n\n## Next\n\n- [Field types](../fields.md): every built-in type and its properties.\n- [Linked records](./linked-records.md): fields that point at records of another table.\n- [Formulas](./formulas.md): computed fields written as formulas.\n- [Views](./views.md): grid, Kanban, calendar, gallery and chart views of the same records.\n- [Forms](./forms.md): a form view that adds records.\n- [Keep a base in a database](./data.md): load the tables and write every change back.\n"
5196
5196
  },
5197
5197
  {
5198
5198
  "slug": "help/base/data",
5199
5199
  "path": "docs/help/base/data.md",
5200
5200
  "title": "Keep a base in a database",
5201
5201
  "section": "SvGrid Base",
5202
- "markdown": "# Keep a base in a database\n\nA base can keep its records in your database instead of the page. Give a\ntable a `source` and `<SvBase>` loads the records from it when it opens, then\nwrites every change back: a cell edit, a pasted range, a new record, a\ndeleted one, a link made from either side. Postgres, MySQL, SQLite, Supabase\nand any REST API work the same way, because each is reached through the\ndata source contract the grid already uses.\n\n<div data-docs-demo=\"516-base-postgres\" data-height=\"760\"></div>\n\nThe demo above is SvGrid Base on a real Postgres database running in the\npage (PGlite). Change a status and the log shows the `UPDATE` it ran.\n\n## How it fits together\n\n![SvGrid Base keeps every record in the page. The writer compares the tables with what the database holds and sends inserts, updates and deletes through baseSource to a data source, which talks to Postgres, Supabase or your API.](/docs-media/base-data-flow.svg)\n\n- **The page holds every record of a table.** Views, filters, groups,\n summaries, colors and automations all work on those records, so they stay\n instant. A table with more records than the page should hold (10,000 by\n default) is read from the database a part at a time instead, by itself: see\n [A table of millions](#a-table-of-millions).\n- **The writer sends what changed.** After each change it compares the\n tables with what the database holds and queues an `INSERT` for a new\n record, an `UPDATE` with only the fields that changed, or a `DELETE`. The\n writes go one at a time, in order, so an update never overtakes the insert\n of its own record.\n- **A refused write goes back.** When the database says no (a constraint, a\n row level security policy, a network error), the cell returns to what the\n database holds, the end of the tab bar says why in plain words, and a notice\n names the record and the field. What the database said is in the tooltip, or\n through an endpoint in the server log, because it names tables and values:\n [what a person reads](./postgres.md#3-one-endpoint-per-table).\n- **Changes made elsewhere come in.** With `realtime`, the base reads a table\n again when someone else changed it and takes what moved, without writing\n anything back.\n\n## Give a table a source\n\n`baseSource` wraps any data source:\n\n```ts\nimport { baseSource, type BaseTable } from '@svgrid/enterprise/base'\nimport { createRestDataSource } from '@svgrid/enterprise'\n\nconst tasks: BaseTable = {\n id: 'tasks',\n name: 'Tasks',\n primaryField: 'name',\n data: [],\n columns: [\n { field: 'name', header: 'Task', fieldType: 'text' },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: [{ value: 'Todo' }, { value: 'Done' }] } },\n ],\n source: baseSource(createRestDataSource({ url: '/api/tasks' })),\n}\n```\n\n```svelte\n<script lang=\"ts\">\n import { SvToaster } from '@svgrid/grid'\n import { SvBase, enableBase, type BaseTable } from '@svgrid/enterprise/base'\n\n enableBase()\n let { tables }: { tables: BaseTable[] } = $props()\n</script>\n\n<SvBase bind:tables editable />\n<SvToaster />\n```\n\n`data` can start empty: the base fills it. A write that is refused shows a notice\nand the tab bar says so; see [Notices](./quick-start.md#notices).\n\nThe data sources that ship with `@svgrid/enterprise`:\n\n| Source | Talks to | Tutorial |\n|---|---|---|\n| `createSupabaseDataSource` | Supabase (PostgREST), from the browser | [Build a base on Supabase](./supabase.md) |\n| `createSqlDataSource` | Any SQL database, through an `execute(sql, params)` you write. It reads and writes the columns on its schema and no others; `wholeRecords: true` reads and writes every column, for a source in the same process as the page, and `createKitHandlers` refuses one | [Build a base on Postgres](./postgres.md) |\n| `createKitDataSource` | Your SvelteKit endpoint, which runs `createSqlDataSource` on the server | [Build a base on Postgres](./postgres.md) |\n| `createRestDataSource` | A JSON API: `GET`, `POST`, `PATCH`, `DELETE` | [Build a base on your API](./rest-api.md) |\n| `createInMemoryDataSource` | An array, for tests and prototypes | |\n\nYour own source works too: an object with `getRows`, `createRow`,\n`updateRow` and `deleteRow` (the grid's `ServerDataSource`).\n\n## Link fields\n\n![How link fields are stored. A link to one record is a foreign key column. A link to many records is a junction table with one row per pair. The field that links back is not stored.](/docs-media/base-links.svg)\n\nIn the page a link field holds record ids. `links` says how each one is kept:\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport { createRestDataSource } from '@svgrid/enterprise'\n\nconst taskTags = createRestDataSource({ url: '/api/task_tags' })\n\nconst source = baseSource(createRestDataSource({ url: '/api/tasks' }), {\n links: {\n // A link to one record: the column with the linked record's id.\n project: { column: 'project_id' },\n // A link to many records: a junction table, one row per pair.\n tags: { through: taskTags, from: 'task_id', to: 'tag_id' },\n },\n})\n```\n\nThe field that links back (a project's Tasks, when tasks have a Project\nfield) is worked out in the page from the other side and never stored. Link\na task from the project's side and the base writes that task's\n`project_id`. A link field that is neither in `links` nor the other side of\none is written as it is, a list of ids, which suits a Postgres `text[]` or a\n`jsonb` column.\n\nA list (a multiple select, a vote, a sparkline) goes to its column as that column takes\nit: an array to `text[]`, JSON to `json` and `jsonb`, and JSON text to a text column, on\nMySQL, SQL Server and SQLite too. It reads back as the list it was.\n\n## Keys the base can use\n\nA record needs one column that names it. When a base reads a database:\n\n- A table with a primary key of one column is a table.\n- A table keyed by both ends, `order_tags (order_id, tag_id, primary key (order_id, tag_id))`,\n with both columns foreign keys, is a link to many, like a junction with its own `id`. A\n link goes by the whole key, so removing a tag deletes that one pair. A tenant column in\n the key, named in `tenantColumn`, is part of the key and no end of the link. Other columns\n on such a table are not read and not written: a new link gives them their defaults, one\n that is `NOT NULL` with no default refuses the link, and the server log names them. A\n pair whose two ends the base does not hold, by their keys, is left out.\n- Any other table with a composite key, or with no primary key, is left out of the base, a\n view included. The rest of the base opens, and the log names the table and says why.\n Nothing runs on part of a key, because a delete by one column of two reaches every row\n that shares it.\n\n`keys` brings such a table in, when one of its columns is unique:\n\n```ts\nexport const { GET, POST } = createBaseEndpoint({ execute, allTables: true, public: true, keys: { v_orders: 'id' } })\n```\n\nThat is your word for it: nothing checks, and a delete reaches every row that holds the\nvalue. `baseFromDatabase` and `baseFromSupabase` take it too. On Supabase the key comes\nfrom the API description; a project that does not show that to the key in use has only a\ncolumn named `id` to go by.\n\n### Foreign keys of more than one column\n\nA foreign key of more than one column, as a database keeps tenants apart with,\n`foreign key (tenant, customer_id) references customers (tenant, id)`, is a link through\n`customer_id` to that tenant's customer, when `tenantColumn` names `tenant` on both tables.\nThe tenant column stays the read-only tenant, not a link. It works the same in a pair\ntable, `(tenant, order_id) -> orders (tenant, id)`. Any other key of more than one column,\nthree columns or two without the tenant, links nothing: its columns are plain values and\nthe log names the key. A key links only to a table its bare name reaches: on Postgres a\ntable of any schema on the search path, unless a table of that name in an earlier schema\nhides it (a key to a hidden table stays a plain value); on SQL Server the user's default\nschema; on MySQL the connected database. The other table still has to be one the base\nserves. Supabase does not describe a key of more than one column, so there\nneither column is a link.\n\n## Record ids\n\nA new record gets an id in the page at once, so it can be edited and linked\nbefore the database answers. By default that id is its key in the database\ntoo: give the key column a text type.\n\nWhen the database makes the key (a `serial` column, a `uuid` with a default),\nset `serverIds: true`. The record is inserted without the page's id, and the\nwriter keeps the page id for the grid and translates to the database id for\nevery later write and every link that points at the record.\n\n## What a write answers\n\nThe row a server answers is the truth. After a write the cell, the record and the views\nshow what was stored: a value a `before*` hook or a trigger changed, a tenant the scope\nstamped, a default on a new record. A number the column rounded shows rounded. Undo goes\nback from what was stored. A block of Edit fields or of a range on a big table answers with\nwhat hooks changed, and a trigger's work on such a block shows when the view reads the rows\nagain. On a table the page holds, a paste, a clear or an Undo of 20 records or more is\nwritten in blocks, and each block's records are read back once it lands, so a trigger's\nwork shows at once there too.\n\nThe answer is the row the write left, not a teammate's write a moment later. Postgres and\nSQLite say it in the write itself (`RETURNING`). SQL Server writes and reads it back in one\ntransaction of one batch. MySQL has no way to do either in one statement: give\n`createSqlDataSource` (or `baseFromDatabase`, `createBaseEndpoint`) its `transaction` and the\nwrite and its read go in one. Without it a teammate's write to the same cell in the moment\nbetween the two is taken for yours, and an Undo of yours can then write over theirs.\n\n## What is not written\n\nFields the page works out are never sent: formula, lookup, rollup, count and\nbutton fields, and the field that links back. Add any other field to `skip`\nto keep it in the page only.\n\n## What a cell holds\n\nA base reads dates, times, bigints and decimals as the database writes them, as\ntext. A SQL source asks for a text copy of each such column\n(`to_json(col) #>> '{}'` on Postgres, `CAST(col AS CHAR)` on MySQL) beside the\ncolumn itself, so sorting and filtering still use the real column.\n\n| Column | What the page holds | What people see |\n|---|---|---|\n| `date` | `2026-03-08` | That day, in any zone |\n| `timestamp`, `datetime` | `2026-03-08T02:30:00.123456` | That wall clock, in any zone |\n| MySQL `timestamp` | The wall clock of the connection's `time_zone` | That wall clock, in any zone |\n| `timestamptz` | `2026-03-14T22:00:00.654321+00:00` | That moment, in the viewer's zone |\n| `bigint`, `numeric`, `decimal` | Text past what a JS number holds | Every digit |\n| `json`, `jsonb` | The text the database writes | Every digit: a driver's `JSON.parse` would round a number past 2^53 |\n\nA time someone types keeps its wall clock on a column without a zone, and is that\nmoment on a `timestamptz`. An undo or a restore writes back the text it read, so\nnothing moves by a zone or loses digits past the millisecond.\n\nThe days of a filter and the months of a chart are the viewer's. On a column of moments\n(Postgres `timestamptz`, SQL Server `datetimeoffset`, SQLite text that ends in a zone)\n\"is 8 March\", \"today\" and \"within the past week\" keep the records whose moment falls on\nthose days where the person is, and a chart counts each record in the month it reads as\nthere, across a change of the clocks too. The page sends each date condition with the\nmoments its days start and end (`moments` on the condition), since a server has no zone\nof the viewer's, and `createSqlDataSource` compares them on such a column. A date and a\nwall clock, a MySQL `TIMESTAMP` among them, compare their days and read the same everywhere.\nOn Supabase, say which columns hold moments: `supabaseFieldFilter(columns, { moments: (def) => def.field === 'starts_at' })`.\nWithout that, a condition there compares the database's days.\n\nOver PostgREST (Supabase) a date already arrives as text. A JSON number cannot\ncarry more than about 15 digits, so the first read that shows a longer one reads\nthat column again with `col::text`, and later reads bring the copy along. A change\nthat arrives through Realtime is plain JSON: such a number in it is rounded until\nthe next read.\n\nJSON in a `text` or `longText` field shows as JSON text. A change goes back as\nJSON; text that is not valid JSON is refused with a message and the cell goes back. A\n`jsonb` column writes its own spaces and key order, and that is no change: nobody is told\nof one. Over PostgREST a JSON column arrives as JSON, so a number past 2^53 in it is rounded.\n\nOpening a cell and leaving it writes nothing, whatever the field shows.\n\n## Changes made elsewhere\n\n![An edit, step by step: Ana's base sends the field that changed, the database tells every subscriber, and Ben's base reads the table again and takes what moved.](/docs-media/base-realtime.svg)\n\n`realtime` takes a function that calls `apply` with what changed and returns a\nfunction that stops listening. With Supabase, hand it Realtime's events as\nthey come:\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport {\n createSupabaseDataSource,\n createSupabaseRealtime,\n type EntitySchema,\n type SupabaseClientLike,\n type SupabaseRealtimeClientLike,\n} from '@svgrid/enterprise'\n\n// Your supabase-js client: createClient(url, anonKey).\ndeclare const client: SupabaseClientLike & SupabaseRealtimeClientLike\ndeclare const schema: EntitySchema\n\nconst source = baseSource(createSupabaseDataSource({ client, table: 'tasks', schema }), {\n realtime: (apply) => createSupabaseRealtime({ client, table: 'tasks', onChange: apply }),\n})\n```\n\nThe base updates just the records a change names: the other records keep\ntheir objects, so the grid redraws one row, not 50,000, and the table is not\nread again. A burst of changes (an agent updating 200 records) goes into the\npage as one update.\n\nAny other feed works the same way: a WebSocket message, a server-sent event,\na poll. Call `apply` with the changes, as the database has the records\n(its column names, its ids):\n\n```ts\nimport { baseSource, type BaseChangeInput } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\ndeclare const tasksSource: ServerDataSource<Record<string, unknown>>\n\nconst source = baseSource(tasksSource, {\n realtime: (apply: (changes?: BaseChangeInput) => void) => {\n const socket = new WebSocket('wss://example.com/tasks')\n socket.onmessage = (m) => {\n const { type, id, row, changed } = JSON.parse(m.data)\n if (type === 'update' && changed) apply([{ type: 'update', id, fields: changed }])\n else if (type === 'delete') apply([{ type: 'delete', id }])\n else apply([{ type, record: row }])\n }\n return () => socket.close()\n },\n})\n```\n\n| A change | Carries |\n|---|---|\n| `{ type: 'insert', record }` | The new record, with its key column. |\n| `{ type: 'update', id, record }` | The whole record as it is now. |\n| `{ type: 'update', id, fields }` | Just the columns that changed. |\n| `{ type: 'delete', id }` | The key of the record that went. |\n\nA change that comes back from the page's own write (the database tells every\nlistener, the page that wrote it included) moves nothing and is not reported.\nAn update with `fields` for a record the page does not have, and `apply()`\nwith nothing, read the whole table again. An update with `id` alone reads that one record.\n\n### A feed on an endpoint\n\n`baseFromEndpoint` takes one `feed` for the whole base. It opens when the first table\nlistens and closes with the last. Each message names its table, a junction table\nincluded, and a change in a junction reaches the record it starts from:\n\n```ts\nimport { baseFromEndpoint } from '@svgrid/enterprise/base'\n\nconst tables = await baseFromEndpoint('/api/base', {\n feed: (apply) => {\n const events = new EventSource('/api/base/feed')\n events.onmessage = (e) => {\n // { \"table\": \"orders\", \"changes\": [{ \"type\": \"update\", \"id\": \"o1\", \"fields\": { \"status\": \"Paid\" } }] }\n const { table, changes } = JSON.parse(e.data)\n apply(table, changes)\n }\n // After a reconnect, changes may have been missed: every table reads again.\n let opened = false\n events.onopen = () => {\n if (opened) apply()\n opened = true\n }\n return () => events.close()\n },\n})\n```\n\n`apply(table)` with no changes reads that table again, and `apply()` reads every table again.\n\nKeep it to one connection. A browser opens at most six connections to one host over\nHTTP/1.1, which is what a Vite or Node dev server speaks, and a connection held open by a\nfeed is one of them. With a feed per table on a base of six tables and junctions, no write\nreaches the server. `realtime: (table, apply) => stop`, called per table, is still taken\nfor a feed that shares one connection itself, such as channels on one socket.\n\nThe feed is yours to write and yours to scope: with a tenant `scope`, send each person their\ntenant's rows only.\n\n### What a feed has to cover\n\nA feed is per table, and some of what a person changes is not in that table:\n\n- **A link to many lives in its junction table.** Adding a tag writes a row to `order_tags`\n and nothing to `orders`, so a feed of `orders` alone never tells teammates. Listen to the\n junction too, on the table that stores the link, and name the record it starts from:\n\n ```ts\n import { baseSource } from '@svgrid/enterprise/base'\n import type { ServerDataSource } from '@svgrid/grid'\n\n declare const ordersSource: ServerDataSource<Record<string, unknown>>\n declare const orderTags: ServerDataSource<Record<string, unknown>>\n // Your feed: calls back with what changed in that table, and returns how to stop.\n declare function listen(table: string, on: (change: any) => void): () => void\n\n const source = baseSource(ordersSource, {\n links: { tags: { through: orderTags, from: 'order_id', to: 'tag_id' } },\n realtime: (apply) => {\n const orders = listen('orders', (change) => apply(change))\n const pairs = listen('order_tags', (row) => apply(row?.order_id == null ? undefined : { type: 'update', id: row.order_id }))\n return () => { orders(); pairs() }\n },\n })\n ```\n\n A feed that only says \"something changed\" calls `apply()` for both. `baseFromSupabase`\n and `baseFromEndpoint` listen to the junction tables by themselves, and needs them in the `supabase_realtime`\n publication. A delete's event there has only the key unless the junction table has\n `REPLICA IDENTITY FULL`, so a removed link reads the table again.\n- **A soft delete reaches a feed as an update.** The base reads an update that sets the mark\n as a delete, so send the column. Through `createKitDataSource` the page learns the column\n from the server's first answer. A restore reads as an update of a record the page does not\n have. A page-held table takes it in, and reads the table if it has links through a junction.\n A table read a part at a time reads the rows on screen, at most once a second, because a\n feed cannot tell a restore from an edit of a record scrolled out of view.\n- **A feed that goes down says nothing.** Nothing here can tell silence from no changes.\n\n### What people see\n\nA change made elsewhere does not just appear. Until someone opens the record:\n\n- in a grid view, the cells that changed are tinted, and a record added\n elsewhere is tinted across the row (Kanban and gallery cards are not);\n- the row number has a dot; hover it for \"Ana Ruiz changed this record 2\n minutes ago\";\n- the view bar counts them, \"3 changed by Ana Ruiz\", and a click on the count\n marks them all seen.\n\nOpening the record clears its marks. Its activity lists each change as\n\"Ana Ruiz changed Priority\", with the value before crossed out and the value\nnow: Low → Urgent.\n\nThe name comes from the record itself: its `lastModifiedBy` field (for a\nrecord added elsewhere, `createdBy` first). Whoever writes to the database\nwrites their user id to that column, and the base looks it up in\n`setFieldUsers`. Without such a field the change is credited to \"Someone\nelse\". Your own writes are never reported back to you: the base compares what\nit reads with what it last wrote.\n\n### An AI agent writing to the base\n\nThe same path serves an agent: a script, a worker calling a model, or Claude\nCode with your database's MCP server. Give the agent a user id, add it to\n`setFieldUsers` with a name and an avatar, and have it set `modified_by` (or\nwhichever column your `lastModifiedBy` field reads) on every row it writes.\nPeople then see exactly what it changed, review it in a view filtered on a\n\"Needs review\" field, correct what it got wrong, and approve the rest with a\nbutton field and an automation.\n\n[Let an AI agent work in your base](./ai-agents.md) walks through it with\nClaude Code and Supabase, and [Review what an agent changed](./review-queue.md)\ncovers the review side on any database. Here the agent is a stand-in made of\nkeyword rules, so the demo runs without a model or a key. It writes to the database with SQL; press \"Run the triage\nagent\", then open a ticket.\n\n<div data-docs-demo=\"700-base-ai-triage\" data-height=\"900\"></div>\n\nTo act on changes made elsewhere in code (a notice, a log, a count of your\nown), pass `onRemoteChange` to `<SvBase>`. It gets each change once it is in\nthe page:\n\n```ts\nimport type { BaseRemoteChange } from '@svgrid/enterprise/base'\n\nfunction onRemoteChange(changes: BaseRemoteChange[]) {\n for (const c of changes) {\n // c.type: 'insert' | 'update' | 'delete'; c.table, c.id, c.record\n // c.fields (updates): [{ field: 'priority', from: 'Low', to: 'Urgent' }]\n if (c.type === 'update') console.log(c.id, c.fields)\n }\n}\n```\n\n## Status and errors\n\nThe end of the tab bar says what the base is doing:\n\n| It reads | When |\n|---|---|\n| Loading records... | The tables are loading |\n| Saving... | Writes are on their way |\n| Still saving: the server has not answered yet | A write has been waiting a while |\n| All changes saved | Every write went through |\n| Not saved: *reason* | A write was refused; the cell went back |\n\n![A refused write: the database rejects a status, the cell goes back, and the tab bar says the database does not accept this value.](/docs-media/base-refused-write.png)\n\nA table that cannot be read says so where its records would be, for as long as it lasts,\nwith why and a **Try again**. No empty table is shown for it, and nothing can be added to\nit. The other tables of the base open as usual.\n\nA table read a part at a time (`serverRows`) whose next part cannot be read says so in the\ntab bar, \"Could not load *table*: *reason*\", with **Try again**, for as long as a part of the\nview is unread, wherever it is scrolled to. The rows that did not come are a tinted band with\n\"Could not load these rows.\" and **Retry** on its first row on screen. A read again of rows\nalready on screen, after a teammate's change, that fails leaves those rows as they were: the\ntab bar says so with **Try again**, which reads them again.\n\nWhen the server sent a reason with its error (`{ \"error\": \"...\" }` and no `code`), that reason\nis what the notice says. `createRestDataSource` passes on an API's `{ \"error\" }` or\n`{ \"message\" }` the same way.\n\nA failed write is told in a notice that names the record and the field, says why in\nplain words and shows what was entered, with **Try again** where that can work. The record\nis named as it is stored, never by the value that was refused. A foreign key of more than one\nfield that refuses a write names all of its fields. The notice stays until it is dismissed,\nand \"Not saved\" stays until that record is written: a later write of another record does not\nclear it. A click, a tap or Enter on the status line opens it in full: the reason, and the\nserver's own words when it sent some and the reason does not already say them. The text of\nthe request itself (a status code, a JSON body) is never shown; it is in the browser's\nnetwork log and the server's log.\n\nA new record the database refused takes no edit while it is still listed (while another cell or\nrecord is being edited, or a draft on a table read a part at a time whose refusal no retry can\nchange, such as an id column too narrow): a click, a key or a paste on\nit shows its notice again instead of an editor. An editor open on it when the refusal comes closes\nwith what it holds, and the notice names it. A draft refused for a reason a retry may get past keeps\ntaking edits, and its next edit sends it again.\n\n**Try again** reads the record first, as an undo does. A value that changed since the\nwrite failed, whether a teammate wrote it or you typed it again, keeps that value and the\npage says so. A record deleted since is reported with what you entered. If the record\ncannot be read, nothing is written and the notice offers Try again once more.\n\nA change of many records at once (a paste, a clear or an Undo of 20 or more) that fails\nis one notice with the number of records, and the same **Try again**: it reads them all\nfirst, sends what is still as it was in blocks, and says how many cells changed since and\nhow many records were deleted since. Part of a block may have landed; the page reads the\ntable again to show which. An Undo of some of those cells takes them out of what Try\nagain would send. A refusal of your permission offers no Try again and leaves no step to\nundo, and so does a refusal of the values, an answer of `400`, `409` or `422`: the same\nvalues would be refused again.\n\nA write that gets no answer is not given up on: it may still land, or be refused. After\n`writeTimeout` (60 seconds by default, on `<SvBase>`) writes to other records go on, and\nlater writes to the same record wait behind it, so none can overtake it or be overwritten\nby it landing late. The status says \"Still saving\" for as long as it has no answer, and\nstarts warning at a sixth of `writeTimeout`. The page keeps what was entered, and goes back\nonly if the write is refused. A delete with no answer keeps its copy in the trash until the\noutcome is known, and leaving the page asks first.\n\nWhile someone is typing in a cell or in a field of a record, changes from others wait\nuntil the editor closes, so no row moves under it and no text changes under the cursor.\nThat holds for a live feed and for a read of the table, on any table. If the record was\ndeleted meanwhile, saving it says so and shows the text.\n\n## Options\n\n`baseSource(dataSource, options)`:\n\n| Option | Default | What it does |\n|---|---|---|\n| `links` | none | How link fields are stored: `{ column }` or `{ through, from, to }` |\n| `columns` | none | Column names that differ from field names: `{ dueDate: 'due_date' }` |\n| `idField` | `'id'` | The key column |\n| `serverIds` | `false` | The database makes the key of a new record |\n| `realtime` | none | Hear of changes made elsewhere: `(apply) => stop`; `apply(changes)` updates those records in place |\n| `skip` | none | Fields kept in the page only |\n| `holdInPage` | `10000` | How many records a table may hold in the page. It is where a table changes how it is read, not a cap: a bigger one is read from the database a part at a time, and no record is left out. (`limit` is the old name.) |\n| `convertField` | none | A field of a big table changed type: rewrite its values in the database and resolve with `true`. Without it the base rewrites them a block at a time. |\n| `duplicate` | none | Copy a big table in the database and resolve with the new table's source. |\n| `serverRows` | automatic | `true` always reads from the database, and the first screen comes sooner on a table you know is big; `{ blockSize }` sets the rows per request (200). `false` always holds every record in the page, however many. Left out, the base decides by `holdInPage`. |\n\n## Build a base from your database\n\n`baseTablesFromSchemas` makes the tables from your database's own\ndescription of them: a number column becomes a number field, an enum a\nsingle select with its values, a foreign key a link with a field in the\nother table that links back. Read the descriptions with\n`introspectSupabaseTable`, `introspectDatabase` or `introspectOpenApi`, or\nwrite them out:\n\n```ts\nimport { baseTablesFromSchemas } from '@svgrid/enterprise/base'\nimport { createRestDataSource, type EntitySchema } from '@svgrid/enterprise'\n\nconst schemas: EntitySchema[] = [\n {\n name: 'deals',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'title', type: 'text' },\n { field: 'stage', type: 'enum', options: [{ value: 'Open', label: 'Open' }, { value: 'Won', label: 'Won' }] },\n { field: 'company_id', type: 'relation', relation: { entity: 'companies', labelField: 'name' } },\n ],\n },\n { name: 'companies', fields: [{ field: 'id', type: 'text', primaryKey: true }, { field: 'name', type: 'text' }] },\n]\n\nconst tables = baseTablesFromSchemas(schemas, (schema) => createRestDataSource({ url: `/api/${schema.name}`, schema }))\n```\n\nThe result is ordinary `BaseTable` objects: change a field's type, add views,\nrename a table. [Build a base on your API](./rest-api.md) does this end to end.\n\nThe other way round, `baseToSql(tables)` writes a base as Postgres SQL:\ntables, foreign keys, the tables behind links to many, and the records, with\n`svgrid:` comments for the field settings, views and automations that SQL\nhas no type for. The tables it makes have text keys with a\n`gen_random_uuid()` default, so a record added later gets its key from the\ndatabase. See [Take a base to a database](./quick-start.md#take-a-base-to-a-database).\n\n## A table of millions\n\nA table has no size limit, and nothing here is a setting you have to get right.\nA table of the thousands of records a team keeps in a tracker, a CRM or an\ninventory is held in the page. One that grows past `holdInPage` is read from the\ndatabase a part at a time, the next time it is read, and looks and works the\nsame: every view, link, lookup, rollup, automation and search reads all of it.\nThe page then holds none of its records, only what is on screen.\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\ndeclare const orders: ServerDataSource<Record<string, unknown>>\n\n// Nothing to set: it reads from the database once the table outgrows the page.\nconst source = baseSource(orders)\n\n// A table you know is big: skip the first look, and size the requests.\nconst big = baseSource(orders, { serverRows: { blockSize: 200 } })\n```\n\n<div data-docs-demo=\"723-base-million-orders\" data-height=\"620\"></div>\n\nHow each part of a base reads such a table:\n\n| | Read a part at a time |\n|---|---|\n| Grid views | Read through the server row model. Each request carries the view's filter (`filterModel.expression`), its sort, its groups (one row per key with its count, the records under an opened key) and the search. The count in the view bar is the server's. A grouped view shows select groups in option order and each group as the field draws it, with the records it holds at every level, and the totals of the fields with a sum, average, lowest or highest picked in the footer. Whenever the view is read again (your delete, Restore, Undo or Redo, a new record, an edit once you leave the record, an option renamed, a field converted, a teammate's change), the group rows and the loaded records of every open group are read, with or without a feed: a deleted record leaves its group, a restored or new one shows in its group, and the counts follow. A group closed at the time reads again when it is opened. A teammate's change that a feed sends with its row patches that row where it is; when it changes a field the view groups or sorts by (any field, when the view groups or sorts by a formula or lookup), the view is read again at once, so the record takes its place in the order and in its group. A filtered view reads again within a second after any change, so a record that left the filter goes. |\n| Kanban | Each lane shows its true size, from one grouped request, and reads its cards as they scroll into sight. A card moved to another lane is written by id and sits at the top of that lane until the lane is read again. Sorted by a number field, dropping a card elsewhere in its lane writes that number, so the order is the table's. |\n| Calendar | Reads the dates on screen and no others. A range of up to 400 records is read whole. A busier one is read a day at a time: each day's first records and its size, from one grouped request where the backend groups and the view has no end date, so an empty day costs nothing and a write reads only the days whose size moved. \"+N more\" lists every record of that day. The days are the viewer's: on a column of moments (Postgres `timestamptz`, SQL Server `datetimeoffset`) the sizes, each day's records and that list follow the page's zone, also across a change of the clocks. A date and a wall clock (`timestamp`, `datetime`, a MySQL `TIMESTAMP`, which reads as the server's clock) show the same everywhere, and their days follow what the cell shows. |\n| Gantt | Rows by start date, read as they scroll into sight. The axis runs from the table's first start to its last end. Dependency arrows and rescheduling reach the rows read so far. |\n| Gallery, list, inbox | Only the records in sight are drawn and read. A grouped list shows each group's size and reads its records when opened. |\n| Chart, dashboard | The server groups and adds up, over every record. A date axis is read a month at a time, in the viewer's months on a column of moments: one grouped request for each part of the records' span with one UTC offset, two a year in a zone with summer time. The records behind a tile show the first 200 and how many there are. |\n| Form | Stores its record through the source, and reads nothing. |\n| Links | A link cell reads the records it names by id, for the cells on screen, in one request. The picker searches the whole table, and a name pasted into the cell is found there. A link edits from either side: on the side that holds no list (a customer's orders), each tick changes that order. |\n| Lookups, rollups and counts | Read the linked records by id. Over a link stored on the big table (a customer's orders), the server counts and adds up: one grouped request a screen, whatever the size. |\n| Search across tables | Each big table searches on the server, and says how many match. |\n| Find in a grid | The search box filters on the server: a find bar could only mark the rows loaded. |\n| Automations | Run on any record, read by id. A scheduled one runs on every matching record, block by block. |\n| Share links | Show the whole table, without the fields the share leaves out. |\n| Autonumber | The next number is the table's largest plus one. Two people can reach it together, so each record is checked once stored: of any two holding a number the later one takes the next, and a column with a UNIQUE constraint refuses the second and the write tries again. Numbers can skip. A number a person deleted in this page is not given out again while it stays open, so an undo finds its number free. After a reload, or on a teammate's page, the number of a deleted record can be given out again, since the next one follows the largest left. A restored record whose number was given out meanwhile takes the next one. |\n| Changing a field's type | Asks first, with the number of records. Then `convertField` does it in one statement (`createSqlDataSource` with `schemaChanges: true` has one on Postgres), or the base reads a block at a time and writes the records whose value changes, sixteen at a time. The other choice changes the field only. |\n| Download as SQL | Writes every record, a block at a time, so the file can be larger than the page could hold. |\n| Duplicate table | Copies in the database with the `duplicate` option (`schemaChanges: true` gives a Postgres source one). Without it the records are read into the page a block at a time, with the count in the tab bar, and the copy lives in the page, so a very large table needs the option. |\n| Edits | Written by id: a change to the 900,000th record is one `update`. |\n| Edit fields | On a selection, in a dialog with progress and Stop. The selection is read first, and only records whose value changes are written, a block of 1,000 at a time. Past 10,000 records it asks first, with the exact number, because Undo cannot take that many back. Stop ends it after the block under way. A block that fails ends it too: the page reads that block back, says how many records changed, and Undo takes those back. Records that landed past 10,000 stay changed. |\n| A range of cells | Paste, one value pasted over a range, Delete, the fill handle and Ctrl+Enter reach every record the selection covers, loaded or not. Up to 200 cells of rows the grid holds are written at once, cell by cell. Anything larger, or reaching rows not loaded, runs in the Edit fields dialog: the records are read in view order, then written a block of 1,000 at a time (200 on a source that writes a request a record, such as Supabase, so the count moves and Stop answers within a block), with the same progress, Stop, count at the end, question past 10,000 and one Undo step. A block the database refuses part way ends the job: the count says how many records changed, and Undo takes those back. If records moved while the range was read (a teammate's edit changed the order or the set of a sorted view), nothing is written or copied: the page says, for a paste, \"Records moved while this was read. Nothing was changed. Try again pastes from \"Acme\" down over the rows as they are now.\", naming the first record selected. Try again finds that record by its id where the list has it at the click, goes from it over as many rows below it as before, and selects those rows first, so the screen shows what it writes; its dialog lies over them, so it names them too (\"From \"Acme\" down over 2,500 rows, selected in the view.\"). If the record is gone or no longer among the rows shown, it writes nothing and says so, and so it does when fewer rows stand below the record now than the first try covered (the record fell to the end of a sorted view): it never turns lines into new records the first try wrote over records that were there. To tell that records moved, the whole id sequence is checked, in requests that ask for ids alone: a range longer than a block reads its ids first and each block must bring the same ids in the same places, each block also reads the last record of the one before again, a record read twice counts as moved, and once the last block is in, the ids of the whole range are read again and must be the very ids read, in the same order. Each of those reads asks for the whole range at once (up to 20,000 ids); a server that caps an answer (PostgREST's `db-max-rows`, 1,000 on Supabase) is read on from where its answer ended, a page of its own size at a time, and only a different id sequence, or a list that ends before the range does, counts as moved. On Supabase and PostgREST these reads select the key alone. Any difference counts, a move inside the part already read too. Try again acts only in the view the cells were chosen in, as it was then: after a switch to another view, or a change of its sort, filter, groups, open groups or shown fields, it writes nothing and says \"That view is no longer open, so nothing was changed.\" or \"The view has changed since, so nothing was changed.\" An answer to a question the job asked (past 10,000 records, new records past the last row, a delete) does the same. A new record that is not stored yet sits on top of the list and in no block of the table: a range over it leaves it out and says how many it left, a paste keeps its line for it, and a copy has its line. A drag of a range over rows not loaded is refused, with the number. A paste while the view is still loading is refused too, and never makes a record. |\n| Copy | Rows not loaded are read first, in a dialog with progress and Cancel, each line a real record with its worked out fields. A copy holds up to 2,000,000 cells: that many, 20 MB of text, took 4.6 s with a longest frame of 450 ms in Chrome on a production build, reading from the page. Past that nothing is copied, the page says how many cells were asked for, and Download gets all of it. If a browser will not take the text so long after the key press, the dialog asks for one more click. |\n| A paste past the last row | A line where no value fits its field makes no record. Up to 10 new records are added at once; past 10 the page asks first, with the number. They go in blocks of 1,000 (`createRows` makes a block one statement), and one Undo takes them out the same way, with the cells the paste wrote. A record someone has worked on since stays, and the page says so. |\n| \"+\" | In every view of records. It opens a new record, filled in from the view's filter (Region is EU in a view of EU orders). Nothing is stored until it holds a value, and one closed empty is gone. If the database refuses it, it stays in the grid with its values and the next edit tries again. What the database fills in, such as an order number, comes back into the record. Ctrl+Z takes a stored record out again. |\n| Insert record, Duplicate | Insert opens a draft, as \"+\" does. Duplicate stores the copy and shows it at the top of the grid. |\n| A new field | The \"+\" at the right of the header row, for whoever may change the schema (`columnAdder={false}` hides it). `createSqlDataSource` with `schemaChanges: true` adds the column on Postgres, MySQL and SQLite. A new field starts empty: when a column by that name is already in the table it takes the next free name, and says so. A column that cannot be added takes the field back. A source that cannot add one, such as Supabase, shows the `ALTER TABLE` to run, and `addField` on `baseSource` is your own way to make it. There, Duplicate field copies nothing: it shows the statement for the copy's column, to run before duplicating. An Autonumber or Created time field stays blank on the records already there. |\n| Delete | By id, to the trash, as the database holds the record at that moment; Ctrl+Z puts it back whole, columns the base shows no field for included, under the key it had, on a table the page holds and on a big one alike. A column off the source's schema never reaches the page, so it takes its default, unless the source reads whole records (`wholeRecords` on `createSqlDataSource`). The record stays in the trash until the database has it. What only the database may write stays the database's: a generated column works itself out again, and a `GENERATED ALWAYS` key gets a new value. Time columns go back as the database wrote them, past the millisecond and at a local time the clocks skipped, on Postgres and MySQL with any driver. A record the database refuses to delete stays in the table and out of the trash, and so does one a row-level policy skipped without an error: after a delete the base looks. |\n| Soft delete | `softDelete: 'deleted_at'` marks a record instead of removing it, and a restore clears the mark, so nothing is rewritten. See [Undo, trash and import](./undo-trash-import.md#soft-delete). |\n| Select all | Selects every record of the view, read or not, and the count says so; a group's tick selects all of the group. A new filter clears the selection. With `selectionBar={{ actions: ['deleteRows'] }}` (and `enableSelectionBar()`), Delete first reads what the selection covers, then asks with the exact number. The records go a block at a time, with a Stop button: what went before you stopped stays gone and stays restorable. `deleteRows` and `createRows` on your data source make a block one statement. Up to 10,000 go to the trash and can be undone; above that the question says they cannot be restored. A group that reads back a different size from its group row stops the job with nothing deleted, since what was left out of the selection could not be told apart. A record that leaves the view after you confirm is still deleted, since you confirmed that set. |\n| Changes made elsewhere | `apply(changes)` updates those rows in the grid, highlighted and credited, whether the page holds them or not. |\n| Formulas | Worked out in the page for the rows on screen. |\n\nAll of this is `getRows` with what a grid view already sends: a range, the\nview's filter, a sort, `groupBy` and `groupKeys`, an `in` list on the key\n(`filterModel.columns.id.selectedValues`), and for a chart or a total\n`aggregations` and `needsGrandTotal`. Grouping a link for a count from its other\nside sets `groupRawKeys`: group by the linked id, which a foreign key does. A calendar's\n`groupDays` and a chart's `groupMonths` ask for a date level by day (`2026-03-14`) or month\n(`2026-03`), on a column of moments in the zone `dayOffset` minutes ahead of UTC. A source that passes over them answers by value and the page adds the groups\nup, which on a time field is a group a record. The data source applies the view's filter\nand acknowledges it with `appliedExpression: true`. The\nSupabase, REST and SQL sources already answer all of these, and\n`createSqlDataSource` also runs a view's filter, sort, groups and totals on a\nformula, lookup or rollup in the database\n([computed fields in SQL](./server-compute.md#computed-fields-in-sql)). On your own\nsource, index the fields views group and place records by: a busy calendar\nmonth is a request per day.\n\n### Index what views sort on\n\nA sort on a column with no index sorts the table again for every block a person\nscrolls to, and that is what limits how many people a big table serves. Give each\nfield a view sorts on an index with the table's key after it:\n\n```sql\nCREATE INDEX orders_amount ON orders (amount, id);\n```\n\nThe key is there because a read breaks ties on it, in the sort's direction\n(`ORDER BY amount DESC, id DESC`). Leave both columns ascending, as above: one index\nthen serves a sort either way. `(amount DESC, id)` matches neither direction, and on\nMySQL it goes unused.\n\nThrough `createBaseEndpoint` on a table of 1,000,000 records, sorted on a numeric\ncolumn, blocks of 100, default database settings, on a shared development machine:\n\n| Median time for a block | No index | With the index |\n|---|---|---|\n| Postgres, one person | 2,263 ms | 567 ms |\n| MySQL, one person | 2,476 ms | 42 ms |\n| Postgres, 30 people at once | 23,576 ms | 2,118 ms |\n| MySQL, 30 people at once | 22,968 ms | 713 ms |\n\nA formula or a lookup has no column to index: a sort on one costs a sort of the\ntable for every block, so a view with many people on it is better sorted on a stored\nfield. A filter narrows what is sorted, and an index that starts with the filtered\nfield and ends with the sorted one serves both.\n\n### What the database cannot run\n\nA view never goes unanswered. When a filter, sort, group or total needs a field\nthe data source cannot compute, or the source does not acknowledge a filter, or\nit fails on a computed field, the page reads the table a block at a time, works\nthe field out with the same code the cells use, and keeps the ids and the values\nthe request needs. The view then reads its records by id. The tab bar shows how\nfar it has got.\n\nIt gives what the cells say, on any backend, and it is slower than the database:\nabout a minute for a lookup over the 1,000,000 records of the demo on a dev\nserver. It is read once. Scrolling, sorting again and the lanes of a board cost\nnothing more, a record you or someone else changes is updated where it stands,\nand changing a field reads again. Taking the sort or filter away stops the read.\nRecords changing while it reads are read again once, and if the count still moves\nthe base says so.\n\nTell the base what your source can work out with `computes`, and that field stays\nin the database. Without it, only `createSqlDataSource` is asked: a `queryRecords`\nbackend needs `computes: () => true` or it is right but slow.\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\ndeclare const orders: ServerDataSource<Record<string, unknown>>\n\nconst source = baseSource(orders, { computes: (def) => def.field === 'tax' })\n```\n\n`@svgrid/enterprise/base/compute` has both halves:\n[`queryRecords`](./server-compute.md#answer-a-paged-table) for records a\nserver holds, and [`fieldFilterToSql`](./server-compute.md#the-views-filter-in-sql)\nfor a database.\n"
5202
+ "markdown": "# Keep a base in a database\n\nA base can keep its records in your database instead of the page. Give a\ntable a `source` and `<SvBase>` loads the records from it when it opens, then\nwrites every change back: a cell edit, a pasted range, a new record, a\ndeleted one, a link made from either side. Postgres, MySQL, SQLite, Supabase\nand any REST API work the same way, because each is reached through the\ndata source contract the grid already uses.\n\n<div data-docs-demo=\"516-base-postgres\" data-height=\"760\"></div>\n\nThe demo above is SvGrid Base on a real Postgres database running in the\npage (PGlite). Change a status and the log shows the `UPDATE` it ran.\n\n## How it fits together\n\n![SvGrid Base keeps every record in the page. The writer compares the tables with what the database holds and sends inserts, updates and deletes through baseSource to a data source, which talks to Postgres, Supabase or your API.](/docs-media/base-data-flow.svg)\n\n- **The page holds every record of a table.** Views, filters, groups,\n summaries, colors and automations all work on those records, so they stay\n instant. A table with more records than the page should hold (10,000 by\n default) is read from the database a part at a time instead, by itself: see\n [A table of millions](#a-table-of-millions).\n- **The writer sends what changed.** After each change it compares the\n tables with what the database holds and queues an `INSERT` for a new\n record, an `UPDATE` with only the fields that changed, or a `DELETE`. The\n writes go one at a time, in order, so an update never overtakes the insert\n of its own record.\n- **A refused write goes back.** When the database says no (a constraint, a\n row level security policy, a network error), the cell returns to what the\n database holds, the end of the tab bar says why in plain words, and a notice\n names the record and the field. What the database said is in the tooltip, or\n through an endpoint in the server log, because it names tables and values:\n [what a person reads](./postgres.md#3-one-endpoint-per-table).\n- **Changes made elsewhere come in.** With `realtime`, the base reads a table\n again when someone else changed it and takes what moved, without writing\n anything back.\n\n## Give a table a source\n\n`baseSource` wraps any data source:\n\n```ts\nimport { baseSource, type BaseTable } from '@svgrid/enterprise/base'\nimport { createRestDataSource } from '@svgrid/enterprise'\n\nconst tasks: BaseTable = {\n id: 'tasks',\n name: 'Tasks',\n primaryField: 'name',\n data: [],\n columns: [\n { field: 'name', header: 'Task', fieldType: 'text' },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: [{ value: 'Todo' }, { value: 'Done' }] } },\n ],\n source: baseSource(createRestDataSource({ url: '/api/tasks' })),\n}\n```\n\n```svelte\n<script lang=\"ts\">\n import { SvToaster } from '@svgrid/grid'\n import { SvBase, enableBase, type BaseTable } from '@svgrid/enterprise/base'\n\n enableBase()\n let { tables }: { tables: BaseTable[] } = $props()\n</script>\n\n<SvBase bind:tables editable />\n<SvToaster />\n```\n\n`data` can start empty: the base fills it. A write that is refused shows a notice\nand the tab bar says so; see [Notices](./quick-start.md#notices).\n\nThe data sources that ship with `@svgrid/enterprise`:\n\n| Source | Talks to | Tutorial |\n|---|---|---|\n| `createSupabaseDataSource` | Supabase (PostgREST), from the browser | [Build a base on Supabase](./supabase.md) |\n| `createSqlDataSource` | Any SQL database, through an `execute(sql, params)` you write. It reads and writes the columns on its schema and no others; `wholeRecords: true` reads and writes every column, for a source in the same process as the page, and `createKitHandlers` refuses one | [Build a base on Postgres](./postgres.md) |\n| `createKitDataSource` | Your SvelteKit endpoint, which runs `createSqlDataSource` on the server | [Build a base on Postgres](./postgres.md) |\n| `createRestDataSource` | A JSON API: `GET`, `POST`, `PATCH`, `DELETE` | [Build a base on your API](./rest-api.md) |\n| `createInMemoryDataSource` | An array, for tests and prototypes | |\n\nYour own source works too: an object with `getRows`, `createRow`,\n`updateRow` and `deleteRow` (the grid's `ServerDataSource`).\n\n## Tables your server makes, copies and deletes\n\nA table with a source lives in your database, so a new one should too. Give\n`<SvBase>` the four hooks and the tab bar's Add table, Duplicate table and\nDelete table, and the trash's Restore for a table, go through your server:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, baseSource, type BaseTable } from '@svgrid/enterprise/base'\n import { createRestDataSource } from '@svgrid/enterprise'\n\n let { tables } = $props()\n\n // What your API answers for a table, as a BaseTable reading its records from it.\n const asTable = (t: any): BaseTable => ({ ...t, data: [], source: baseSource(createRestDataSource({ url: `/api/tables/${t.id}/rows` })) })\n const post = async (url: string, body?: unknown) => {\n const res = await fetch(url, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify(body ?? {}) })\n if (!res.ok) throw new Error((await res.json()).error ?? 'The table was not saved.')\n return res.json()\n }\n</script>\n\n<SvBase\n bind:tables\n editable\n onAddTable={async (name) => asTable(await post('/api/tables', { name }))}\n onDuplicateTable={async (table, name) => asTable(await post(`/api/tables/${table.id}/copy`, { name }))}\n onDeleteTable={async (table) => (await post(`/api/tables/${table.id}/trash`)).item}\n onRestoreTable={async (item) => asTable(await post(`/api/tables/${item.table.id}/restore`))}\n/>\n```\n\n- **Add table** calls `onAddTable` with the name the base would give the\n table (\"Table\", \"Table 2\"...). The table it answers is added and opened,\n and its `source` is read at once, as the first tables' were. Nothing\n answered adds nothing. A rejection's message is shown.\n- **Import a file as a table** calls `onImportTable` with the table the\n import built: its fields (with the types it found) and its records in\n `data`. Keep it, and answer the table to add, with its `source`; it opens and\n is read from there. Nothing answered adds nothing; a rejection's message is\n shown. Without the hook the table is made in the page.\n- **Duplicate table** calls `onDuplicateTable` with the table and\n \"Name copy\". The copy it answers is added and opened.\n- **Delete table** asks the person first, as always, then calls\n `onDeleteTable`. `false`, or a rejection, keeps the table. A trash item it\n answers is what the trash shows for the table, and your `trashStore` is not\n handed it again. Otherwise the base keeps its own item.\n- **Restore** of a table in the trash calls `onRestoreTable` with the item.\n The table it answers goes back where it was, read from its `source`, and\n opens. Nothing answered leaves the item in the trash.\n\nA table made, copied or brought back this way is yours, not a step of the\npage's history: Ctrl+Z does not take it away. Without a hook, the base makes,\ncopies, deletes and restores tables in the page, as before.\n\nAny table you put in `tables` after the base opened, hooks or not, is read\nfrom its `source` and kept in sync like the first ones.\n\n## Link fields\n\n![How link fields are stored. A link to one record is a foreign key column. A link to many records is a junction table with one row per pair. The field that links back is not stored.](/docs-media/base-links.svg)\n\nIn the page a link field holds record ids. `links` says how each one is kept:\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport { createRestDataSource } from '@svgrid/enterprise'\n\nconst taskTags = createRestDataSource({ url: '/api/task_tags' })\n\nconst source = baseSource(createRestDataSource({ url: '/api/tasks' }), {\n links: {\n // A link to one record: the column with the linked record's id.\n project: { column: 'project_id' },\n // A link to many records: a junction table, one row per pair.\n tags: { through: taskTags, from: 'task_id', to: 'tag_id' },\n },\n})\n```\n\nThe field that links back (a project's Tasks, when tasks have a Project\nfield) is worked out in the page from the other side and never stored. Link\na task from the project's side and the base writes that task's\n`project_id`. A link field that is neither in `links` nor the other side of\none is written as it is, a list of ids, which suits a Postgres `text[]` or a\n`jsonb` column.\n\nA list (a multiple select, a vote, a sparkline) goes to its column as that column takes\nit: an array to `text[]`, JSON to `json` and `jsonb`, and JSON text to a text column, on\nMySQL, SQL Server and SQLite too. It reads back as the list it was.\n\n## Keys the base can use\n\nA record needs one column that names it. When a base reads a database:\n\n- A table with a primary key of one column is a table.\n- A table keyed by both ends, `order_tags (order_id, tag_id, primary key (order_id, tag_id))`,\n with both columns foreign keys, is a link to many, like a junction with its own `id`. A\n link goes by the whole key, so removing a tag deletes that one pair. A tenant column in\n the key, named in `tenantColumn`, is part of the key and no end of the link. Other columns\n on such a table are not read and not written: a new link gives them their defaults, one\n that is `NOT NULL` with no default refuses the link, and the server log names them. A\n pair whose two ends the base does not hold, by their keys, is left out.\n- Any other table with a composite key, or with no primary key, is left out of the base, a\n view included. The rest of the base opens, and the log names the table and says why.\n Nothing runs on part of a key, because a delete by one column of two reaches every row\n that shares it.\n\n`keys` brings such a table in, when one of its columns is unique:\n\n```ts\nexport const { GET, POST } = createBaseEndpoint({ execute, allTables: true, public: true, keys: { v_orders: 'id' } })\n```\n\nThat is your word for it: nothing checks, and a delete reaches every row that holds the\nvalue. `baseFromDatabase` and `baseFromSupabase` take it too. On Supabase the key comes\nfrom the API description; a project that does not show that to the key in use has only a\ncolumn named `id` to go by.\n\n### Foreign keys of more than one column\n\nA foreign key of more than one column, as a database keeps tenants apart with,\n`foreign key (tenant, customer_id) references customers (tenant, id)`, is a link through\n`customer_id` to that tenant's customer, when `tenantColumn` names `tenant` on both tables.\nThe tenant column stays the read-only tenant, not a link. It works the same in a pair\ntable, `(tenant, order_id) -> orders (tenant, id)`. Any other key of more than one column,\nthree columns or two without the tenant, links nothing: its columns are plain values and\nthe log names the key. A key links only to a table its bare name reaches: on Postgres a\ntable of any schema on the search path, unless a table of that name in an earlier schema\nhides it (a key to a hidden table stays a plain value); on SQL Server the user's default\nschema; on MySQL the connected database. The other table still has to be one the base\nserves. Supabase does not describe a key of more than one column, so there\nneither column is a link.\n\n## Record ids\n\nA new record gets an id in the page at once, so it can be edited and linked\nbefore the database answers. By default that id is its key in the database\ntoo: give the key column a text type.\n\nWhen the database makes the key (a `serial` column, a `uuid` with a default),\nset `serverIds: true`. The record is inserted without the page's id, and the\nwriter keeps the page id for the grid and translates to the database id for\nevery later write and every link that points at the record.\n\n## What a write answers\n\nThe row a server answers is the truth. After a write the cell, the record and the views\nshow what was stored: a value a `before*` hook or a trigger changed, a tenant the scope\nstamped, a default on a new record. A number the column rounded shows rounded. Undo goes\nback from what was stored. A block of Edit fields or of a range on a big table answers with\nwhat hooks changed, and a trigger's work on such a block shows when the view reads the rows\nagain. On a table the page holds, a paste, a clear or an Undo of 20 records or more is\nwritten in blocks, and each block's records are read back once it lands, so a trigger's\nwork shows at once there too.\n\nThe answer is the row the write left, not a teammate's write a moment later. Postgres and\nSQLite say it in the write itself (`RETURNING`). SQL Server writes and reads it back in one\ntransaction of one batch. MySQL has no way to do either in one statement: give\n`createSqlDataSource` (or `baseFromDatabase`, `createBaseEndpoint`) its `transaction` and the\nwrite and its read go in one. Without it a teammate's write to the same cell in the moment\nbetween the two is taken for yours, and an Undo of yours can then write over theirs.\n\n## What is not written\n\nFields the page works out are never sent: formula, lookup, rollup, count and\nbutton fields, and the field that links back. Add any other field to `skip`\nto keep it in the page only.\n\n## What a cell holds\n\nA base reads dates, times, bigints and decimals as the database writes them, as\ntext. A SQL source asks for a text copy of each such column\n(`to_json(col) #>> '{}'` on Postgres, `CAST(col AS CHAR)` on MySQL) beside the\ncolumn itself, so sorting and filtering still use the real column.\n\n| Column | What the page holds | What people see |\n|---|---|---|\n| `date` | `2026-03-08` | That day, in any zone |\n| `timestamp`, `datetime` | `2026-03-08T02:30:00.123456` | That wall clock, in any zone |\n| MySQL `timestamp` | The wall clock of the connection's `time_zone` | That wall clock, in any zone |\n| `timestamptz` | `2026-03-14T22:00:00.654321+00:00` | That moment, in the viewer's zone |\n| `bigint`, `numeric`, `decimal` | Text past what a JS number holds | Every digit |\n| `json`, `jsonb` | The text the database writes | Every digit: a driver's `JSON.parse` would round a number past 2^53 |\n\nA time someone types keeps its wall clock on a column without a zone, and is that\nmoment on a `timestamptz`. An undo or a restore writes back the text it read, so\nnothing moves by a zone or loses digits past the millisecond.\n\nThe days of a filter and the months of a chart are the viewer's. On a column of moments\n(Postgres `timestamptz`, SQL Server `datetimeoffset`, SQLite text that ends in a zone)\n\"is 8 March\", \"today\" and \"within the past week\" keep the records whose moment falls on\nthose days where the person is, and a chart counts each record in the month it reads as\nthere, across a change of the clocks too. The page sends each date condition with the\nmoments its days start and end (`moments` on the condition), since a server has no zone\nof the viewer's, and `createSqlDataSource` compares them on such a column. A date and a\nwall clock, a MySQL `TIMESTAMP` among them, compare their days and read the same everywhere.\nOn Supabase, say which columns hold moments: `supabaseFieldFilter(columns, { moments: (def) => def.field === 'starts_at' })`.\nWithout that, a condition there compares the database's days.\n\nOver PostgREST (Supabase) a date already arrives as text. A JSON number cannot\ncarry more than about 15 digits, so the first read that shows a longer one reads\nthat column again with `col::text`, and later reads bring the copy along. A change\nthat arrives through Realtime is plain JSON: such a number in it is rounded until\nthe next read.\n\nJSON in a `text` or `longText` field shows as JSON text. A change goes back as\nJSON; text that is not valid JSON is refused with a message and the cell goes back. A\n`jsonb` column writes its own spaces and key order, and that is no change: nobody is told\nof one. Over PostgREST a JSON column arrives as JSON, so a number past 2^53 in it is rounded.\n\nOpening a cell and leaving it writes nothing, whatever the field shows.\n\n## Changes made elsewhere\n\n![An edit, step by step: Ana's base sends the field that changed, the database tells every subscriber, and Ben's base reads the table again and takes what moved.](/docs-media/base-realtime.svg)\n\n`realtime` takes a function that calls `apply` with what changed and returns a\nfunction that stops listening. With Supabase, hand it Realtime's events as\nthey come:\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport {\n createSupabaseDataSource,\n createSupabaseRealtime,\n type EntitySchema,\n type SupabaseClientLike,\n type SupabaseRealtimeClientLike,\n} from '@svgrid/enterprise'\n\n// Your supabase-js client: createClient(url, anonKey).\ndeclare const client: SupabaseClientLike & SupabaseRealtimeClientLike\ndeclare const schema: EntitySchema\n\nconst source = baseSource(createSupabaseDataSource({ client, table: 'tasks', schema }), {\n realtime: (apply) => createSupabaseRealtime({ client, table: 'tasks', onChange: apply }),\n})\n```\n\nThe base updates just the records a change names: the other records keep\ntheir objects, so the grid redraws one row, not 50,000, and the table is not\nread again. A burst of changes (an agent updating 200 records) goes into the\npage as one update. A change to a record this page wrote in the last few\nseconds may be its own write coming back late, or someone else's edit after\nit: that record alone is read again, by id, to know which.\n\nAny other feed works the same way: a WebSocket message, a server-sent event,\na poll. Call `apply` with the changes, as the database has the records\n(its column names, its ids):\n\n```ts\nimport { baseSource, type BaseChangeInput } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\ndeclare const tasksSource: ServerDataSource<Record<string, unknown>>\n\nconst source = baseSource(tasksSource, {\n realtime: (apply: (changes?: BaseChangeInput) => void) => {\n const socket = new WebSocket('wss://example.com/tasks')\n socket.onmessage = (m) => {\n const { type, id, row, changed } = JSON.parse(m.data)\n if (type === 'update' && changed) apply([{ type: 'update', id, fields: changed }])\n else if (type === 'delete') apply([{ type: 'delete', id }])\n else apply([{ type, record: row }])\n }\n return () => socket.close()\n },\n})\n```\n\n| A change | Carries |\n|---|---|\n| `{ type: 'insert', record }` | The new record, with its key column. |\n| `{ type: 'update', id, record }` | The whole record as it is now. |\n| `{ type: 'update', id, fields }` | Just the columns that changed. |\n| `{ type: 'delete', id }` | The key of the record that went. |\n\nA change that comes back from the page's own write (the database tells every\nlistener, the page that wrote it included) moves nothing and is not reported.\nAn update with `fields` for a record the page does not have, and `apply()`\nwith nothing, read the whole table again. An update with `id` alone reads that one record.\n\n### A feed on an endpoint\n\n`baseFromEndpoint` takes one `feed` for the whole base. It opens when the first table\nlistens and closes with the last. Each message names its table, a junction table\nincluded, and a change in a junction reaches the record it starts from:\n\n```ts\nimport { baseFromEndpoint } from '@svgrid/enterprise/base'\n\nconst tables = await baseFromEndpoint('/api/base', {\n feed: (apply) => {\n const events = new EventSource('/api/base/feed')\n events.onmessage = (e) => {\n // { \"table\": \"orders\", \"changes\": [{ \"type\": \"update\", \"id\": \"o1\", \"fields\": { \"status\": \"Paid\" } }] }\n const { table, changes } = JSON.parse(e.data)\n apply(table, changes)\n }\n // After a reconnect, changes may have been missed: every table reads again.\n let opened = false\n events.onopen = () => {\n if (opened) apply()\n opened = true\n }\n return () => events.close()\n },\n})\n```\n\n`apply(table)` with no changes reads that table again, and `apply()` reads every table again.\n\nKeep it to one connection. A browser opens at most six connections to one host over\nHTTP/1.1, which is what a Vite or Node dev server speaks, and a connection held open by a\nfeed is one of them. With a feed per table on a base of six tables and junctions, no write\nreaches the server. `realtime: (table, apply) => stop`, called per table, is still taken\nfor a feed that shares one connection itself, such as channels on one socket.\n\nThe feed is yours to write and yours to scope: with a tenant `scope`, send each person their\ntenant's rows only.\n\n### What a feed has to cover\n\nA feed is per table, and some of what a person changes is not in that table:\n\n- **A link to many lives in its junction table.** Adding a tag writes a row to `order_tags`\n and nothing to `orders`, so a feed of `orders` alone never tells teammates. Listen to the\n junction too, on the table that stores the link, and name the record it starts from:\n\n ```ts\n import { baseSource } from '@svgrid/enterprise/base'\n import type { ServerDataSource } from '@svgrid/grid'\n\n declare const ordersSource: ServerDataSource<Record<string, unknown>>\n declare const orderTags: ServerDataSource<Record<string, unknown>>\n // Your feed: calls back with what changed in that table, and returns how to stop.\n declare function listen(table: string, on: (change: any) => void): () => void\n\n const source = baseSource(ordersSource, {\n links: { tags: { through: orderTags, from: 'order_id', to: 'tag_id' } },\n realtime: (apply) => {\n const orders = listen('orders', (change) => apply(change))\n const pairs = listen('order_tags', (row) => apply(row?.order_id == null ? undefined : { type: 'update', id: row.order_id }))\n return () => { orders(); pairs() }\n },\n })\n ```\n\n A feed that only says \"something changed\" calls `apply()` for both. `baseFromSupabase`\n and `baseFromEndpoint` listen to the junction tables by themselves, and needs them in the `supabase_realtime`\n publication. A delete's event there has only the key unless the junction table has\n `REPLICA IDENTITY FULL`, so a removed link reads the table again.\n- **A soft delete reaches a feed as an update.** The base reads an update that sets the mark\n as a delete, so send the column. Through `createKitDataSource` the page learns the column\n from the server's first answer. A restore reads as an update of a record the page does not\n have. A page-held table takes it in, and reads the table if it has links through a junction.\n A table read a part at a time reads the rows on screen, at most once a second, because a\n feed cannot tell a restore from an edit of a record scrolled out of view.\n- **A feed that goes down says nothing.** Nothing here can tell silence from no changes.\n\n### What people see\n\nA change made elsewhere does not just appear. Until someone opens the record:\n\n- in a grid view, the cells that changed are tinted, and a record added\n elsewhere is tinted across the row (Kanban and gallery cards are not);\n- the row number has a dot; hover it for \"Ana Ruiz changed this record 2\n minutes ago\";\n- the view bar counts them, \"3 changed by Ana Ruiz\", and a click on the count\n marks them all seen.\n\nOpening the record clears its marks. Its activity lists each change as\n\"Ana Ruiz changed Priority\", with the value before crossed out and the value\nnow: Low → Urgent.\n\nThe name comes from the record itself: its `lastModifiedBy` field (for a\nrecord added elsewhere, `createdBy` first). Whoever writes to the database\nwrites their user id to that column, and the base looks it up in\n`setFieldUsers`. Without such a field the change is credited to \"Someone\nelse\". Your own writes are never reported back to you: the base compares what\nit reads with what it last wrote.\n\n### An AI agent writing to the base\n\nThe same path serves an agent: a script, a worker calling a model, or Claude\nCode with your database's MCP server. Give the agent a user id, add it to\n`setFieldUsers` with a name and an avatar, and have it set `modified_by` (or\nwhichever column your `lastModifiedBy` field reads) on every row it writes.\nPeople then see exactly what it changed, review it in a view filtered on a\n\"Needs review\" field, correct what it got wrong, and approve the rest with a\nbutton field and an automation.\n\n[Let an AI agent work in your base](./ai-agents.md) walks through it with\nClaude Code and Supabase, and [Review what an agent changed](./review-queue.md)\ncovers the review side on any database. Here the agent is a stand-in made of\nkeyword rules, so the demo runs without a model or a key. It writes to the database with SQL; press \"Run the triage\nagent\", then open a ticket.\n\n<div data-docs-demo=\"700-base-ai-triage\" data-height=\"900\"></div>\n\nTo act on changes made elsewhere in code (a notice, a log, a count of your\nown), pass `onRemoteChange` to `<SvBase>`. It gets each change once it is in\nthe page:\n\n```ts\nimport type { BaseRemoteChange } from '@svgrid/enterprise/base'\n\nfunction onRemoteChange(changes: BaseRemoteChange[]) {\n for (const c of changes) {\n // c.type: 'insert' | 'update' | 'delete'; c.table, c.id, c.record\n // c.fields (updates): [{ field: 'priority', from: 'Low', to: 'Urgent' }]\n if (c.type === 'update') console.log(c.id, c.fields)\n }\n}\n```\n\n## Status and errors\n\nThe end of the tab bar says what the base is doing:\n\n| It reads | When |\n|---|---|\n| Loading records... | The tables are loading |\n| Saving... | Writes are on their way |\n| Still saving: the server has not answered yet | A write has been waiting a while |\n| All changes saved | Every write went through |\n| Not saved: *reason* | A write was refused; the cell went back |\n\n![A refused write: the database rejects a status, the cell goes back, and the tab bar says the database does not accept this value.](/docs-media/base-refused-write.png)\n\nA table that cannot be read says so where its records would be, for as long as it lasts,\nwith why and a **Try again**. No empty table is shown for it, and nothing can be added to\nit. The other tables of the base open as usual.\n\nA table read a part at a time (`serverRows`) whose next part cannot be read says so in the\ntab bar, \"Could not load *table*: *reason*\", with **Try again**, for as long as a part of the\nview is unread, wherever it is scrolled to. The rows that did not come are a tinted band with\n\"Could not load these rows.\" and **Retry** on its first row on screen. A read again of rows\nalready on screen, after a teammate's change, that fails leaves those rows as they were: the\ntab bar says so with **Try again**, which reads them again.\n\nWhen the server sent a reason with its error (`{ \"error\": \"...\" }` and no `code`), that reason\nis what the notice says. `createRestDataSource` passes on an API's `{ \"error\" }` or\n`{ \"message\" }` the same way.\n\nA failed write is told in a notice that names the record and the field, says why in\nplain words and shows what was entered, with **Try again** where that can work. The record\nis named as it is stored, never by the value that was refused. A foreign key of more than one\nfield that refuses a write names all of its fields. The notice stays until it is dismissed,\nand \"Not saved\" stays until that record is written: a later write of another record does not\nclear it. A click, a tap or Enter on the status line opens it in full: the reason, and the\nserver's own words when it sent some and the reason does not already say them. The text of\nthe request itself (a status code, a JSON body) is never shown; it is in the browser's\nnetwork log and the server's log.\n\nA new record the database refused takes no edit while it is still listed (while another cell or\nrecord is being edited, or a draft on a table read a part at a time whose refusal no retry can\nchange, such as an id column too narrow): a click, a key or a paste on\nit shows its notice again instead of an editor. An editor open on it when the refusal comes closes\nwith what it holds, and the notice names it. A draft refused for a reason a retry may get past keeps\ntaking edits, and its next edit sends it again.\n\n**Try again** reads the record first, as an undo does. A value that changed since the\nwrite failed, whether a teammate wrote it or you typed it again, keeps that value and the\npage says so. A record deleted since is reported with what you entered. If the record\ncannot be read, nothing is written and the notice offers Try again once more.\n\nA change of many records at once (a paste, a clear or an Undo of 20 or more) that fails\nis one notice with the number of records, and the same **Try again**: it reads them all\nfirst, sends what is still as it was in blocks, and says how many cells changed since and\nhow many records were deleted since. Part of a block may have landed; the page reads the\ntable again to show which. An Undo of some of those cells takes them out of what Try\nagain would send. A refusal of your permission offers no Try again and leaves no step to\nundo, and so does a refusal of the values, an answer of `400`, `409` or `422`: the same\nvalues would be refused again.\n\nA write that gets no answer is not given up on: it may still land, or be refused. After\n`writeTimeout` (60 seconds by default, on `<SvBase>`) writes to other records go on, and\nlater writes to the same record wait behind it, so none can overtake it or be overwritten\nby it landing late. The status says \"Still saving\" for as long as it has no answer, and\nstarts warning at a sixth of `writeTimeout`. The page keeps what was entered, and goes back\nonly if the write is refused. A delete with no answer keeps its copy in the trash until the\noutcome is known, and leaving the page asks first.\n\nWhile someone is typing in a cell or in a field of a record, changes from others wait\nuntil the editor closes, so no row moves under it and no text changes under the cursor.\nThat holds for a live feed and for a read of the table, on any table. If the record was\ndeleted meanwhile, saving it says so and shows the text.\n\n## Options\n\n`baseSource(dataSource, options)`:\n\n| Option | Default | What it does |\n|---|---|---|\n| `links` | none | How link fields are stored: `{ column }` or `{ through, from, to }` |\n| `columns` | none | Column names that differ from field names: `{ dueDate: 'due_date' }` |\n| `idField` | `'id'` | The key column |\n| `serverIds` | `false` | The database makes the key of a new record |\n| `realtime` | none | Hear of changes made elsewhere: `(apply) => stop`; `apply(changes)` updates those records in place |\n| `skip` | none | Fields kept in the page only |\n| `holdInPage` | `10000` | How many records a table may hold in the page. It is where a table changes how it is read, not a cap: a bigger one is read from the database a part at a time, and no record is left out. (`limit` is the old name.) |\n| `convertField` | none | A field of a big table changed type: rewrite its values in the database and resolve with `true`. Without it the base rewrites them a block at a time. |\n| `duplicate` | none | Copy a big table in the database and resolve with the new table's source. |\n| `serverRows` | automatic | `true` always reads from the database, and the first screen comes sooner on a table you know is big; `{ blockSize }` sets the rows per request (200). `false` always holds every record in the page, however many. Left out, the base decides by `holdInPage`. |\n\n## Build a base from your database\n\n`baseTablesFromSchemas` makes the tables from your database's own\ndescription of them: a number column becomes a number field, an enum a\nsingle select with its values, a foreign key a link with a field in the\nother table that links back. Read the descriptions with\n`introspectSupabaseTable`, `introspectDatabase` or `introspectOpenApi`, or\nwrite them out:\n\n```ts\nimport { baseTablesFromSchemas } from '@svgrid/enterprise/base'\nimport { createRestDataSource, type EntitySchema } from '@svgrid/enterprise'\n\nconst schemas: EntitySchema[] = [\n {\n name: 'deals',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'title', type: 'text' },\n { field: 'stage', type: 'enum', options: [{ value: 'Open', label: 'Open' }, { value: 'Won', label: 'Won' }] },\n { field: 'company_id', type: 'relation', relation: { entity: 'companies', labelField: 'name' } },\n ],\n },\n { name: 'companies', fields: [{ field: 'id', type: 'text', primaryKey: true }, { field: 'name', type: 'text' }] },\n]\n\nconst tables = baseTablesFromSchemas(schemas, (schema) => createRestDataSource({ url: `/api/${schema.name}`, schema }))\n```\n\nThe result is ordinary `BaseTable` objects: change a field's type, add views,\nrename a table. [Build a base on your API](./rest-api.md) does this end to end.\n\nThe other way round, `baseToSql(tables)` writes a base as Postgres SQL:\ntables, foreign keys, the tables behind links to many, and the records, with\n`svgrid:` comments for the field settings, views and automations that SQL\nhas no type for. The tables it makes have text keys with a\n`gen_random_uuid()` default, so a record added later gets its key from the\ndatabase. See [Take a base to a database](./quick-start.md#take-a-base-to-a-database).\n\n## A table of millions\n\nA table has no size limit, and nothing here is a setting you have to get right.\nA table of the thousands of records a team keeps in a tracker, a CRM or an\ninventory is held in the page. One that grows past `holdInPage` is read from the\ndatabase a part at a time, the next time it is read, and looks and works the\nsame: every view, link, lookup, rollup, automation and search reads all of it.\nThe page then holds none of its records, only what is on screen.\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\ndeclare const orders: ServerDataSource<Record<string, unknown>>\n\n// Nothing to set: it reads from the database once the table outgrows the page.\nconst source = baseSource(orders)\n\n// A table you know is big: skip the first look, and size the requests.\nconst big = baseSource(orders, { serverRows: { blockSize: 200 } })\n```\n\n<div data-docs-demo=\"723-base-million-orders\" data-height=\"620\"></div>\n\nHow each part of a base reads such a table:\n\n| | Read a part at a time |\n|---|---|\n| Grid views | Read through the server row model. Each request carries the view's filter (`filterModel.expression`), its sort, its groups (one row per key with its count, the records under an opened key) and the search. The count in the view bar is the server's. A grouped view shows select groups in option order and each group as the field draws it, with the records it holds at every level, and the totals of the fields with a sum, average, lowest or highest picked in the footer. Whenever the view is read again (your delete, Restore, Undo or Redo, a new record, an edit once you leave the record, an option renamed, a field converted, a teammate's change), the group rows and the loaded records of every open group are read, with or without a feed: a deleted record leaves its group, a restored or new one shows in its group, and the counts follow. A group closed at the time reads again when it is opened. A teammate's change that a feed sends with its row patches that row where it is; when it changes a field the view groups or sorts by (any field, when the view groups or sorts by a formula or lookup), the view is read again at once, so the record takes its place in the order and in its group. A filtered view reads again within a second after any change, so a record that left the filter goes. |\n| Kanban | Each lane shows its true size, from one grouped request, and reads its cards as they scroll into sight. A card moved to another lane is written by id and sits at the top of that lane until the lane is read again. Sorted by a number field, dropping a card elsewhere in its lane writes that number, so the order is the table's. |\n| Calendar | Reads the dates on screen and no others. A range of up to 400 records is read whole. A busier one is read a day at a time: each day's first records and its size, from one grouped request where the backend groups and the view has no end date, so an empty day costs nothing and a write reads only the days whose size moved. \"+N more\" lists every record of that day. The days are the viewer's: on a column of moments (Postgres `timestamptz`, SQL Server `datetimeoffset`) the sizes, each day's records and that list follow the page's zone, also across a change of the clocks. A date and a wall clock (`timestamp`, `datetime`, a MySQL `TIMESTAMP`, which reads as the server's clock) show the same everywhere, and their days follow what the cell shows. |\n| Gantt | Rows by start date, read as they scroll into sight. The axis runs from the table's first start to its last end. Dependency arrows and rescheduling reach the rows read so far. |\n| Gallery, list, inbox | Only the records in sight are drawn and read. A grouped list shows each group's size and reads its records when opened. |\n| Chart, dashboard | The server groups and adds up, over every record. A date axis is read a month at a time, in the viewer's months on a column of moments: one grouped request for each part of the records' span with one UTC offset, two a year in a zone with summer time. The records behind a tile show the first 200 and how many there are. |\n| Form | Stores its record through the source, and reads nothing. |\n| Links | A link cell reads the records it names by id, for the cells on screen, in one request. The picker searches the whole table, and a name pasted into the cell is found there. A link edits from either side: on the side that holds no list (a customer's orders), each tick changes that order. |\n| Lookups, rollups and counts | Read the linked records by id. Over a link stored on the big table (a customer's orders), the server counts and adds up: one grouped request a screen, whatever the size. |\n| Search across tables | Each big table searches on the server, and says how many match. |\n| Find in a grid | The search box filters on the server: a find bar could only mark the rows loaded. |\n| Automations | Run on any record, read by id. A scheduled one runs on every matching record, block by block. |\n| Share links | Show the whole table, without the fields the share leaves out. |\n| Autonumber | The next number is the table's largest plus one. Two people can reach it together, so each record is checked once stored: of any two holding a number the later one takes the next, and a column with a UNIQUE constraint refuses the second and the write tries again. Numbers can skip. A number a person deleted in this page is not given out again while it stays open, so an undo finds its number free. After a reload, or on a teammate's page, the number of a deleted record can be given out again, since the next one follows the largest left. A restored record whose number was given out meanwhile takes the next one. |\n| Changing a field's type | Asks first, with the number of records. Then `convertField` does it in one statement (`createSqlDataSource` with `schemaChanges: true` has one on Postgres), or the base reads a block at a time and writes the records whose value changes, sixteen at a time. The other choice changes the field only. |\n| Download as SQL | Writes every record, a block at a time, so the file can be larger than the page could hold. |\n| Duplicate table | Copies in the database with the `duplicate` option (`schemaChanges: true` gives a Postgres source one). Without it the records are read into the page a block at a time, with the count in the tab bar, and the copy lives in the page, so a very large table needs the option. |\n| Edits | Written by id: a change to the 900,000th record is one `update`. |\n| Edit fields | On a selection, in a dialog with progress and Stop. The selection is read first, and only records whose value changes are written, a block of 1,000 at a time. Past 10,000 records it asks first, with the exact number, because Undo cannot take that many back. Stop ends it after the block under way. A block that fails ends it too: the page reads that block back, says how many records changed, and Undo takes those back. Records that landed past 10,000 stay changed. |\n| A range of cells | Paste, one value pasted over a range, Delete, the fill handle and Ctrl+Enter reach every record the selection covers, loaded or not. Up to 200 cells of rows the grid holds are written at once, cell by cell. Anything larger, or reaching rows not loaded, runs in the Edit fields dialog: the records are read in view order, then written a block of 1,000 at a time (200 on a source that writes a request a record, such as Supabase, so the count moves and Stop answers within a block), with the same progress, Stop, count at the end, question past 10,000 and one Undo step. A block the database refuses part way ends the job: the count says how many records changed, and Undo takes those back. If records moved while the range was read (a teammate's edit changed the order or the set of a sorted view), nothing is written or copied: the page says, for a paste, \"Records moved while this was read. Nothing was changed. Try again pastes from \"Acme\" down over the rows as they are now.\", naming the first record selected. Try again finds that record by its id where the list has it at the click, goes from it over as many rows below it as before, and selects those rows first, so the screen shows what it writes; its dialog lies over them, so it names them too (\"From \"Acme\" down over 2,500 rows, selected in the view.\"). If the record is gone or no longer among the rows shown, it writes nothing and says so, and so it does when fewer rows stand below the record now than the first try covered (the record fell to the end of a sorted view): it never turns lines into new records the first try wrote over records that were there. To tell that records moved, the whole id sequence is checked, in requests that ask for ids alone: a range longer than a block reads its ids first and each block must bring the same ids in the same places, each block also reads the last record of the one before again, a record read twice counts as moved, and once the last block is in, the ids of the whole range are read again and must be the very ids read, in the same order. Each of those reads asks for the whole range at once (up to 20,000 ids); a server that caps an answer (PostgREST's `db-max-rows`, 1,000 on Supabase) is read on from where its answer ended, a page of its own size at a time, and only a different id sequence, or a list that ends before the range does, counts as moved. On Supabase and PostgREST these reads select the key alone. Any difference counts, a move inside the part already read too. Try again acts only in the view the cells were chosen in, as it was then: after a switch to another view, or a change of its sort, filter, groups, open groups or shown fields, it writes nothing and says \"That view is no longer open, so nothing was changed.\" or \"The view has changed since, so nothing was changed.\" An answer to a question the job asked (past 10,000 records, new records past the last row, a delete) does the same. A new record that is not stored yet sits on top of the list and in no block of the table: a range over it leaves it out and says how many it left, a paste keeps its line for it, and a copy has its line. A drag of a range over rows not loaded is refused, with the number. A paste while the view is still loading is refused too, and never makes a record. |\n| Copy | Rows not loaded are read first, in a dialog with progress and Cancel, each line a real record with its worked out fields. A copy holds up to 2,000,000 cells: that many, 20 MB of text, took 4.6 s with a longest frame of 450 ms in Chrome on a production build, reading from the page. Past that nothing is copied, the page says how many cells were asked for, and Download gets all of it. If a browser will not take the text so long after the key press, the dialog asks for one more click. |\n| A paste past the last row | A line where no value fits its field makes no record. Up to 10 new records are added at once; past 10 the page asks first, with the number. They go in blocks of 1,000 (`createRows` makes a block one statement), and one Undo takes them out the same way, with the cells the paste wrote. A record someone has worked on since stays, and the page says so. |\n| \"+\" | In every view of records. It opens a new record, filled in from the view's filter (Region is EU in a view of EU orders). Nothing is stored until it holds a value, and one closed empty is gone. If the database refuses it, it stays in the grid with its values and the next edit tries again. What the database fills in, such as an order number, comes back into the record. Ctrl+Z takes a stored record out again. |\n| Insert record, Duplicate | Insert opens a draft, as \"+\" does. Duplicate stores the copy and shows it at the top of the grid. |\n| A new field | The \"+\" at the right of the header row, for whoever may change the schema (`columnAdder={false}` hides it). `createSqlDataSource` with `schemaChanges: true` adds the column on Postgres, MySQL and SQLite. A new field starts empty, or with its default: on a table held in the page the base writes the default to every record; on a table read a part at a time `addField` gets it as `default`, to fill the stored records (the one `createSqlDataSource` makes does not yet). When a column by that name is already in the table it takes the next free name, and says so. A column that cannot be added takes the field back. A source that cannot add one, such as Supabase, shows the `ALTER TABLE` to run, and `addField` on `baseSource` is your own way to make it. There, Duplicate field copies nothing: it shows the statement for the copy's column, to run before duplicating. An Autonumber or Created time field stays blank on the records already there. |\n| Delete | By id, to the trash, as the database holds the record at that moment; Ctrl+Z puts it back whole, columns the base shows no field for included, under the key it had, on a table the page holds and on a big one alike. A column off the source's schema never reaches the page, so it takes its default, unless the source reads whole records (`wholeRecords` on `createSqlDataSource`). The record stays in the trash until the database has it. What only the database may write stays the database's: a generated column works itself out again, and a `GENERATED ALWAYS` key gets a new value. Time columns go back as the database wrote them, past the millisecond and at a local time the clocks skipped, on Postgres and MySQL with any driver. A record the database refuses to delete stays in the table and out of the trash, and so does one a row-level policy skipped without an error: after a delete the base looks. |\n| Soft delete | `softDelete: 'deleted_at'` marks a record instead of removing it, and a restore clears the mark, so nothing is rewritten. See [Undo, trash and import](./undo-trash-import.md#soft-delete). |\n| Select all | Selects every record of the view, read or not, and the count says so; a group's tick selects all of the group. A new filter clears the selection. With `selectionBar={{ actions: ['deleteRows'] }}` (and `enableSelectionBar()`), Delete first reads what the selection covers, then asks with the exact number. The records go a block at a time, with a Stop button: what went before you stopped stays gone and stays restorable. `deleteRows` and `createRows` on your data source make a block one statement. Up to 10,000 go to the trash and can be undone; above that the question says they cannot be restored. A group that reads back a different size from its group row stops the job with nothing deleted, since what was left out of the selection could not be told apart. A record that leaves the view after you confirm is still deleted, since you confirmed that set. |\n| Changes made elsewhere | `apply(changes)` updates those rows in the grid, highlighted and credited, whether the page holds them or not. |\n| Formulas | Worked out in the page for the rows on screen. |\n\nAll of this is `getRows` with what a grid view already sends: a range, the\nview's filter, a sort, `groupBy` and `groupKeys`, an `in` list on the key\n(`filterModel.columns.id.selectedValues`), and for a chart or a total\n`aggregations` and `needsGrandTotal`. Grouping a link for a count from its other\nside sets `groupRawKeys`: group by the linked id, which a foreign key does. A calendar's\n`groupDays` and a chart's `groupMonths` ask for a date level by day (`2026-03-14`) or month\n(`2026-03`), on a column of moments in the zone `dayOffset` minutes ahead of UTC. A source that passes over them answers by value and the page adds the groups\nup, which on a time field is a group a record. The data source applies the view's filter\nand acknowledges it with `appliedExpression: true`. The\nSupabase, REST and SQL sources already answer all of these, and\n`createSqlDataSource` also runs a view's filter, sort, groups and totals on a\nformula, lookup or rollup in the database\n([computed fields in SQL](./server-compute.md#computed-fields-in-sql)). On your own\nsource, index the fields views group and place records by: a busy calendar\nmonth is a request per day.\n\n### Index what views sort on\n\nA sort on a column with no index sorts the table again for every block a person\nscrolls to, and that is what limits how many people a big table serves. Give each\nfield a view sorts on an index with the table's key after it:\n\n```sql\nCREATE INDEX orders_amount ON orders (amount, id);\n```\n\nThe key is there because a read breaks ties on it, in the sort's direction\n(`ORDER BY amount DESC, id DESC`). Leave both columns ascending, as above: one index\nthen serves a sort either way. `(amount DESC, id)` matches neither direction, and on\nMySQL it goes unused.\n\nThrough `createBaseEndpoint` on a table of 1,000,000 records, sorted on a numeric\ncolumn, blocks of 100, default database settings, on a shared development machine:\n\n| Median time for a block | No index | With the index |\n|---|---|---|\n| Postgres, one person | 2,263 ms | 567 ms |\n| MySQL, one person | 2,476 ms | 42 ms |\n| Postgres, 30 people at once | 23,576 ms | 2,118 ms |\n| MySQL, 30 people at once | 22,968 ms | 713 ms |\n\nA formula or a lookup has no column to index: a sort on one costs a sort of the\ntable for every block, so a view with many people on it is better sorted on a stored\nfield. A filter narrows what is sorted, and an index that starts with the filtered\nfield and ends with the sorted one serves both.\n\n### What the database cannot run\n\nA view never goes unanswered. When a filter, sort, group or total needs a field\nthe data source cannot compute, or the source does not acknowledge a filter, or\nit fails on a computed field, the page reads the table a block at a time, works\nthe field out with the same code the cells use, and keeps the ids and the values\nthe request needs. The view then reads its records by id. The tab bar shows how\nfar it has got.\n\nIt gives what the cells say, on any backend, and it is slower than the database:\nabout a minute for a lookup over the 1,000,000 records of the demo on a dev\nserver. It is read once. Scrolling, sorting again and the lanes of a board cost\nnothing more, a record you or someone else changes is updated where it stands,\nand changing a field reads again. Taking the sort or filter away stops the read.\nRecords changing while it reads are read again once, and if the count still moves\nthe base says so.\n\nTell the base what your source can work out with `computes`, and that field stays\nin the database. Without it, only `createSqlDataSource` is asked: a `queryRecords`\nbackend needs `computes: () => true` or it is right but slow.\n\n```ts\nimport { baseSource } from '@svgrid/enterprise/base'\nimport type { ServerDataSource } from '@svgrid/grid'\n\ndeclare const orders: ServerDataSource<Record<string, unknown>>\n\nconst source = baseSource(orders, { computes: (def) => def.field === 'tax' })\n```\n\n`@svgrid/enterprise/base/compute` has both halves:\n[`queryRecords`](./server-compute.md#answer-a-paged-table) for records a\nserver holds, and [`fieldFilterToSql`](./server-compute.md#the-views-filter-in-sql)\nfor a database.\n"
5203
5203
  },
5204
5204
  {
5205
5205
  "slug": "help/base/forms",
@@ -5234,7 +5234,7 @@ export const docs = [
5234
5234
  "path": "docs/help/base/permissions.md",
5235
5235
  "title": "Permissions: what each person may do",
5236
5236
  "section": "SvGrid Base",
5237
- "markdown": "# Permissions: what each person may do\n\nYour app already knows its roles: who is an editor, who only comments, who\nsees the margin column. `permissions` hands that to the base, and the screen\noffers only what the signed-in person may do. A commenter gets no \"+\" row and\nno Delete in the record menu; a sales rep never sees the Cost field, or the\nformula that reads it.\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, setCurrentUser, type BaseTable, type BasePermissions } from '@svgrid/enterprise/base'\n\n enableBase()\n let { data } = $props()\n setCurrentUser(data.user)\n\n let tables = $state<BaseTable[]>(data.tables)\n // From your server: the role of the person signed in.\n const permissions: BasePermissions =\n data.user.role === 'editor'\n ? { canEditSchema: false }\n : { canCreate: false, canDelete: false, canEditSchema: false, canEditViews: false, fields: { 'deals.cost': 'hidden' } }\n</script>\n\n<SvBase bind:tables {permissions} editable fieldMenu columnAdder expandRows />\n```\n\nEverything is allowed by default. Each key turns one thing off:\n\n| Permission | Off means |\n|---|---|\n| `canEdit` | No cell, record view, card drag or date drag changes a value. Adding and deleting records are off too. |\n| `canCreate` | No \"+\" row, no group \"+\", no Insert record above or below, no Duplicate record, no \"+\" on board lanes or calendar days, no form views, no \"Import records from a file\". |\n| `canDelete` | No Delete record in the row and card menus, no Delete in the selection bar, no trash button. |\n| `canEditSchema` | No field editor or \"+\" field column, no Add, Rename, Duplicate or Delete table, no \"Import a file as a table\", no Automations button, no SQL export. The field menu keeps Filter, Group, Hide and Color by. |\n| `canEditViews` | No Create, Rename, Duplicate, Delete, Lock or Unlock on shared views. Filters, sorts and groups the person changes stay in their page; they may still make personal views, which are saved. |\n| `canComment` | Comments are read, not written: no comment box, no Comment in the menus. |\n| `fields` | Per field: `'hidden'` or `'readOnly'`. A key is a column id (in every table) or `tableId.columnId`. |\n| `viewsOnly` | Only these views (a view id, or `tableId.viewId`). Tables with none of them are not shown, the field editor is off, and view changes stay in the page. |\n\n## A table's own permissions\n\n`BaseTable.permissions` wins over the base's, key by key, and its `fields`\nare merged over the base's `fields`:\n\n```ts\nconst tables: BaseTable[] = [\n // Everyone may add deals; only the base's permissions decide the rest.\n { id: 'deals', name: 'Deals', columns, data, permissions: { canCreate: true } },\n // The price list is read-only for the person signed in.\n { id: 'prices', name: 'Prices', columns: priceColumns, data: prices, permissions: { canEdit: false } },\n]\n```\n\n## Hidden fields\n\nA hidden field leaves the grid, the record view, cards, search and lookups\nfrom other tables. So does every field that would show what it holds:\n\n- a formula that reads it (`{Amount} - {Cost}` goes with Cost),\n- a lookup or rollup of it in another table,\n- a link to a table the person cannot see, and the lookups that read that link.\n\nThe records keep the values. When the person edits a field, the stored record\nis the same object with Cost still in it, so the database never sees Cost\ncleared. A field change in the field editor comes back with the hidden fields\nin their places, and the columns your app stores never take the\n`editable: false` the screen uses for read-only fields.\n\nHiding is for display. A page that loads the whole table has the values in\nmemory, and a person with the browser's developer tools can read them. For a\nfield that is secret, leave it out of what your server sends, as\n`sharedTables` does for a [read-only share](./collaboration.md#share-a-base-read-only).\n\n## Views a person may change\n\nWithout `canEditViews` the views menu shows \"Create a personal view\" and\nRename, Duplicate and Delete only on the person's own personal views (those\nneed `setCurrentUser`). A filter or sort they set on a shared view shows in\ntheir page and is not saved; `onTablesChange` hands your app the views as\nstored, with their personal views added.\n\nA locked view keeps its settings for everyone. With `canEditViews`, a person\nmay unlock it; without it, they may not.\n\n## A narrow window: `viewsOnly`\n\nA support agent who needs the \"Open tickets\" view and nothing else:\n\n```svelte\n<SvBase bind:tables permissions={{ viewsOnly: ['tickets.open'], canDelete: false }} editable expandRows />\n```\n\nThey see the Tickets tab with one view, edit records in it, and nothing else\nof the base. Their filter and sort changes stay in their page.\n\n## One table\n\nWith `data` and `columns` instead of `tables`, `permissions` works the same.\n`onColumnAdd`, `onColumnChange` and `onColumnRemove` hand your app the\ncolumns as you declared them, hidden fields included.\n\n## In code\n\n```ts\nimport { permittedRecords, tablePermissions, gridPermissions } from '@svgrid/enterprise/base'\n\n// The tables as one person sees them, the hidden fields' values gone from the\n// records: on your server, to send only those.\nconst shown = permittedRecords(tables, permissions)\n\n// Every permission decided for a table (defaults filled in).\ntablePermissions({ canEdit: false }) // { canEdit: false, canCreate: false, canDelete: false, ... }\n\n// What the person may do in a grid of the base: for your own buttons and cells.\nif (gridPermissions(ctrl).canDelete) showDeleteButton()\n```\n\n`permittedTables` gives the same tables with the records as they were, hidden\nvalues included, for code in the page that writes them back. Do not send its\nresult to anyone.\n\n`<sv-base>` takes the same object as its `permissions` property or as JSON in\nthe `permissions` attribute; see [`<sv-base>`](../web-components/sv-base.md).\n\n## Pitfalls\n\n- **The browser follows these; your server decides.** A base on a database\n still needs the API or row-level security to refuse a delete from a role\n that may not delete. The screen not offering it is not the same as the\n server refusing it.\n- **`canCreate` and `canDelete` need `canEdit`.** A person who may not change\n values cannot add or delete records either.\n- **`readOnly` wins.** A base shown with `readOnly` (a share link) is read-only\n whatever `permissions` says.\n- **`fields` keys are column ids.** A field's header can change; its id\n does not. Use `tableId.columnId` when two tables have a field with the same id.\n\n## Next\n\n- [People, comments and files](./collaboration.md): who is signed in, protected\n fields, a read-only share.\n- [Views](./views.md): personal and locked views.\n- [Formulas on a server](./server-compute.md): the values people see, worked\n out in Node for an API or an AI tool.\n"
5237
+ "markdown": "# Permissions: what each person may do\n\nYour app already knows its roles: who is an editor, who only comments, who\nsees the margin column. `permissions` hands that to the base, and the screen\noffers only what the signed-in person may do. A commenter gets no \"+\" row and\nno Delete in the record menu; a sales rep never sees the Cost field, or the\nformula that reads it.\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, setCurrentUser, type BaseTable, type BasePermissions } from '@svgrid/enterprise/base'\n\n enableBase()\n let { data } = $props()\n setCurrentUser(data.user)\n\n let tables = $state<BaseTable[]>(data.tables)\n // From your server: the role of the person signed in.\n const permissions: BasePermissions =\n data.user.role === 'editor'\n ? { canEditSchema: false }\n : { canCreate: false, canDelete: false, canEditSchema: false, canEditViews: false, fields: { 'deals.cost': 'hidden' } }\n</script>\n\n<SvBase bind:tables {permissions} editable fieldMenu columnAdder expandRows />\n```\n\nEverything is allowed by default. Each key turns one thing off:\n\n| Permission | Off means |\n|---|---|\n| `canEdit` | No cell, record view, card drag or date drag changes a value. Adding and deleting records are off too. |\n| `canCreate` | No \"+\" row, no group \"+\", no Insert record above or below, no Duplicate record, no \"+\" on board lanes or calendar days, no form views, no \"Import records from a file\". |\n| `canDelete` | No Delete record in the row and card menus, no Delete in the selection bar, no trash button. |\n| `canEditSchema` | No field editor or \"+\" field column, no Add, Rename, Duplicate or Delete table, no \"Import a file as a table\", no Automations button, no SQL export. The field menu keeps Filter, Group, Hide and Color by. |\n| `canEditViews` | No Create, Rename, Duplicate, Delete, Lock or Unlock on shared views. Filters, sorts and groups the person changes stay in their page; they may still make personal views, which are saved. |\n| `canComment` | Comments are read, not written: no comment box, no Comment in the menus. |\n| `fields` | Per field: `'hidden'` or `'readOnly'`. A key is a column id (in every table) or `tableId.columnId`. |\n| `viewsOnly` | Only these views (a view id, or `tableId.viewId`). Tables with none of them are not shown, the field editor is off, and view changes stay in the page. |\n\n## A table's own permissions\n\n`BaseTable.permissions` wins over the base's, key by key, and its `fields`\nare merged over the base's `fields`:\n\n```ts\nconst tables: BaseTable[] = [\n // Everyone may add deals; only the base's permissions decide the rest.\n { id: 'deals', name: 'Deals', columns, data, permissions: { canCreate: true } },\n // The price list is read-only for the person signed in.\n { id: 'prices', name: 'Prices', columns: priceColumns, data: prices, permissions: { canEdit: false } },\n]\n```\n\n## Hidden fields\n\nA hidden field leaves the grid, the record view, cards, search and lookups\nfrom other tables. So does every field that would show what it holds:\n\n- a formula that reads it (`{Amount} - {Cost}` goes with Cost),\n- a lookup or rollup of it in another table,\n- a link to a table the person cannot see, and the lookups that read that link.\n\nThe records keep the values. When the person edits a field, the stored record\nis the same object with Cost still in it, so the database never sees Cost\ncleared. A field change in the field editor comes back with the hidden fields\nin their places, and the columns your app stores never take the\n`editable: false` the screen uses for read-only fields.\n\nHiding is for display. A page that loads the whole table has the values in\nmemory, and a person with the browser's developer tools can read them. For a\nfield that is secret, leave it out of what your server sends, as\n`sharedTables` does for a [read-only share](./collaboration.md#share-a-base-read-only).\n\n## Views a person may change\n\nWithout `canEditViews` the views menu shows \"Create a personal view\" and\nRename, Duplicate and Delete only on the person's own personal views (those\nneed `setCurrentUser`). A filter or sort they set on a shared view shows in\ntheir page and is not saved; `onTablesChange` hands your app the views as\nstored, with their personal views added.\n\nA locked view keeps its settings for everyone. With `canEditViews`, a person\nmay unlock it; without it, they may not.\n\n## A narrow window: `viewsOnly`\n\nA support agent who needs the \"Open tickets\" view and nothing else:\n\n```svelte\n<SvBase bind:tables permissions={{ viewsOnly: ['tickets.open'], canDelete: false }} editable expandRows />\n```\n\nThey see the Tickets tab with one view, edit records in it, and nothing else\nof the base. Their filter and sort changes stay in their page.\n\n## A view as a page of an app: `chrome=\"page\"`\n\nFor an app built on the base, where each page is one view (an interface, a\nportal), drop the base's own chrome:\n\n```svelte\n<SvBase bind:tables chrome=\"page\" activeTable=\"tickets\" permissions={{ viewsOnly: ['tickets.open'], canEditSchema: false }} editable />\n```\n\n`chrome=\"page\"` shows the view alone: no table tabs, no views menu. The rest\nof the view bar stays (fields, filter, sort, group, search), so people still\nnarrow what they see; with `canEditViews: false` that stays in their page,\nand a dashboard offers no \"Edit dashboard\". The default, `chrome=\"base\"`, is\nthe base as it always was.\n\n## One table\n\nWith `data` and `columns` instead of `tables`, `permissions` works the same.\n`onColumnAdd`, `onColumnChange` and `onColumnRemove` hand your app the\ncolumns as you declared them, hidden fields included.\n\n## In code\n\n```ts\nimport { permittedRecords, tablePermissions, gridPermissions } from '@svgrid/enterprise/base'\n\n// The tables as one person sees them, the hidden fields' values gone from the\n// records: on your server, to send only those.\nconst shown = permittedRecords(tables, permissions)\n\n// Every permission decided for a table (defaults filled in).\ntablePermissions({ canEdit: false }) // { canEdit: false, canCreate: false, canDelete: false, ... }\n\n// What the person may do in a grid of the base: for your own buttons and cells.\nif (gridPermissions(ctrl).canDelete) showDeleteButton()\n```\n\n`permittedTables` gives the same tables with the records as they were, hidden\nvalues included, for code in the page that writes them back. Do not send its\nresult to anyone.\n\n`<sv-base>` takes the same object as its `permissions` property or as JSON in\nthe `permissions` attribute; see [`<sv-base>`](../web-components/sv-base.md).\n\n## Pitfalls\n\n- **The browser follows these; your server decides.** A base on a database\n still needs the API or row-level security to refuse a delete from a role\n that may not delete. The screen not offering it is not the same as the\n server refusing it.\n- **`canCreate` and `canDelete` need `canEdit`.** A person who may not change\n values cannot add or delete records either.\n- **`readOnly` wins.** A base shown with `readOnly` (a share link) is read-only\n whatever `permissions` says.\n- **`fields` keys are column ids.** A field's header can change; its id\n does not. Use `tableId.columnId` when two tables have a field with the same id.\n\n## Next\n\n- [People, comments and files](./collaboration.md): who is signed in, protected\n fields, a read-only share.\n- [Views](./views.md): personal and locked views.\n- [Formulas on a server](./server-compute.md): the values people see, worked\n out in Node for an API or an AI tool.\n"
5238
5238
  },
5239
5239
  {
5240
5240
  "slug": "help/base/postgres",
@@ -5255,7 +5255,7 @@ export const docs = [
5255
5255
  "path": "docs/help/base/rest-api.md",
5256
5256
  "title": "Build a base on your API",
5257
5257
  "section": "SvGrid Base",
5258
- "markdown": "# Build a base on your API\n\nWhen your data sits behind an API you already have, a base can use it as it\nis: it reads a table with one `GET`, and writes each change with `POST`,\n`PATCH` or `DELETE`. This tutorial builds a small CRM, Deals, Companies and\nContacts, from the API's own description of its tables.\n\n<div data-docs-demo=\"518-base-rest-api\" data-height=\"720\"></div>\n\n![A CRM base over a JSON API. Under the base, the request log: a PATCH of a deal answered with 422 and the reason, and the GET of each table when the base opened.](/docs-media/base-rest-api.png)\n\n## 1. What the API answers\n\nFor each table, four routes. The shapes below are the defaults; the next\nsection shows how to read others.\n\n| Request | Answer |\n|---|---|\n| `GET /api/deals?offset=0&limit=10000` | `200` with `{ \"rows\": [...], \"rowCount\": 5 }`, or a plain array |\n| `POST /api/deals` with the new record | `201` with the record as stored, its `id` included |\n| `PATCH /api/deals/d3` with the changed fields only | `200` with the record as stored |\n| `DELETE /api/deals/d3` | `204` |\n\nA foreign key is a plain field: a deal has `company_id`. To refuse a write,\nanswer with a `4xx` status and `{ \"error\": \"Amount must be 0 or more\" }`\n(or `{ \"message\": ... }`); the base puts the cell back and shows that sentence. A\n`400`, `409` or `422` is a refusal of the values, so it offers no Try again.\n\n## 2. Read another shape\n\n`createRestDataSource` takes `buildQuery` (the query string for a read) and\n`parse` (the rows and count from the answer). For an API that pages with\n`page` and `per_page` and answers `{ \"data\": [...], \"meta\": { \"total\": 5 } }`:\n\n```ts\nimport { createRestDataSource } from '@svgrid/enterprise'\n\nconst deals = createRestDataSource({\n url: 'https://api.example.com/v2/deals',\n headers: () => ({ authorization: `Bearer ${localStorage.getItem('token') ?? ''}` }),\n buildQuery: (request) => ({ page: '1', per_page: String(request.endRow - request.startRow) }),\n parse: (body) => {\n const b = body as { data: Record<string, unknown>[]; meta: { total: number } }\n return { rows: b.data, rowCount: b.meta.total }\n },\n})\n```\n\n`offsetLimitAdapter`, `jsonServerAdapter` and `dummyJsonAdapter` are\nready-made pairs for common shapes.\n\n## 3. Describe the tables\n\n`baseTablesFromSchemas` makes base tables from a description of each table:\nits columns and their types. Write the description out, or read it from an\nOpenAPI document with `introspectOpenApi(spec).entities`:\n\n```ts\n// src/lib/crm.ts\nimport { baseTablesFromSchemas, type BaseTable } from '@svgrid/enterprise/base'\nimport { createRestDataSource, type EntitySchema } from '@svgrid/enterprise'\n\nconst STAGES = ['Qualified', 'Proposal', 'Negotiation', 'Won', 'Lost'].map((value) => ({ value, label: value }))\n\nconst SCHEMAS: EntitySchema[] = [\n {\n name: 'deals',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'title', type: 'text' },\n { field: 'stage', type: 'enum', options: STAGES },\n { field: 'amount', type: 'number' },\n { field: 'close_date', type: 'date', label: 'Close date' },\n { field: 'company_id', type: 'relation', relation: { entity: 'companies', labelField: 'name' } },\n ],\n },\n {\n name: 'companies',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'name', type: 'text' },\n { field: 'website', type: 'text', format: 'url' },\n ],\n },\n {\n name: 'contacts',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'name', type: 'text' },\n { field: 'email', type: 'text', format: 'email' },\n { field: 'company_id', type: 'relation', relation: { entity: 'companies', labelField: 'name' } },\n ],\n },\n]\n\nexport function crmTables(): BaseTable[] {\n return baseTablesFromSchemas(\n SCHEMAS,\n (schema) => createRestDataSource({ url: `/api/${schema.name}`, schema }),\n // The API picks the id of a new record.\n () => ({ serverIds: true }),\n )\n}\n```\n\nWhat comes out:\n\n| In the description | In the base |\n|---|---|\n| `text`, with `format: 'url'` or `'email'` | Text, URL or Email field |\n| `number` | Number field |\n| `date`, `datetime` | Date field, with the time for `datetime` |\n| `enum` with options | Single select with those options |\n| `boolean` | Checkbox |\n| `relation` (`company_id`) | A Company link, stored as `company_id`, and a field in Companies that links back |\n| The primary key | The record's id, not a field |\n\nThe first text field is each table's primary field. Companies gets a Deals\nfield and a Contacts field that link back; they are worked out in the page\nand never sent to the API.\n\n## 4. Adjust the fields\n\nThe tables are plain objects. Show Amount as currency with a total, and give\nDeals a Kanban view by stage:\n\n```ts\nimport type { BaseTable } from '@svgrid/enterprise/base'\nimport { crmTables } from '$lib/crm'\n\nconst tables: BaseTable[] = crmTables().map((t) =>\n t.id !== 'deals'\n ? t\n : {\n ...t,\n columns: t.columns.map((c) =>\n c.field === 'amount' ? { ...c, fieldProps: { format: 'currency', symbol: '$', precision: 0, summary: 'sum' } } : c,\n ),\n views: [\n { id: 'grid', name: 'All deals', type: 'grid', color: { mode: 'field', field: 'stage' } },\n { id: 'pipeline', name: 'Pipeline', type: 'kanban', stackBy: 'stage' },\n ],\n },\n)\n```\n\n## 5. The page\n\n```svelte\n<script lang=\"ts\">\n import { SvToaster } from '@svgrid/grid'\n import { SvBase, enableBase } from '@svgrid/enterprise/base'\n import { crmTables } from '$lib/crm'\n\n enableBase()\n let tables = $state.raw(crmTables())\n</script>\n\n<div style=\"height: 640px\">\n <SvBase bind:tables editable sortable summary containerHeight=\"100%\" />\n</div>\n<SvToaster />\n```\n\nOpen it and the base makes one `GET` per table. Change a stage and it sends\n`PATCH /api/deals/d3` with `{ \"stage\": \"Won\" }`. Link a contact to a company\nfrom the Companies tab and it sends `PATCH /api/contacts/p2` with\n`{ \"company_id\": \"c1\" }`.\n\n## Next\n\n- [Keep a base in a database](./data.md): links, record ids, changes made\n elsewhere and every option.\n- [Build a base on Supabase](./supabase.md) and\n [on Postgres](./postgres.md).\n"
5258
+ "markdown": "# Build a base on your API\n\nWhen your data sits behind an API you already have, a base can use it as it\nis: it reads a table with one `GET`, and writes each change with `POST`,\n`PATCH` or `DELETE`. This tutorial builds a small CRM, Deals, Companies and\nContacts, from the API's own description of its tables.\n\n<div data-docs-demo=\"518-base-rest-api\" data-height=\"720\"></div>\n\n![A CRM base over a JSON API. Under the base, the request log: a PATCH of a deal answered with 422 and the reason, and the GET of each table when the base opened.](/docs-media/base-rest-api.png)\n\n## 1. What the API answers\n\nFor each table, four routes. The shapes below are the defaults; the next\nsection shows how to read others.\n\n| Request | Answer |\n|---|---|\n| `GET /api/deals?offset=0&limit=10000` | `200` with `{ \"rows\": [...], \"rowCount\": 5 }`, or a plain array |\n| `POST /api/deals` with the new record | `201` with the record as stored, its `id` included |\n| `PATCH /api/deals/d3` with the changed fields only | `200` with the record as stored |\n| `DELETE /api/deals/d3` | `204` |\n\nA view with an advanced filter (conditions in and/or groups) adds `filter`,\nthe filter as JSON. Apply it in full and answer `\"appliedExpression\": true`\nwith the rows; otherwise leave that out, and the base knows the filter did\nnot run on the server.\n\nA foreign key is a plain field: a deal has `company_id`. To refuse a write,\nanswer with a `4xx` status and `{ \"error\": \"Amount must be 0 or more\" }`\n(or `{ \"message\": ... }`); the base puts the cell back and shows that sentence. A\n`400`, `409` or `422` is a refusal of the values, so it offers no Try again.\n\n## 2. Read another shape\n\n`createRestDataSource` takes `buildQuery` (the query string for a read) and\n`parse` (the rows and count from the answer). For an API that pages with\n`page` and `per_page` and answers `{ \"data\": [...], \"meta\": { \"total\": 5 } }`:\n\n```ts\nimport { createRestDataSource } from '@svgrid/enterprise'\n\nconst deals = createRestDataSource({\n url: 'https://api.example.com/v2/deals',\n headers: () => ({ authorization: `Bearer ${localStorage.getItem('token') ?? ''}` }),\n buildQuery: (request) => ({ page: '1', per_page: String(request.endRow - request.startRow) }),\n parse: (body) => {\n const b = body as { data: Record<string, unknown>[]; meta: { total: number } }\n return { rows: b.data, rowCount: b.meta.total }\n },\n})\n```\n\n`offsetLimitAdapter`, `jsonServerAdapter` and `dummyJsonAdapter` are\nready-made pairs for common shapes.\n\n## 3. Describe the tables\n\n`baseTablesFromSchemas` makes base tables from a description of each table:\nits columns and their types. Write the description out, or read it from an\nOpenAPI document with `introspectOpenApi(spec).entities`:\n\n```ts\n// src/lib/crm.ts\nimport { baseTablesFromSchemas, type BaseTable } from '@svgrid/enterprise/base'\nimport { createRestDataSource, type EntitySchema } from '@svgrid/enterprise'\n\nconst STAGES = ['Qualified', 'Proposal', 'Negotiation', 'Won', 'Lost'].map((value) => ({ value, label: value }))\n\nconst SCHEMAS: EntitySchema[] = [\n {\n name: 'deals',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'title', type: 'text' },\n { field: 'stage', type: 'enum', options: STAGES },\n { field: 'amount', type: 'number' },\n { field: 'close_date', type: 'date', label: 'Close date' },\n { field: 'company_id', type: 'relation', relation: { entity: 'companies', labelField: 'name' } },\n ],\n },\n {\n name: 'companies',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'name', type: 'text' },\n { field: 'website', type: 'text', format: 'url' },\n ],\n },\n {\n name: 'contacts',\n fields: [\n { field: 'id', type: 'text', primaryKey: true },\n { field: 'name', type: 'text' },\n { field: 'email', type: 'text', format: 'email' },\n { field: 'company_id', type: 'relation', relation: { entity: 'companies', labelField: 'name' } },\n ],\n },\n]\n\nexport function crmTables(): BaseTable[] {\n return baseTablesFromSchemas(\n SCHEMAS,\n (schema) => createRestDataSource({ url: `/api/${schema.name}`, schema }),\n // The API picks the id of a new record.\n () => ({ serverIds: true }),\n )\n}\n```\n\nWhat comes out:\n\n| In the description | In the base |\n|---|---|\n| `text`, with `format: 'url'` or `'email'` | Text, URL or Email field |\n| `number` | Number field |\n| `date`, `datetime` | Date field, with the time for `datetime` |\n| `enum` with options | Single select with those options |\n| `boolean` | Checkbox |\n| `relation` (`company_id`) | A Company link, stored as `company_id`, and a field in Companies that links back |\n| The primary key | The record's id, not a field |\n\nThe first text field is each table's primary field. Companies gets a Deals\nfield and a Contacts field that link back; they are worked out in the page\nand never sent to the API.\n\n## 4. Adjust the fields\n\nThe tables are plain objects. Show Amount as currency with a total, and give\nDeals a Kanban view by stage:\n\n```ts\nimport type { BaseTable } from '@svgrid/enterprise/base'\nimport { crmTables } from '$lib/crm'\n\nconst tables: BaseTable[] = crmTables().map((t) =>\n t.id !== 'deals'\n ? t\n : {\n ...t,\n columns: t.columns.map((c) =>\n c.field === 'amount' ? { ...c, fieldProps: { format: 'currency', symbol: '$', precision: 0, summary: 'sum' } } : c,\n ),\n views: [\n { id: 'grid', name: 'All deals', type: 'grid', color: { mode: 'field', field: 'stage' } },\n { id: 'pipeline', name: 'Pipeline', type: 'kanban', stackBy: 'stage' },\n ],\n },\n)\n```\n\n## 5. The page\n\n```svelte\n<script lang=\"ts\">\n import { SvToaster } from '@svgrid/grid'\n import { SvBase, enableBase } from '@svgrid/enterprise/base'\n import { crmTables } from '$lib/crm'\n\n enableBase()\n let tables = $state.raw(crmTables())\n</script>\n\n<div style=\"height: 640px\">\n <SvBase bind:tables editable sortable summary containerHeight=\"100%\" />\n</div>\n<SvToaster />\n```\n\nOpen it and the base makes one `GET` per table. Change a stage and it sends\n`PATCH /api/deals/d3` with `{ \"stage\": \"Won\" }`. Link a contact to a company\nfrom the Companies tab and it sends `PATCH /api/contacts/p2` with\n`{ \"company_id\": \"c1\" }`.\n\n## Next\n\n- [Keep a base in a database](./data.md): links, record ids, changes made\n elsewhere and every option.\n- [Build a base on Supabase](./supabase.md) and\n [on Postgres](./postgres.md).\n"
5259
5259
  },
5260
5260
  {
5261
5261
  "slug": "help/base/review-queue",
@@ -5297,21 +5297,21 @@ export const docs = [
5297
5297
  "path": "docs/help/base/undo-trash-import.md",
5298
5298
  "title": "Undo, the trash and importing files",
5299
5299
  "section": "SvGrid Base",
5300
- "markdown": "# Undo, the trash and importing files\n\nThree things people reach for in their first week with a base: Ctrl+Z after\na slip, getting back a record someone deleted yesterday, and pulling in the\nspreadsheet they kept until now. All three work on `<SvBase tables>` with no\nsetup; this page says what each one does and how an app keeps it.\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\n## Undo and redo\n\nA step undoes what it did and nothing else: what teammates changed since stays, whether\nthe change came through a live feed or a read of the table. A cell someone changed after\nyou keeps their value, on undo and on redo, cell by cell in a paste, and a record you added\nthat someone has worked on since stays, and the page says so. Delete it if it should go.\nFor a range the page says how many: \"Someone has changed 1 of these records since, so it\nkeeps their value.\" once all its blocks are written, or \"1 of these cells\" for a small range\nundone cell by cell.\nA cell undo on a table in a database reads the record first, with a live feed or without\none, so this holds for a change the page never saw; a date with a time is compared as the\npage last heard of it. When the record cannot be read, or a write of it has not answered\nyet, the undo is refused with a message and the step stays the next one to try. A number\nthe column rounded (1.234 in a column of two decimals) counts as what you typed. Moving a\nrecord by drag is a step too.\n\nOn a table in a database one Undo or Redo runs at a time. While one is still checking or\nwriting, another press changes nothing and says \"Undo is still running.\" (or \"Redo is still\nrunning.\"); the buttons show it busy, and leaving the page asks first. One that writes more\nthan one block (1,000 records, or 1,000 cells of a table the page holds) runs in the dialog a\nbig paste has, with its count and Stop: a range, Edit fields, a delete, the records a paste\nadded, and cells. Stopped, an Undo, or a Redo of cells or of added records, stays the next\nstep, and pressing it again writes only what is left; a Redo of a delete that is stopped\nkeeps the records it deleted, and Undo brings those back. A turn that writes a record the\nview no longer shows is done when that write has answered. What an Undo or Redo did is said\nonce it is done, with the number it changed: cells someone else changed, records a teammate\nrestored first or that were deleted forever, records out of reach and writes the database\nrefused are not counted. A paste of cells the grid holds says its count the same way, once\nthe database has answered.\n\nA record you add to a table read a part at a time has its \"add record\" step from the\nmoment its row is there. What you type into it is part of that step. An Undo before the\ndatabase has stored it waits for the insert, then takes the record out; a new record that\nwas never stored (left empty, or refused) just goes, with no Redo. Its activity starts\nonce it is stored: a refused record has none.\n\nAn undo or a redo puts values back. It runs no automation, and the row returns to its\nplace in the view at once.\n\nAn undo of a delete that fails (the network dropped, the database refused) leaves the\nrecord in the trash and the step in place: press Ctrl+Z again, or restore it from the\ntrash.\n\nCtrl+Z (Cmd+Z on a Mac) undoes what was done to the shown table, in the\norder it was done. Ctrl+Shift+Z or Ctrl+Y redoes it. The keys work anywhere\nin the base, and also with nothing focused on the page (after a dialog\ncloses, or a click beside the base): then the base used last takes them. A\nkey typed in a field of your own page stays yours. One history holds:\n\n- cell edits, pastes, fills and Edit fields on a selection, each one step however many\n cells it wrote (the grid's own history). Edit fields on more than 10,000 records of a\n table read a part at a time has no step: it says so and asks first. If it is stopped or\n fails with 10,000 or fewer changed, those have a step. A paste, Delete or fill over a large\n range goes the same way, and its undo reads and writes in blocks too;\n- records added, deleted or imported from a file;\n- fields added, changed (a new type, new options) or deleted;\n- tables added, renamed, duplicated, deleted or imported from a file.\n\nUndoing a deleted record puts it back at the place it had, and the links the\ndelete cleared come back too: delete a task and its project stops listing\nit; undo, and the project lists it again. The step restores the records of\nevery table the change touched.\n\nUndoing a deleted table brings it back as a tab, with its records and the\nfields of other tables that linked to it, and takes it out of the trash.\nUndoing a table you added takes it away again; the step lives on the new\ntable's history, so there is no redo once it is gone.\n\nThe Undo and Redo buttons at the end of the tab bar say what the next press\ndoes (\"Undo delete 2 records\"). `undoRedo={false}` hides them; the keys still\nwork. Each table keeps its history while another table is shown, so going\nback to a table carries on where you left it. View settings (a filter, a\nsort) are not part of the history: a view keeps them as you leave it.\n\n<div data-docs-demo=\"705-base-undo-redo\" data-height=\"620\"></div>\n\n## Soft delete\n\nUndo of a delete writes the record back. Point `softDelete` at a timestamp column and it\ndoes not have to: a delete sets the column and leaves the row, and a restore clears it.\nNothing about the record is rewritten, so columns off the schema, exact times and big\nnumbers come back as they were, and what a teammate wrote before the delete stays.\n\n```sql\nALTER TABLE orders ADD COLUMN deleted_at timestamptz;\n```\n\n```ts\nconst source = createSqlDataSource({ schema, table: 'orders', softDelete: 'deleted_at', dialect, execute })\n```\n\n`createSupabaseDataSource` takes the same option, and so do `createBaseEndpoint`,\n`baseFromDatabase` and `baseFromSupabase`:\n\n```ts\nconst endpoint = createBaseEndpoint({ execute, tables: ['orders', 'tags'], authorize, softDelete: 'deleted_at' })\n```\n\nPer table it is a function: `softDelete: (table) => (table === 'orders' ? 'removed_at' : null)`.\n\nThe string form skips a table without the column. The function form names a column that\nmust be there. Those three keep the column out of the fields. By hand, leave it off the\nschema: no write sets it either way, so a client cannot create a record already deleted or\nmark one through an update, and `createKitHandlers` answers `422` to an update of it alone.\nReads, counts, groups, totals, lookups and rollups skip marked records, and a write to one\nis told the record is gone.\n\n- The trash of a table with `softDelete` reads the table: every marked row is there, after\n a reload, for whoever may delete, whoever marked it. It lists the latest 1,000 per table\n and says how many there are. No trash store is needed for those records.\n- A record's links through a junction table stay. A delete leaves the junction rows, at\n either end of the link, a marked record shows in no link, and a restore finds them.\n- A foreign key in another table stays too. A delete writes nothing to that table, the\n record shows in no link while it is marked, and a restore finds its links where they were,\n whoever restores it and after any reload. A link a teammate changed in the meantime stays\n as they left it. While the record is marked the link reads as empty everywhere: \"is empty\"\n and \"is not empty\", one Empty group, sort, count and the link picker. The key stays in the\n column, so a query of your own still sees it. A link written to a marked record, from a\n picker opened before the delete, is stored, shows empty, and the person is told why.\n- Delete forever and Empty trash are final. Each asks first, the item stays in the trash\n until the database has confirmed, and if it refuses (another table still needs the record)\n the person is told and the record can still be restored. In one Empty trash the others go.\n Junction rows go with the record, and the foreign keys that point at it are emptied with it,\n in the tables of this base: see [Delete forever is one step](#delete-forever-is-one-step). A\n `NOT NULL` key cannot be emptied, so the record stays and the person is told other records\n depend on it. A row itself marked deleted keeps its key: a foreign key constraint refuses\n the purge until that row has gone. Undo cannot bring a deleted record back, in that window\n or another. An undo to a link whose record was deleted forever says so, and offers no Try again.\n- A trash item reaching `trashDays` is deleted for real only when the page's trash holds\n it, that is, when a trash store or `bind:trash` kept it. A row found marked in the table\n stays until someone deletes it, so opening a page never purges what another tool marked.\n For a limit on those, run it on a schedule:\n `DELETE FROM orders WHERE deleted_at < now() - interval '30 days'`.\n- A delete too big for the trash marks and purges a block of 1,000 at a time. A page closed\n half way leaves at most the block under way marked, and it shows in the trash.\n- A marked row still holds its value in a UNIQUE column, and still counts for a foreign key.\n- On Supabase a delete is an UPDATE to row-level security, so write the policy as one.\n- Autonumber reads the highest number over every row, marked ones too, on every source, so\n a marked record keeps its number and a new record takes the next.\n- A trash store item from before `softDelete` was turned on has no mark to clear: Restore\n says it was deleted forever.\n\n### Delete forever is one step\n\nEmptying the keys that hold a record and deleting it are several writes. Which of them can\nstop half way depends on how the base reaches the database:\n\n| The base is on | Delete forever is |\n|---|---|\n| `createBaseEndpoint` with `transaction` | One request and one transaction: all of it or none of it |\n| `createBaseEndpoint` without | One request: the server runs the steps, and a closed page changes nothing |\n| `createKitHandlers` with `linkedFrom` | The same two, by whether the sources share a `transaction` |\n| `baseFromDatabase` with `transaction` | One transaction, for the tables that share it |\n| `createKitHandlers` without `linkedFrom`, `baseFromDatabase` without `transaction` | Steps the page runs, over several requests |\n| Supabase and PostgREST | Steps the page runs. PostgREST has no transaction across requests |\n\nWhere it is steps, they are safe in this order. First the record's mark changes to\n`1970-01-03`, which is a claim: from then on no Restore or undo reaches it, by any path.\nThen the keys go, each only where it still holds this record, then the record. A failure\nputs the keys back, only where a key is still empty, so a link a teammate made meanwhile\nstays, and gives the claim up, with the mark set to that moment. If that fails too, or the\npage closes, the claim stays: the record stays in the trash marked as not finished, Restore\nsays why it cannot, and Delete forever finishes the job. Its links may be gone by then, so\nit does not come back with fewer than it had. A purge never reports a record it did not delete.\n\nGive the `transaction` next to `execute`, the same function for every table of a database. It\nruns `run` in a transaction, hands it an `execute` on that connection, and resolves to what\n`run` resolved to. A throw must roll back.\n\n```ts\n// node-postgres\ntransaction: async (run) => {\n const client = await pool.connect()\n try {\n await client.query('BEGIN')\n const out = await run((sql, params) => client.query(sql, params).then((r) => r.rows))\n await client.query('COMMIT')\n return out\n } catch (e) {\n await client.query('ROLLBACK').catch(() => {})\n throw e\n } finally {\n client.release()\n }\n}\n\n// postgres.js\ntransaction: (run) => sql.begin((tx) => run((text, params) => tx.unsafe(text, params)))\n\n// mysql2\ntransaction: async (run) => {\n const c = await pool.getConnection()\n try {\n await c.beginTransaction()\n const out = await run((sql, params) => c.query(sql, params).then(([rows]) => rows))\n await c.commit()\n return out\n } catch (e) {\n await c.rollback().catch(() => {})\n throw e\n } finally {\n c.release()\n }\n}\n\n// mssql\ntransaction: async (run) => {\n const tx = new mssql.Transaction(pool)\n await tx.begin()\n try {\n const out = await run(async (text, params) => {\n const request = new mssql.Request(tx)\n params.forEach((value, i) => request.input(`p${i + 1}`, value))\n return (await request.query(text)).recordset ?? []\n })\n await tx.commit()\n return out\n } catch (e) {\n await tx.rollback().catch(() => {})\n throw e\n }\n}\n```\n\nOn SQLite run `BEGIN IMMEDIATE`, `run` and `COMMIT` or `ROLLBACK` on the one connection, and\ntake no other write on it meanwhile. A teammate's write to a row the transaction holds waits\nfor it, then lands. Under a transaction a Restore of the record waits too, and finds it gone.\n\nIn the server's step each table's own `authorize`, `scope`, `validate`, hooks and `audit` apply\nto the keys it empties, as the page's requests met them, with `afterUpdate` and the audit only\nfor what stayed emptied. A `beforeUpdate` hook can refuse. What it changes in the patch is not\nwritten here.\n\nWith routes wired by hand, tell each table who holds its keys, and which of its fields lead to a\ntable that marks. Without `liveLinks` a group on such a field has an Empty group for each\ntrashed key.\n\n```ts\nconst orders = createKitHandlers({\n schema: orderSchema,\n source: createSqlDataSource({ ...orderConfig, transaction, liveLinks: { customer_id: { table: 'customers', key: 'id', softDelete: 'deleted_at' } } }),\n})\nconst customers = createKitHandlers({\n schema: customerSchema,\n source: createSqlDataSource({ ...customerConfig, transaction }),\n linkedFrom: [{ handlers: orders, column: 'customer_id' }],\n})\n```\n\n### Index for soft delete\n\n`deleted_at IS NULL` is in every read, so a count or a group can no longer be answered from\nan index that does not hold the column. Scrolling, sorting and filtering a page barely\nnotice; counts and groups do, and MySQL more than Postgres. Measured on a million orders,\nPostgres 16 and MySQL 8.4 on default memory settings:\n\n| Statement, ms | Postgres, plain to soft | MySQL, plain to soft |\n|---|---|---|\n| Count all | 19 to 32 | 55 to 229 |\n| Group by status | 44 to 73 | 112 to 888 |\n| Count through a lookup filter | 14 to 61 | 33 to 465 |\n\nWith a third of 633,000 rows marked, the lookup count on MySQL took about 0.9 s warm and\n13.3 s cold; an index on `(account, deleted_at)` brought it to 18 ms. So add the mark to the\nindexes your views filter and group on, last:\n\n```sql\nCREATE INDEX orders_account_live ON orders (account, deleted_at);\n```\n\nOn Postgres a partial index does the same for less space:\n`CREATE INDEX orders_account_live ON orders (account) WHERE deleted_at IS NULL`. The trash\nreads the latest marks, so index the column by itself on a table with many of them.\n\n## The trash\n\nA deleted view goes there too, with its filter, sorts and settings, and \"Restore\"\nputs it back at its place in the views menu. Undo does the same.\n\nA deleted record, field or table goes to the trash. The trash button at the\nend of the tab bar shows how many items it holds and opens a list, newest\nfirst, with who deleted each one (`setCurrentUser`) and when. Restore puts an\nitem back:\n\n- a record in its table, at the place it had, with its links;\n- a field with its values (the values stay in the records while the field is\n in the trash);\n- a table as a tab, with the fields of other tables that linked to it made\n again from its side.\n\nThe notice that appears after a delete has a Restore button too. A record\nwhose table is in the trash waits until the table is restored; the dialog\nsays so.\n\nThe trash is plain data. Bind it and store it next to the tables:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, type BaseTable, type BaseTrashItem } from '@svgrid/enterprise/base'\n\n enableBase()\n let tables = $state.raw<BaseTable[]>(loadTables())\n let trash = $state<BaseTrashItem[]>(loadTrash())\n</script>\n\n<SvBase bind:tables bind:trash trashDays={30} editable onTablesChange={saveTables} onTrashChange={saveTrash} />\n```\n\nWith neither `onTrashChange` nor a store the trash lives in the page: it goes when the\npage is closed or reloaded, and the panel says so in place of a number of days. If the\napp keeps a `bind:trash` some other way, pass `trashKept` to show the days.\n\nTo keep the trash where the records are, so it survives a reload and every\nperson of the base gets the same one, give it a store. `list` is read when the\nbase opens (a promise is fine), so someone else's deletes appear in the panel when\nthe page is next opened; `add` and `remove` hear each item deleted, restored, undone\nor emptied, by id:\n\n```ts\nimport type { BaseTrashItem, BaseTrashStore } from '@svgrid/enterprise/base'\n\nconst trashStore: BaseTrashStore = {\n list: async () => (await (await fetch('/api/trash')).json()) as BaseTrashItem[],\n add: (items) => fetch('/api/trash', { method: 'POST', body: JSON.stringify(items) }),\n remove: (ids) => fetch('/api/trash', { method: 'DELETE', body: JSON.stringify(ids) }),\n}\n```\n\nPass it as `<SvBase {trashStore}>`. A store that fails says so in a notice;\nthe page's trash stands either way.\n\nItems older than `trashDays` (30 by default, 0 for no limit) are dropped when\nthe base opens, the same as Delete forever: on a table with `softDelete` the row goes\nwith them, and one the database refuses stays in the trash. `restoreItems(tables, items)` puts items back in code, for an\nadmin screen that restores from a stored trash.\n\n| `BaseTrashItem.kind` | Also holds |\n|---|---|\n| `'record'` | `table` (id), `row`, `index` |\n| `'field'` | `table` (id), `column`, `index` |\n| `'table'` | `table` (the whole `BaseTable`), `index` |\n\nEvery item has `id`, `deletedAt` (ISO time) and, when someone is signed in,\n`deletedBy` (a user id).\n\n<div data-docs-demo=\"706-base-trash\" data-height=\"620\"></div>\n\n## Import a CSV or Excel file\n\nThe tab bar's menus import a CSV, TSV, Excel (.xlsx) or JSON file. The file\nis read in the page; nothing is uploaded. An Excel cell is read as it shows: a\nnumber in a percent cell as a percent, a number in a date cell as that date.\n\n- **As a new table** (the list of tables, \"Import a file as a table\"): each\n column becomes a field, the first one naming the records. The type comes\n from the column's values and the dialog shows each guess to change before\n the import.\n- **Into the shown table** (the table's menu, \"Import records from a file\"):\n each column goes to the field it names, matched by name or id in any case.\n A column that matches nothing is skipped unless you point it at a field. A\n value a select does not have becomes a new option. A link field reads the\n linked records' names, comma-separated, with a name that holds a comma in\n quotes. A person field reads names or emails.\n\nA CSV is read for what it is. The separator (comma, semicolon, tab, pipe) and\nthe encoding are found from the file, and the dialog shows both to change. A\ndecimal comma stays a decimal: `1,50` is 1.5, and `1,500` alone is one thousand\nfive hundred.\n\nEach value is read the way the field reads a paste. A date keeps the calendar day\nit is written with, in every time zone, in any form a person writes one\n(`October 9, 2026`, `12/02/2026`, `3 December 2026`, `2026-10-09`). `12/02/2026`\nreads by the field's date format, then the viewer's language, unless a number past\n12 in the same column settles it: with `13/02/2026` in the column, all of it is day\nfirst. A percent reads as the grid reads it: `25%` is 25%, and a bare `25` in a\npercent field is 25% too. A person matches by name or email. A name that matches\nnobody is not stored, and neither is a linked record's name that no record has: the\ndialog says how many cells that is before the import, and again after it.\n\nThe dialog says what it found before anything is imported: a file that is not\nUTF-8, a quote that is never closed (the quote stays as text and the rows after\nit are read as usual), rows longer or shorter than the header (extra cells get\ncolumns of their own), values a field type cannot hold, names that match no\nperson or record, and records that repeat a name already in the table. A file\nthat is not text is refused.\n\n## Download a view\n\nThe list of tables has \"Download this view as CSV\" and \"Download this view as\nExcel\". The file holds the view's fields in its order and its records, filtered\nand sorted, each cell as it reads on screen: a formula's result, the names of\nlinked records and people, a checklist's items, attachments' file names. Numbers\nstay numbers, at every digit they hold, whether the page or a database driver gave\nthem. A percent is written `12.5%` and a duration `1:30:15`, so the file imports back to\nthe same values. A person sees only what they may see, in the file too. Text that would\nstart a formula in a spreadsheet is guarded in a CSV.\n\nIn Excel a number, a percent, a date and a time of day are real cells with the field's\nnumber format: they sum and sort, and read as on screen, in the date format and at\nthe places the field shows. That goes for a formula with a result of that kind, a count, a\nrollup, and a lookup with one value. A moment reads at the time its cell shows, in the\nviewer's zone or in GMT for a GMT field. These stay text: a number of more digits than a spreadsheet\nkeeps (16 and up), so every digit stays; a date before March 1900, which Excel has no day for;\na lookup of several values; a duration; and a yes or no field, as the word `true` or `false`\non every driver.\n\n`recordsAsShown(columns, rows)` gives the same values in code. `api.exportCsv`\nwrites what is stored: nothing for a formula, ids for links and people.\n\n| Values in the column | Field type |\n|---|---|\n| `$1,200`, `€80.50` | Number as currency, with the symbol |\n| `25%` | Number as percent (0 to 100) |\n| `yes` / `no`, `true` / `false`, `x` | Checkbox |\n| `42`, `3.75` | Number (integer or decimal) |\n| `ana@example.com` | Email |\n| `https://...` | URL |\n| `2026-11-02`, `02/11/2026` | Date (with the time when there is one) |\n| `+1 415 555 0100` | Phone |\n| A few short values, repeated | Single select, a color per value |\n| `Design, Copy`, items repeated | Multiple select |\n| Long text or line breaks | Long text |\n| Anything else | Single line text |\n\nAn import into a table is one undo step. An app that already has a file (a\ndrop zone of its own, a download) opens the same dialog with it:\n\n```svelte\n<script lang=\"ts\">\n let base = $state<ReturnType<typeof SvBase> | null>(null)\n</script>\n\n<SvBase bind:this={base} bind:tables editable />\n<button onclick={() => base?.openImport('append', file)}>Import</button>\n```\n\n`openImport('new', file)` makes a new table from it. In code, without the\ndialog, `tableFromMatrix(rows, tables, name)` builds a table from a header row\nand rows of text, and `recordsFromMatrix(rows, table, matchColumns(header,\ntable))` turns them into records of an existing table. `fileImport={false}`\nremoves the menu items.\n\n<div data-docs-demo=\"709-base-import-csv-excel\" data-height=\"620\"></div>\n\n## Next\n\n- [Views for every team](./views.md): personal and locked views, and the\n layout on a phone.\n- [Collaboration](./collaboration.md): who is signed in, comments and\n @mentions.\n- [Keep a base in a database](./data.md): a restored or imported record is\n written back like any other change.\n"
5300
+ "markdown": "# Undo, the trash and importing files\n\nThree things people reach for in their first week with a base: Ctrl+Z after\na slip, getting back a record someone deleted yesterday, and pulling in the\nspreadsheet they kept until now. All three work on `<SvBase tables>` with no\nsetup; this page says what each one does and how an app keeps it.\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\n## Undo and redo\n\nA step undoes what it did and nothing else: what teammates changed since stays, whether\nthe change came through a live feed or a read of the table. A cell someone changed after\nyou keeps their value, on undo and on redo, cell by cell in a paste, and a record you added\nthat someone has worked on since stays, and the page says so. Delete it if it should go.\nFor a range the page says how many: \"Someone has changed 1 of these records since, so it\nkeeps their value.\" once all its blocks are written, or \"1 of these cells\" for a small range\nundone cell by cell.\nA cell undo on a table in a database reads the record first, with a live feed or without\none, so this holds for a change the page never saw; a date with a time is compared as the\npage last heard of it. When the record cannot be read, or a write of it has not answered\nyet, the undo is refused with a message and the step stays the next one to try. A number\nthe column rounded (1.234 in a column of two decimals) counts as what you typed. Moving a\nrecord by drag is a step too.\n\nOn a table in a database one Undo or Redo runs at a time. While one is still checking or\nwriting, another press changes nothing and says \"Undo is still running.\" (or \"Redo is still\nrunning.\"); the buttons show it busy, and leaving the page asks first. One that writes more\nthan one block (1,000 records, or 1,000 cells of a table the page holds) runs in the dialog a\nbig paste has, with its count and Stop: a range, Edit fields, a delete, the records a paste\nadded, and cells. Stopped, an Undo, or a Redo of cells or of added records, stays the next\nstep, and pressing it again writes only what is left; a Redo of a delete that is stopped\nkeeps the records it deleted, and Undo brings those back. A turn that writes a record the\nview no longer shows is done when that write has answered. What an Undo or Redo did is said\nonce it is done, with the number it changed: cells someone else changed, records a teammate\nrestored first or that were deleted forever, records out of reach and writes the database\nrefused are not counted. A paste of cells the grid holds says its count the same way, once\nthe database has answered.\n\nA record you add to a table read a part at a time has its \"add record\" step from the\nmoment its row is there. What you type into it is part of that step. An Undo before the\ndatabase has stored it waits for the insert, then takes the record out; a new record that\nwas never stored (left empty, or refused) just goes, with no Redo. Its activity starts\nonce it is stored: a refused record has none.\n\nAn undo or a redo puts values back. It runs no automation, and the row returns to its\nplace in the view at once.\n\nAn undo of a delete that fails (the network dropped, the database refused) leaves the\nrecord in the trash and the step in place: press Ctrl+Z again, or restore it from the\ntrash.\n\nCtrl+Z (Cmd+Z on a Mac) undoes what was done to the shown table, in the\norder it was done. Ctrl+Shift+Z or Ctrl+Y redoes it. The keys work anywhere\nin the base, and also with nothing focused on the page (after a dialog\ncloses, or a click beside the base): then the base used last takes them. A\nkey typed in a field of your own page stays yours. One history holds:\n\n- cell edits, pastes, fills and Edit fields on a selection, each one step however many\n cells it wrote (the grid's own history). Edit fields on more than 10,000 records of a\n table read a part at a time has no step: it says so and asks first. If it is stopped or\n fails with 10,000 or fewer changed, those have a step. A paste, Delete or fill over a large\n range goes the same way, and its undo reads and writes in blocks too;\n- records added, deleted or imported from a file;\n- fields added, changed (a new type, new options) or deleted;\n- tables added, renamed, duplicated, deleted or imported from a file.\n\nUndoing a deleted record puts it back at the place it had, and the links the\ndelete cleared come back too: delete a task and its project stops listing\nit; undo, and the project lists it again. The step restores the records of\nevery table the change touched.\n\nUndoing a deleted table brings it back as a tab, with its records and the\nfields of other tables that linked to it, and takes it out of the trash.\nUndoing a table you added takes it away again; the step lives on the new\ntable's history, so there is no redo once it is gone.\n\nThe Undo and Redo buttons at the end of the tab bar say what the next press\ndoes (\"Undo delete 2 records\"). `undoRedo={false}` hides them; the keys still\nwork. Each table keeps its history while another table is shown, so going\nback to a table carries on where you left it. View settings (a filter, a\nsort) are not part of the history: a view keeps them as you leave it.\n\n<div data-docs-demo=\"705-base-undo-redo\" data-height=\"620\"></div>\n\n## Soft delete\n\nUndo of a delete writes the record back. Point `softDelete` at a timestamp column and it\ndoes not have to: a delete sets the column and leaves the row, and a restore clears it.\nNothing about the record is rewritten, so columns off the schema, exact times and big\nnumbers come back as they were, and what a teammate wrote before the delete stays.\n\n```sql\nALTER TABLE orders ADD COLUMN deleted_at timestamptz;\n```\n\n```ts\nconst source = createSqlDataSource({ schema, table: 'orders', softDelete: 'deleted_at', dialect, execute })\n```\n\n`createSupabaseDataSource` takes the same option, and so do `createBaseEndpoint`,\n`baseFromDatabase` and `baseFromSupabase`:\n\n```ts\nconst endpoint = createBaseEndpoint({ execute, tables: ['orders', 'tags'], authorize, softDelete: 'deleted_at' })\n```\n\nPer table it is a function: `softDelete: (table) => (table === 'orders' ? 'removed_at' : null)`.\n\nThe string form skips a table without the column. The function form names a column that\nmust be there. Those three keep the column out of the fields. By hand, leave it off the\nschema: no write sets it either way, so a client cannot create a record already deleted or\nmark one through an update, and `createKitHandlers` answers `422` to an update of it alone.\nReads, counts, groups, totals, lookups and rollups skip marked records, and a write to one\nis told the record is gone.\n\n- The trash of a table with `softDelete` reads the table: every marked row is there, after\n a reload, for whoever may delete, whoever marked it. It lists the latest 1,000 per table\n and says how many there are. No trash store is needed for those records.\n- A record's links through a junction table stay. A delete leaves the junction rows, at\n either end of the link, a marked record shows in no link, and a restore finds them.\n- A foreign key in another table stays too. A delete writes nothing to that table, the\n record shows in no link while it is marked, and a restore finds its links where they were,\n whoever restores it and after any reload. A link a teammate changed in the meantime stays\n as they left it. While the record is marked the link reads as empty everywhere: \"is empty\"\n and \"is not empty\", one Empty group, sort, count and the link picker. The key stays in the\n column, so a query of your own still sees it. A link written to a marked record, from a\n picker opened before the delete, is stored, shows empty, and the person is told why.\n- Delete forever and Empty trash are final. Each asks first, the item stays in the trash\n until the database has confirmed, and if it refuses (another table still needs the record)\n the person is told and the record can still be restored. In one Empty trash the others go.\n Junction rows go with the record, and the foreign keys that point at it are emptied with it,\n in the tables of this base: see [Delete forever is one step](#delete-forever-is-one-step). A\n `NOT NULL` key cannot be emptied, so the record stays and the person is told other records\n depend on it. A row itself marked deleted keeps its key: a foreign key constraint refuses\n the purge until that row has gone. Undo cannot bring a deleted record back, in that window\n or another. An undo to a link whose record was deleted forever says so, and offers no Try again.\n- A trash item reaching `trashDays` is deleted for real only when the page's trash holds\n it, that is, when a trash store or `bind:trash` kept it. A row found marked in the table\n stays until someone deletes it, so opening a page never purges what another tool marked.\n For a limit on those, run it on a schedule:\n `DELETE FROM orders WHERE deleted_at < now() - interval '30 days'`.\n- A delete too big for the trash marks and purges a block of 1,000 at a time. A page closed\n half way leaves at most the block under way marked, and it shows in the trash.\n- A marked row still holds its value in a UNIQUE column, and still counts for a foreign key.\n- On Supabase a delete is an UPDATE to row-level security, so write the policy as one.\n- Autonumber reads the highest number over every row, marked ones too, on every source, so\n a marked record keeps its number and a new record takes the next.\n- A trash store item from before `softDelete` was turned on has no mark to clear: Restore\n says it was deleted forever.\n\n### Delete forever is one step\n\nEmptying the keys that hold a record and deleting it are several writes. Which of them can\nstop half way depends on how the base reaches the database:\n\n| The base is on | Delete forever is |\n|---|---|\n| `createBaseEndpoint` with `transaction` | One request and one transaction: all of it or none of it |\n| `createBaseEndpoint` without | One request: the server runs the steps, and a closed page changes nothing |\n| `createKitHandlers` with `linkedFrom` | The same two, by whether the sources share a `transaction` |\n| `baseFromDatabase` with `transaction` | One transaction, for the tables that share it |\n| `createKitHandlers` without `linkedFrom`, `baseFromDatabase` without `transaction` | Steps the page runs, over several requests |\n| Supabase and PostgREST | Steps the page runs. PostgREST has no transaction across requests |\n\nWhere it is steps, they are safe in this order. First the record's mark changes to\n`1970-01-03`, which is a claim: from then on no Restore or undo reaches it, by any path.\nThen the keys go, each only where it still holds this record, then the record. A failure\nputs the keys back, only where a key is still empty, so a link a teammate made meanwhile\nstays, and gives the claim up, with the mark set to that moment. If that fails too, or the\npage closes, the claim stays: the record stays in the trash marked as not finished, Restore\nsays why it cannot, and Delete forever finishes the job. Its links may be gone by then, so\nit does not come back with fewer than it had. A purge never reports a record it did not delete.\n\nGive the `transaction` next to `execute`, the same function for every table of a database. It\nruns `run` in a transaction, hands it an `execute` on that connection, and resolves to what\n`run` resolved to. A throw must roll back.\n\n```ts\n// node-postgres\ntransaction: async (run) => {\n const client = await pool.connect()\n try {\n await client.query('BEGIN')\n const out = await run((sql, params) => client.query(sql, params).then((r) => r.rows))\n await client.query('COMMIT')\n return out\n } catch (e) {\n await client.query('ROLLBACK').catch(() => {})\n throw e\n } finally {\n client.release()\n }\n}\n\n// postgres.js\ntransaction: (run) => sql.begin((tx) => run((text, params) => tx.unsafe(text, params)))\n\n// mysql2\ntransaction: async (run) => {\n const c = await pool.getConnection()\n try {\n await c.beginTransaction()\n const out = await run((sql, params) => c.query(sql, params).then(([rows]) => rows))\n await c.commit()\n return out\n } catch (e) {\n await c.rollback().catch(() => {})\n throw e\n } finally {\n c.release()\n }\n}\n\n// mssql\ntransaction: async (run) => {\n const tx = new mssql.Transaction(pool)\n await tx.begin()\n try {\n const out = await run(async (text, params) => {\n const request = new mssql.Request(tx)\n params.forEach((value, i) => request.input(`p${i + 1}`, value))\n return (await request.query(text)).recordset ?? []\n })\n await tx.commit()\n return out\n } catch (e) {\n await tx.rollback().catch(() => {})\n throw e\n }\n}\n```\n\nOn SQLite run `BEGIN IMMEDIATE`, `run` and `COMMIT` or `ROLLBACK` on the one connection, and\ntake no other write on it meanwhile. A teammate's write to a row the transaction holds waits\nfor it, then lands. Under a transaction a Restore of the record waits too, and finds it gone.\n\nIn the server's step each table's own `authorize`, `scope`, `validate`, hooks and `audit` apply\nto the keys it empties, as the page's requests met them, with `afterUpdate` and the audit only\nfor what stayed emptied. A `beforeUpdate` hook can refuse. What it changes in the patch is not\nwritten here.\n\nWith routes wired by hand, tell each table who holds its keys, and which of its fields lead to a\ntable that marks. Without `liveLinks` a group on such a field has an Empty group for each\ntrashed key.\n\n```ts\nconst orders = createKitHandlers({\n schema: orderSchema,\n source: createSqlDataSource({ ...orderConfig, transaction, liveLinks: { customer_id: { table: 'customers', key: 'id', softDelete: 'deleted_at' } } }),\n})\nconst customers = createKitHandlers({\n schema: customerSchema,\n source: createSqlDataSource({ ...customerConfig, transaction }),\n linkedFrom: [{ handlers: orders, column: 'customer_id' }],\n})\n```\n\n### Index for soft delete\n\n`deleted_at IS NULL` is in every read, so a count or a group can no longer be answered from\nan index that does not hold the column. Scrolling, sorting and filtering a page barely\nnotice; counts and groups do, and MySQL more than Postgres. Measured on a million orders,\nPostgres 16 and MySQL 8.4 on default memory settings:\n\n| Statement, ms | Postgres, plain to soft | MySQL, plain to soft |\n|---|---|---|\n| Count all | 19 to 32 | 55 to 229 |\n| Group by status | 44 to 73 | 112 to 888 |\n| Count through a lookup filter | 14 to 61 | 33 to 465 |\n\nWith a third of 633,000 rows marked, the lookup count on MySQL took about 0.9 s warm and\n13.3 s cold; an index on `(account, deleted_at)` brought it to 18 ms. So add the mark to the\nindexes your views filter and group on, last:\n\n```sql\nCREATE INDEX orders_account_live ON orders (account, deleted_at);\n```\n\nOn Postgres a partial index does the same for less space:\n`CREATE INDEX orders_account_live ON orders (account) WHERE deleted_at IS NULL`. The trash\nreads the latest marks, so index the column by itself on a table with many of them.\n\n## The trash\n\nA deleted view goes there too, with its filter, sorts and settings, and \"Restore\"\nputs it back at its place in the views menu. Undo does the same.\n\nA deleted record, field or table goes to the trash. The trash button at the\nend of the tab bar shows how many items it holds and opens a list, newest\nfirst, with who deleted each one (`setCurrentUser`) and when. Restore puts an\nitem back:\n\n- a record in its table, at the place it had, with its links;\n- a field with its values (the values stay in the records while the field is\n in the trash);\n- a table as a tab, with the fields of other tables that linked to it made\n again from its side.\n\nThe notice that appears after a delete has a Restore button too. A record\nwhose table is in the trash waits until the table is restored; the dialog\nsays so.\n\nThe trash is plain data. Bind it and store it next to the tables:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, type BaseTable, type BaseTrashItem } from '@svgrid/enterprise/base'\n\n enableBase()\n let tables = $state.raw<BaseTable[]>(loadTables())\n let trash = $state<BaseTrashItem[]>(loadTrash())\n</script>\n\n<SvBase bind:tables bind:trash trashDays={30} editable onTablesChange={saveTables} onTrashChange={saveTrash} />\n```\n\nWith neither `onTrashChange` nor a store the trash lives in the page: it goes when the\npage is closed or reloaded, and the panel says so in place of a number of days. If the\napp keeps a `bind:trash` some other way, pass `trashKept` to show the days.\n\nTo keep the trash where the records are, so it survives a reload and every\nperson of the base gets the same one, give it a store. `list` is read when the\nbase opens (a promise is fine), so someone else's deletes appear in the panel when\nthe page is next opened; `add` and `remove` hear each item deleted, restored, undone\nor emptied, by id:\n\n```ts\nimport type { BaseTrashItem, BaseTrashStore } from '@svgrid/enterprise/base'\n\nconst trashStore: BaseTrashStore = {\n list: async () => (await (await fetch('/api/trash')).json()) as BaseTrashItem[],\n add: (items) => fetch('/api/trash', { method: 'POST', body: JSON.stringify(items) }),\n remove: (ids) => fetch('/api/trash', { method: 'DELETE', body: JSON.stringify(ids) }),\n}\n```\n\nPass it as `<SvBase {trashStore}>`. A store that fails says so in a notice;\nthe page's trash stands either way.\n\nItems older than `trashDays` (30 by default, 0 for no limit) are dropped when\nthe base opens, the same as Delete forever: on a table with `softDelete` the row goes\nwith them, and one the database refuses stays in the trash. `restoreItems(tables, items)` puts items back in code, for an\nadmin screen that restores from a stored trash.\n\n| `BaseTrashItem.kind` | Also holds |\n|---|---|\n| `'record'` | `table` (id), `row`, `index` |\n| `'field'` | `table` (id), `column`, `index` |\n| `'table'` | `table` (the whole `BaseTable`), `index` |\n\nEvery item has `id`, `deletedAt` (ISO time) and, when someone is signed in,\n`deletedBy` (a user id).\n\n<div data-docs-demo=\"706-base-trash\" data-height=\"620\"></div>\n\n## Import a CSV or Excel file\n\nThe tab bar's menus import a CSV, TSV, Excel (.xlsx) or JSON file. The file\nis read in the page; nothing is uploaded. An Excel cell is read as it shows: a\nnumber in a percent cell as a percent, a number in a date cell as that date.\n\n- **As a new table** (the list of tables, \"Import a file as a table\"): each\n column becomes a field, the first one naming the records. The type comes\n from the column's values and the dialog shows each guess to change before\n the import.\n- **Into the shown table** (the table's menu, \"Import records from a file\"):\n each column goes to the field it names, matched by name or id in any case.\n A column that matches nothing is skipped unless you point it at a field. A\n value a select does not have becomes a new option. A link field reads the\n linked records' names, comma-separated, with a name that holds a comma in\n quotes. A person field reads names or emails.\n\nA CSV is read for what it is. The separator (comma, semicolon, tab, pipe) and\nthe encoding are found from the file, and the dialog shows both to change. A\ndecimal comma stays a decimal: `1,50` is 1.5, and `1,500` alone is one thousand\nfive hundred.\n\nEach value is read the way the field reads a paste. A date keeps the calendar day\nit is written with, in every time zone, in any form a person writes one\n(`October 9, 2026`, `12/02/2026`, `3 December 2026`, `2026-10-09`). `12/02/2026`\nreads by the field's date format, then the viewer's language, unless a number past\n12 in the same column settles it: with `13/02/2026` in the column, all of it is day\nfirst. A percent reads as the grid reads it: `25%` is 25%, and a bare `25` in a\npercent field is 25% too. A person matches by name or email. A name that matches\nnobody is not stored, and neither is a linked record's name that no record has: the\ndialog says how many cells that is before the import, and again after it.\n\nThe dialog says what it found before anything is imported: a file that is not\nUTF-8, a quote that is never closed (the quote stays as text and the rows after\nit are read as usual), rows longer or shorter than the header (extra cells get\ncolumns of their own), values a field type cannot hold, names that match no\nperson or record, and records that repeat a name already in the table. A file\nthat is not text is refused.\n\n## Download a view\n\nThe list of tables has \"Download this view as CSV\" and \"Download this view as\nExcel\". The file holds the view's fields in its order and its records, filtered\nand sorted, each cell as it reads on screen: a formula's result, the names of\nlinked records and people, a checklist's items, attachments' file names. Numbers\nstay numbers, at every digit they hold, whether the page or a database driver gave\nthem. A percent is written `12.5%` and a duration `1:30:15`, so the file imports back to\nthe same values. A person sees only what they may see, in the file too. Text that would\nstart a formula in a spreadsheet is guarded in a CSV.\n\nIn Excel a number, a percent, a date and a time of day are real cells with the field's\nnumber format: they sum and sort, and read as on screen, in the date format and at\nthe places the field shows. That goes for a formula with a result of that kind, a count, a\nrollup, and a lookup with one value. A moment reads at the time its cell shows, in the\nviewer's zone or in GMT for a GMT field. These stay text: a number of more digits than a spreadsheet\nkeeps (16 and up), so every digit stays; a date before March 1900, which Excel has no day for;\na lookup of several values; a duration; and a yes or no field, as the word `true` or `false`\non every driver.\n\n`recordsAsShown(columns, rows)` gives the same values in code. `api.exportCsv`\nwrites what is stored: nothing for a formula, ids for links and people.\n\n| Values in the column | Field type |\n|---|---|\n| `$1,200`, `€80.50` | Number as currency, with the symbol |\n| `25%` | Number as percent (0 to 100) |\n| `yes` / `no`, `true` / `false`, `x` | Checkbox |\n| `42`, `3.75` | Number (integer or decimal) |\n| `ana@example.com` | Email |\n| `https://...` | URL |\n| `2026-11-02`, `02/11/2026` | Date (with the time when there is one) |\n| `+1 415 555 0100` | Phone |\n| A few short values, repeated | Single select, a color per value |\n| `Design, Copy`, items repeated | Multiple select |\n| Long text or line breaks | Long text |\n| Anything else | Single line text |\n\nAn import into a table is one undo step. An app that already has a file (a\ndrop zone of its own, a download) opens the same dialog with it:\n\n```svelte\n<script lang=\"ts\">\n let base = $state<ReturnType<typeof SvBase> | null>(null)\n</script>\n\n<SvBase bind:this={base} bind:tables editable />\n<button onclick={() => base?.openImport('append', file)}>Import</button>\n```\n\n`openImport('new', file)` makes a new table from it. A base whose tables live on a server keeps it\nwith `onImportTable` (see [Keep a base in a database](./data.md)). In code, without the\ndialog, `tableFromMatrix(rows, tables, name)` builds a table from a header row\nand rows of text, and `recordsFromMatrix(rows, table, matchColumns(header,\ntable))` turns them into records of an existing table. `fileImport={false}`\nremoves the menu items.\n\nThe base reads CSV, TSV, `.xlsx` (its first sheet) and JSON itself. To take\nother formats (an old `.xls`, `.ods`, `.xlsb`), give your own reader:\n`importReader` gets every file first and answers its rows as text, the header\nrow first, or `null` to leave the file to the base. With it, the file picker\noffers those formats too.\n\n```ts\nimport * as XLSX from 'xlsx'\n\nconst importReader = async (file: File) => {\n if (!/\\.(xls|xlsb|ods)$/i.test(file.name)) return null\n const book = XLSX.read(new Uint8Array(await file.arrayBuffer()), { type: 'array' })\n const sheet = book.Sheets[book.SheetNames[0]]\n return XLSX.utils.sheet_to_json(sheet, { header: 1, raw: false, defval: '' })\n}\n// <SvBase {tables} {importReader} />\n```\n\n<div data-docs-demo=\"709-base-import-csv-excel\" data-height=\"620\"></div>\n\n## Next\n\n- [Views for every team](./views.md): personal and locked views, and the\n layout on a phone.\n- [Collaboration](./collaboration.md): who is signed in, comments and\n @mentions.\n- [Keep a base in a database](./data.md): a restored or imported record is\n written back like any other change.\n"
5301
5301
  },
5302
5302
  {
5303
5303
  "slug": "help/base/views",
5304
5304
  "path": "docs/help/base/views.md",
5305
5305
  "title": "Views for every team",
5306
5306
  "section": "SvGrid Base",
5307
- "markdown": "# Views for every team\n\nOne table of records, many ways to look at it. Product reads the roadmap\ngrouped by quarter, engineering works off a status board, the leads check a\ntimeline, and anyone in the company files a request through a form. Each of\nthose is a saved view: a name, a type, and the filter, sorts, groups, hidden\nfields, row height and colors it had when someone last left it.\n\n![The Features table feeds six saved views: Roadmap by quarter (a grid grouped by Quarter, sorted by Impact, Lead hidden, row height 56, colored by Status), Now (a grid filtered to active work), Status board (a Kanban stacked by Status), Timeline (bars from Start to Target), Effort by team (a bar chart) and Request a feature (a form). onViewsChange hands the whole list to the app, which stores it and loads it back into bind:views.](/docs-media/base-views.svg)\n\n<div data-docs-demo=\"513-base-product-roadmap\" data-height=\"680\"></div>\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\nThis page builds the Features table of that roadmap one view at a time. Each\nstep adds to the `views` list from the step before and works on its own.\n\n## 1. The table\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, setFieldUsers, type FieldView } from '@svgrid/enterprise/base'\n\n enableBase()\n setFieldUsers([\n { id: 'u1', name: 'Ana Ruiz' },\n { id: 'u2', name: 'Ben Okafor' },\n { id: 'u3', name: 'Cleo Park' },\n ])\n\n type Feature = {\n id: number; feature: string; quarter: string; status: string; teams: string[]\n lead: string; impact: number; effort: number; start: string; target: string; votes: number\n }\n\n const QUARTERS = [\n { value: 'Q4 2026', color: '#2d7ff9' },\n { value: 'Q1 2027', color: '#18bfff' },\n { value: 'Later', color: '#94a3b8' },\n ]\n const STATUS = [\n { value: 'Planned', color: '#94a3b8' },\n { value: 'In progress', color: '#2d7ff9' },\n { value: 'In beta', color: '#8b46ff' },\n { value: 'Shipped', color: '#20c933' },\n { value: 'Blocked', color: '#f82b60' },\n ]\n const TEAMS = [\n { value: 'Web', color: '#2d7ff9' },\n { value: 'Mobile', color: '#ff6f2c' },\n { value: 'Platform', color: '#8b46ff' },\n ]\n\n let rows = $state.raw<Feature[]>([\n { id: 1, feature: 'Single sign-on with SAML', quarter: 'Q4 2026', status: 'In progress', teams: ['Platform'], lead: 'u2', impact: 5, effort: 12, start: '2026-10-05', target: '2026-11-13', votes: 182 },\n { id: 2, feature: 'Offline mode for the mobile app', quarter: 'Q4 2026', status: 'Blocked', teams: ['Mobile'], lead: 'u3', impact: 4, effort: 13, start: '2026-10-12', target: '2026-12-04', votes: 96 },\n { id: 3, feature: 'Dark mode', quarter: 'Q4 2026', status: 'Shipped', teams: ['Web'], lead: 'u1', impact: 2, effort: 4, start: '2026-10-01', target: '2026-10-16', votes: 154 },\n { id: 4, feature: 'Audit log export', quarter: 'Q1 2027', status: 'Planned', teams: ['Platform', 'Web'], lead: 'u1', impact: 3, effort: 5, start: '2027-01-11', target: '2027-01-29', votes: 41 },\n ])\n\n const columns = [\n { field: 'feature', header: 'Feature', width: 260, fieldType: 'text' },\n { field: 'quarter', header: 'Quarter', fieldType: 'singleSelect', fieldProps: { options: QUARTERS } },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: STATUS } },\n { field: 'teams', header: 'Teams', fieldType: 'multiSelect', fieldProps: { options: TEAMS } },\n { field: 'lead', header: 'Lead', fieldType: 'user' },\n { field: 'impact', header: 'Impact', fieldType: 'rating', fieldProps: { max: 5 } },\n { field: 'effort', header: 'Effort (days)', fieldType: 'number', fieldProps: { format: 'integer', summary: 'sum' } },\n { field: 'start', header: 'Start', fieldType: 'date', fieldProps: { dateFormat: 'friendly' } },\n { field: 'target', header: 'Target', fieldType: 'date', fieldProps: { dateFormat: 'friendly' } },\n { field: 'votes', header: 'Votes', fieldType: 'number', fieldProps: { format: 'integer', summary: 'sum' } },\n ]\n\n let views = $state<FieldView[]>([{ id: 'all', name: 'All features', type: 'grid' }])\n</script>\n\n<SvBase data={rows} {columns} bind:views primaryField=\"feature\"\n editable sortable showRowNumbers expandRows summary containerHeight={640}\n rowAdder={{ label: 'Add feature', defaults: () => ({ id: Date.now(), status: 'Planned', quarter: 'Later', teams: [] }) }} />\n```\n\nThe view bar over the grid has \"All features\" on the left, then Hide\nfields, Filter, Group, Sort, Color and a row height button, and \"4 records\"\nwith a search button on the right. The footer row shows \"Sum 34\" under Effort\nand \"Sum 473\" under Votes, because those fields set `fieldProps.summary`.\n\nEvery view key that names a field (`stackBy`, `dateField`, a filter\ncondition's `column`, a chart's `x`) takes the column id. A column without an\n`id` uses its `field`, so in this table the two are the same.\n\n## 2. A roadmap grouped by quarter\n\nA view can start with settings. `state` is the grid's own state, the same\nobject `api.getState()` returns; `color` says how rows are colored.\n\n```ts\nconst roadmap: FieldView = {\n id: 'roadmap',\n name: 'Roadmap by quarter',\n type: 'grid',\n color: { mode: 'field', field: 'status' },\n rowHeight: 56,\n state: {\n grouping: ['quarter'],\n sorting: [{ id: 'quarter', desc: false }, { id: 'impact', desc: true }],\n hiddenColumns: ['lead'],\n },\n}\n```\n\nGroups and sorts share the grid's `sorting` list: a grouped field's\ndirection is an entry ahead of the sorts. The view bar writes them that way,\nso write them that way too. Here the bar reads \"1 hidden field\", \"Grouped by\n1 field\" and \"Sorted by 1 field\": the `quarter` entry belongs to the group,\nnot to Sort.\n\nEach group header shows \"Quarter\" over the option's pill, the number of\nrecords, the summaries the fields picked for that group (\"Effort (days) 29\",\n\"Votes 432\"), and \"+\" to add a feature that already has that quarter. Groups\ncome in the select's option order, not alphabetically. Group rows are at\nleast 52px tall so the field's name fits over the value; with\n`rowHeight: 56` they match the records. Each row has a status color stripe\non its row header. 56 is the row height menu's \"Medium\": from 48px, text and\nselect pills wrap.\n\n## 3. A view that starts filtered\n\nThe filter is the grid's advanced filter with its own kind, `fieldFilter`.\nPut it in `state.advancedFilter`:\n\n```ts\nimport { describeFieldFilter, type FieldFilter } from '@svgrid/enterprise/base'\n\nconst activeWork: FieldFilter = {\n kind: 'fieldFilter',\n conjunction: 'and',\n conditions: [\n { column: 'status', kind: 'select', op: 'isAnyOf', value: ['In progress', 'In beta', 'Blocked'] },\n {\n conjunction: 'or',\n conditions: [\n { column: 'target', kind: 'date', op: 'isWithin', value: { span: 'nextDays', days: 30 } },\n { column: 'impact', kind: 'number', op: 'gte', value: 4 },\n ],\n },\n ],\n}\n\nconst now: FieldView = {\n id: 'now',\n name: 'Now',\n type: 'grid',\n state: { sorting: [{ id: 'target', desc: false }], advancedFilter: activeWork },\n}\n\nconst header = (id: string) => columns.find((c) => c.field === id)?.header ?? id\ndescribeFieldFilter(activeWork, header)\n// \"Status is any of In progress, In beta, Blocked and (Target is within the next 30 days or Impact ≥ 4)\"\n```\n\nThe Filter button reads \"Filtered by 3 fields\" (one field reads \"Filtered by\nStatus\") and the record count drops to what the filter keeps.\n`describeFieldFilter` writes a filter in the words the panel uses, for a page\nsubtitle or an audit line. It returns `\"every record\"` for a filter with no\ncomplete conditions.\n\nA condition is a field, an operator and a value; a group is one level deep.\n[Filters in code](#filters-in-code) below lists the operators per kind.\n\nOne view can serve every lead. `ME` (the string `'@me'`) in a person\ncondition stands for whoever `setCurrentUser` names when the filter runs; the\nFilter panel lists it as \"Me\", first among the people.\n\n```ts\nimport { ME, setCurrentUser } from '@svgrid/enterprise/base'\n\nsetCurrentUser({ id: 'u2', name: 'Ben Okafor' })\n\nconst mine: FieldView = {\n id: 'mine',\n name: 'My features',\n type: 'grid',\n state: {\n advancedFilter: {\n kind: 'fieldFilter',\n conjunction: 'and',\n conditions: [{ column: 'lead', kind: 'select', op: 'is', value: ME }],\n },\n },\n}\n```\n\nBen sees \"Single sign-on with SAML\"; Ana, signed in, sees her two. With\nnobody signed in the condition matches no one. `describeFieldFilter` writes\nit as \"Lead is me\".\n\n## 4. A status board\n\n```ts\nconst board: FieldView = { id: 'board', name: 'Status board', type: 'kanban', stackBy: 'status' }\n```\n\n![The roadmap's Status board: Planned, In progress, In beta and Shipped lanes of feature cards showing Quarter, Teams and Lead, with the right-click menu open on \"Single sign-on with SAML\": Expand record, Comment, Move to, Duplicate record, Delete record.](/docs-media/base-views.png)\n\n`stackBy` takes a single select or a user field that holds one person. You\nget one lane per option, in option order and color, plus \"Uncategorized\",\nwhich shows only while a record has no status. Stack by `lead` instead and\nthe lanes are the people from `setFieldUsers`.\n\nSort the view by a number field and a card dropped elsewhere in its lane takes\na number between its neighbours', so the order is the table's and lasts.\n\nDrop a card in another lane and the status changes through the grid's edit\npath, so undo, validation and `onCellValueChange` all see it. \"+ Add a card\"\nadds a record already in that lane, named by the primary field.\n\nA card shows the record's name, then the next three fields the view shows,\nin column order (the stacked field left out), each drawn as its cell draws\nit. Hide a field in Hide fields and it leaves the cards; drag it up there to\nput it on them. Cards are read, not edited: a rating shows its stars\nbut does not take a click. Double-click a card or press F2 to open the\nrecord. Right-click it for its menu: Expand record, Comment, and, when the\nbase is `editable`, Move to (a submenu of the lanes), Duplicate record and\nDelete record, which asks \"Delete record?\" first. A card with comments shows\ntheir count; click it to open the record at its comment box.\n\n## 5. A timeline\n\n```ts\nconst timeline: FieldView = {\n id: 'timeline',\n name: 'Timeline',\n type: 'gantt',\n dateField: 'start',\n endField: 'target',\n color: { mode: 'field', field: 'status' },\n}\n```\n\nEach feature is a bar from its start to its target date, beside a table of\nnames and dates. Drag a bar to move both dates; drag its edge to change the\ntarget. One drag is one Ctrl+Z. `type: 'calendar'` with the same fields shows\nthe features on a month instead: a date field without a time shows as all-day\nevents. The end of a bar or an event is exclusive, so the record keeps the last\nday: a feature dragged to cover the 19th to the 21st gets\n`target: '2026-10-21'`.\n\nWith only `dateField`, events move but cannot be resized. Pick a day (or a\nrange) on the calendar and a new record lands on it, opened with its name\nready to type. Colored by a single select, the events take the option colors\nand the options appear above the calendar as a legend that hides and shows\nthem.\n\nMoving and resizing need `editable` on the base. Without it the calendar is\nread-only.\n\n## 6. Effort by team\n\n```ts\nconst effort: FieldView = { id: 'effort', name: 'Effort by team', type: 'chart', chart: { kind: 'barH', x: 'teams', y: 'effort' } }\nconst mix: FieldView = { id: 'mix', name: 'Features per quarter', type: 'chart', chart: { kind: 'donut', x: 'quarter' } }\n```\n\nA chart groups the records the view shows by `x` and adds up `y` (or counts\nrecords when there is no `y`). `aggregation: 'avg'` averages instead of\nsumming. Kinds are `'bar'`, `'barH'`, `'line'`, `'pie'` and `'donut'`.\n\nHow `x` groups:\n\n- A single or multiple select keeps its options' order and colors. A\n multiple select counts a record in each of its options, so \"Audit log\n export\" adds 5 days to both Platform and Web.\n- A person or a linked record shows by name.\n- A date groups by month (\"Oct 2026\").\n- Records with no value go in an \"Empty\" bar, last.\n\nValues read as the summed field writes them, and a count as a whole number.\nThe series is labeled \"Sum of Effort (days)\", \"Average of ...\" or \"Records\".\nThe bar shows Filter, which narrows what is charted, and \"Customize chart\",\nwhose Type, Group by, Value and Combine (sum or average) are kept in `chart`.\nA chart view declared without `chart` draws \"Choose a field to chart in\nCustomize chart.\" until a field is picked there.\n\n## 7. A gallery of mockups\n\nAdd an attachment field, then a gallery view with it as the cover:\n\n```ts\nconst mockupField = { field: 'mockups', header: 'Mockups', fieldType: 'attachment' }\nconst gallery: FieldView = { id: 'mockups', name: 'Mockups', type: 'gallery', coverField: 'mockups' }\n```\n\nEach card shows the first image of `mockups`, the primary field as its title,\nthen the fields the grid shows, each drawn as its cell draws it. Hide a field\nin this view and it leaves the cards; drag fields in Hide fields to reorder\nthem. A click on a card opens the record. Without `coverField` the cards have\nno cover.\n\n## 8. A request form\n\n```ts\nconst request: FieldView = {\n id: 'request',\n name: 'Request a feature',\n type: 'form',\n formFields: ['feature', 'teams', 'impact'],\n formRequired: ['feature'],\n formTitle: 'Request a feature',\n formDescription: 'What should we build, and who is it for? Product reviews requests every Friday.',\n formSubmitLabel: 'Send request',\n}\n```\n\nThe form asks for the fields in `formFields` (column ids, in order), each with\nthe editor its cell uses. Without it, the form asks for every field a person\ncan fill in, the primary field first; computed fields are left out. A field\nin `formRequired` must be filled before the form sends. `formTitle` defaults\nto the view's name, `formSubmitLabel` to \"Submit\". A sent form adds a record\nthe way \"+ Add row\" does: `rowAdder.defaults` first, so a request starts as\nPlanned in Later, then what was typed, then `onRowAdd`. The bar shows only\nthe view switcher on a form.\n\nAll the views so far, in one list:\n\n```ts\nlet views = $state<FieldView[]>([roadmap, board, timeline, effort, now, mine, gallery, request])\n```\n\n## 9. A dashboard\n\n```ts\nconst OPEN: FieldFilter = { kind: 'fieldFilter', conjunction: 'and', conditions: [{ column: 'status', kind: 'select', op: 'isNot', value: 'Shipped' }] }\nconst SHIPPED: FieldFilter = { kind: 'fieldFilter', conjunction: 'and', conditions: [{ column: 'status', kind: 'select', op: 'is', value: 'Shipped' }] }\n\nconst overview: FieldView = {\n id: 'overview',\n name: 'Overview',\n type: 'dashboard',\n dashboard: [\n { id: 'open', kind: 'number', title: 'Open features', filter: OPEN },\n { id: 'days', kind: 'number', title: 'Days of work left', aggregation: 'sum', field: 'effort', filter: OPEN, caption: 'Team capacity 40 days' },\n { id: 'shipped', kind: 'progress', title: 'Shipped', filter: SHIPPED },\n { id: 'status', kind: 'chart', chart: { kind: 'bar', x: 'status' } },\n { id: 'wanted', kind: 'list', title: 'Most wanted', sortBy: 'votes', sortDesc: true, limit: 3 },\n ],\n}\n```\n\nA dashboard is a grid of tiles. Each tile reads the records of one table,\nkeeps the ones its `filter` matches (the same `FieldFilter` as a view's\nfilter), and shows one of four things:\n\n- **`'number'`** counts the records, or with `aggregation: 'sum' | 'avg' |\n 'min' | 'max'` and a number `field`, adds them up. The number is written the\n way the field writes it, so a currency field shows \"$42,000\". Here \"Open\n features\" reads 3 and \"Days of work left\" reads 30. `caption` is a line\n under the number.\n- **`'chart'`** takes a chart view's `chart` setting (`kind`, `x`, `y`,\n `aggregation`) and groups the tile's records the same way.\n- **`'list'`** shows the first `limit` records (5 by default) sorted by\n `sortBy`, with that field's value beside each name (\"Single sign-on with\n SAML 182\"); `detail` shows another field instead. A click opens the record.\n- **`'progress'`** shows how many of all the table's records match the filter:\n \"25%, 1 of 4 records\" for Shipped.\n\n`size` is `'1x1'`, `'2x1'` or `'2x2'` (columns by rows). A number or progress\ntile is `'1x1'` by default, a chart or list `'2x2'`. Below about 450px the\ntiles stack in one column. A tile without `title` is named after what it\nshows (\"Total Effort (days)\", \"Status is Shipped\").\n\nWith several tables (`<SvBase bind:tables>`), a tile's `table` names any\ntable of the base, so a dashboard on Deals can count Accounts and list the\nActivities due this week. A tile without `table` reads the table the view is\non. Its filter and fields are that table's.\n\nA number or progress tile is a button: a click, or Enter, lists the records\nit counts (\"Open the 3 records\") in a dialog, each with its name and a few\nof its fields. On a chart, a click on a bar or a slice lists the records in\nit: the records the tile's filter keeps that fall in that bucket. A click on a record in the dialog opens it, in its\nown table's tab when the tile reads another table. While the dashboard is\nbeing edited, the tiles open nothing.\n\n\"Edit dashboard\" at the top of the view adds a tile, and on each tile sets its\nsize, moves it (drag it, or the arrows), edits it in a dialog (kind, table,\nvalue, filter, title, size) or removes it. Every change is saved in the view's\n`dashboard` at once and reaches `onViewsChange` and `onTablesChange`. A locked\nview has no \"Edit dashboard\". The view bar shows only the view switcher on a\ndashboard.\n\n<div data-docs-demo=\"721-base-dashboard\" data-height=\"680\"></div>\n\n## 10. A list for the phone\n\n```ts\nconst list: FieldView = {\n id: 'list',\n name: 'Feature list',\n type: 'list',\n listFields: ['status', 'teams', 'votes'],\n state: { sorting: [{ id: 'quarter', desc: false }, { id: 'votes', desc: true }], grouping: ['quarter'], hiddenColumns: [], advancedFilter: null },\n}\n```\n\nA list shows one record per entry: the primary field in bold, and under it\nthe fields in `listFields` (column ids, in order), each drawn as its cell\ndraws it, so Status is a colored pill and Teams are tags. An empty value is\nleft out. Without `listFields` the list shows the next three fields the grid\nshows. Long names stay readable at phone width, where a grid would cut them.\n\nThe list keeps the view's filter, sorts and groups. Here it is grouped by\nQuarter: a header per quarter with the field's name, its value and the\nnumber of records, \"Quarter Q4 2026 3\", and a click on the header folds the\ngroup. Within each quarter the most voted feature comes first. Grouped by\nmore fields (`grouping: ['quarter', 'status']`), each quarter holds a header\nper status, indented under it, with its own count and its own fold.\n\nA click on a record, or Enter, opens it. The list is one tab stop: Up and\nDown move between records and group headers at every level, Home and End\njump, Left and Right fold and unfold a group, and Left on a record or a\nfolded group moves to the group it is in. A record someone else changed and nobody here\nopened yet has a dot before its name.\n\nThe view bar has Hide fields, Filter, Group, Sort and \"Customize list\", which\nturns the fields under the name on and off and saves them in `listFields`.\n\n<div data-docs-demo=\"719-base-list-view\" data-height=\"620\"></div>\n\n## 11. An inbox of changes\n\n```ts\nconst inbox: FieldView = {\n id: 'inbox',\n name: 'Inbox',\n type: 'inbox',\n inboxDateField: 'start',\n listFields: ['status', 'lead'],\n}\n```\n\nAn inbox shows the records on the left and the picked one on the right: its\nfields, each one editable in place with the editor its cell uses, and its\nlatest changes, \"Cleo Park changed Status: Planned → In progress\". \"Expand\"\nopens the full record with its comments.\n\nThe list is newest first by `inboxDateField` (a date field's column id), so\n\"Audit log export\", which starts on January 11, is on top.\nRecords without a date go last. Without `inboxDateField` the inbox keeps the\nview's sort. The view's filter applies either way.\n\nA record someone else changed and nobody here has opened is unread: its name\nis bold and it has a dot. The changes come from a table with a `source` that\nhears of writes made elsewhere ([Changes made\nelsewhere](./data.md#changes-made-elsewhere)). Picking a record marks it read,\nas opening it in any view does, and \"Mark all read\" marks the rest. The count\nat the top says \"2 unread\" or \"All read\".\n\nDown or j picks the next record and Up or k the previous one; Enter opens the\nfull record. Narrower than 640px the inbox shows the list, and a picked record\nfills it with a Back button (Escape goes back too).\n\nThe view bar has Hide fields, Filter, Sort and \"Customize inbox\": the date\nfield it lists by and the fields under each name. A field hidden in the view\nleaves the record on the right too.\n\n<div data-docs-demo=\"720-base-inbox-view\" data-height=\"620\"></div>\n\n## 12. Keep the views\n\n`onViewsChange` fires with the whole list when a view is created, renamed,\nduplicated or deleted, and about 400ms after the shown view's settings stop\nchanging. Switching views stores the one being left first. Store the list\nwherever the app stores things.\n\nIn the browser:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, type FieldView } from '@svgrid/enterprise/base'\n\n enableBase()\n const KEY = 'roadmap:views'\n const DEFAULT_VIEWS: FieldView[] = [{ id: 'all', name: 'All features', type: 'grid' }]\n\n function read<T>(key: string, fallback: T): T {\n try {\n const raw = localStorage.getItem(key)\n return raw ? (JSON.parse(raw) as T) : fallback\n } catch {\n return fallback\n }\n }\n function write(key: string, value: unknown) {\n try {\n localStorage.setItem(key, JSON.stringify(value))\n } catch {\n // Private mode or a full quota: the views still work, unsaved.\n }\n }\n\n let views = $state<FieldView[]>(read(KEY, DEFAULT_VIEWS))\n let activeView = $state<string>(read(`${KEY}:active`, ''))\n\n $effect(() => write(`${KEY}:active`, activeView))\n</script>\n\n<SvBase data={rows} {columns} primaryField=\"feature\" editable\n bind:views bind:activeView onViewsChange={(v: FieldView[]) => write(KEY, v)} />\n```\n\nOn a server, decide who owns a list.\n\n- **Per user.** Each person keeps their own list, so one lead's filter does\n not change another's. Seed it from the team's defaults the first time.\n- **Shared.** One list per table that everyone reads. Write it only for the\n people you allow; for everyone else, keep their changes in memory or in\n their own list. A shared \"My features\" view built on `ME` still shows each\n person their own records.\n\n```ts\ntype Scope = 'me' | 'team'\n\nasync function loadViews(scope: Scope): Promise<FieldView[]> {\n const res = await fetch(`/api/tables/features/views?scope=${scope}`)\n return res.ok ? ((await res.json()) as FieldView[]) : DEFAULT_VIEWS\n}\n\nfunction saveViews(scope: Scope, next: FieldView[]) {\n void fetch(`/api/tables/features/views?scope=${scope}`, {\n method: 'PUT',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify(next),\n })\n}\n```\n\n```svelte\n{#await loadViews('me') then loaded}\n <SvBase data={rows} {columns} primaryField=\"feature\" editable\n views={loaded} onViewsChange={(v: FieldView[]) => saveViews('me', v)} />\n{/await}\n```\n\nOr mount at once and swap the list in when it arrives. When `views` or\n`activeView` changes from outside and the shown view is not the one the grid\nhas, the bar applies it. It never stores the grid's settings into a view it\nhas not applied yet.\n\n```svelte\n<script lang=\"ts\">\n let views = $state<FieldView[]>(DEFAULT_VIEWS)\n let loaded = false\n loadViews('me').then((v) => {\n views = v\n loaded = true\n })\n</script>\n\n<SvBase data={rows} {columns} primaryField=\"feature\" editable\n bind:views onViewsChange={(v: FieldView[]) => loaded && saveViews('me', v)} />\n```\n\nThe `loaded` check matters: about 400ms after mount the bar stores the\ndefault view's settings, and that save must not replace the list on the\nserver before the real one is in.\n\nWith several tables (`<SvBase bind:tables>`), each table keeps its own\n`views`, and `onTablesChange` hands back the whole list of tables, views\nincluded.\n\n## Personal and locked views\n\nOnce several people share a table, two more keys on a view matter:\n\n```ts\nconst views: FieldView[] = [\n // Only Ana sees it, under \"My views\" in the views menu.\n { id: 'ana-open', name: 'My open tasks', type: 'grid', personal: 'u1', state: { /* ... */ } },\n // Everyone sees it; nobody changes its settings until someone unlocks it.\n { id: 'report', name: 'Leadership report', type: 'grid', locked: true, lockedBy: 'u2', state: { /* ... */ } },\n]\n```\n\n- **`personal: userId`** lists the view only for that person, read from\n `setCurrentUser`. With nobody signed in every view is listed. \"Create\" in\n the views menu has a \"Personal view: only I see it\" box once someone is\n signed in.\n- **`locked: true`** keeps the view's filter, sorts, groups, hidden fields,\n colors and row height as saved. Its settings buttons are off, nothing is\n saved into it, and the bar shows \"Locked\" with who locked it. A sort from a\n column header still works for the moment; the bar then says \"Locked, not\n saved\" with a Reset. Records stay editable. A locked view cannot be\n deleted, and a copy of it starts unlocked.\n\nThe lock button in the views menu locks the view as the grid shows it now,\nor unlocks it. `lockViews={false}` on `SvBase` removes it for people who may\nnot lock views. Both keys come back through `onViewsChange` and\n`onTablesChange`, so the app stores a personal view with its owner the same\nway as the others.\n\n<div data-docs-demo=\"710-base-personal-locked-views\" data-height=\"620\"></div>\n\n## On a phone\n\nWhen a table is narrower than 600px (its own width, not the window's), a\ngrid view shows each record as a card: the primary field, then the next four\nfields the view shows, in the view's filter and sort. A tap opens the\nrecord, \"+\" adds one, and on a phone screen the opened record fills it. The\nview bar keeps its icons and the tab bar its tabs.\n\nA card keeps what a row can do. A long press, the button on the card, a\nright-click, the Menu key or Shift+F10 opens the record's menu: expand, comment,\nselect, insert above or below, duplicate and delete. A record selected this way\nshows on its card, as a ring on its list row, and in the selection bar. Copy and\npaste work on a cell, so they stay with the table. A gallery view and a list view\nhave the same menu.\n\nA Kanban lane takes the table's width, the next lane a swipe away and\nsnapped into place. A calendar opens in its agenda, the days listed one under\nanother, since a month of seven columns has no room for a title; Month, Week\nand Day are a tap away. Gallery cards stand one per row, and a form is one\ncolumn at any width.\n\n`phoneLayout` on `SvBase` moves the breakpoint (`phoneLayout={480}`), turns\nthe cards on at every width (`true`) or keeps the grid (`false`).\n\n<div data-docs-demo=\"713-base-phone-layout\" data-height=\"720\"></div>\n\n## Search across tables\n\nThe view bar's search finds records in the view on screen. The box at the\nend of the tab bar searches every table of the base. With the focus in the\nbase, Ctrl+K puts the cursor in it, and so does \"/\" outside the grid and any\ntext box. Neither is taken from the rest of the page.\n\nIt reads a field the way its cell shows it: a single select by its option's\nlabel, a person by name, a date as the field writes it, a link by the names\nof the records it links to, a formula or lookup by its result. Numbers,\ncurrency, percents, durations, ratings, checkboxes, buttons, rollups and\ncounts are not searched. Case and accents do not count: \"tomas\" finds\n\"Tomás\".\n\nEvery word of the query has to be in the record, in one field or another.\nThe results are grouped by table, five records each with \"N more in Deals\"\nunder them, each with its name, the field the words were found in and the\nwords marked. Records rank by:\n\n1. the whole query in one field, before words spread over several;\n2. the primary field, before the others;\n3. a field that is the query, then one that starts with it, then one with a\n word that starts with it, then one that holds it anywhere.\n\nTables are listed by their best record. Up and Down move through the list,\nEnter or a click shows the record's table and opens the record, and Escape\ncloses the list, then clears the box. `search={false}` on `SvBase` leaves the\nbox out. A shared base (`readOnly`) keeps it, over the tables and fields the\nshare shows.\n\nThe same search runs without the box, for a command palette of your own or\non a server:\n\n```ts\nimport { searchBase } from '@svgrid/enterprise/base'\n\nconst groups = searchBase(tables, 'northwind proposal', { limit: 10 })\nfor (const g of groups) {\n console.log(g.name, g.total)\n for (const hit of g.hits) console.log(hit.rowId, hit.title, `${hit.fieldName}: ${hit.text}`, hit.marks)\n}\n```\n\nEach hit has the table and record ids, the record's name (`title`), the\nfield that matched (`field`, `fieldName`, `primary`), its text cut around\nthe match when it is long, and `marks`, the `[start, end)` spans of the\nwords in that text. `field: (table, column) => boolean` in the options picks\nthe fields to search. To open a hit in `SvBase`, show its table\n(`bind:activeTable`) and call `openRecord({ api, rowId })` once that table's\ngrid has called `onApiReady`; the box does this for you.\n\n<div data-docs-demo=\"718-base-search-all-tables\" data-height=\"620\"></div>\n\n## View types and the keys they read\n\n| Type | Keys | View bar buttons |\n|---|---|---|\n| `'grid'` | `state`, `rowHeight`, `color` | Hide fields, Filter, Group, Sort, Color, Row height |\n| `'kanban'` | `stackBy`: a single select, or a user field with one person | Hide fields, Filter, Sort |\n| `'calendar'` | `dateField`, `endField` (optional), `color` | Hide fields, Filter, Sort, Color |\n| `'gantt'` | `dateField`, `endField`, `progressField`, `dependencyField`, `zoom`, `color` | Hide fields, Filter, Sort, Color, Customize timeline |\n| `'gallery'` | `coverField`: an attachment field | Hide fields, Filter, Sort |\n| `'chart'` | `chart`: `kind`, `x`, `y`, `aggregation` | Filter, Customize chart |\n| `'list'` | `listFields`: the fields under each name | Hide fields, Filter, Group, Sort, Customize list |\n| `'inbox'` | `inboxDateField`: a date field, newest first; `listFields` | Hide fields, Filter, Sort, Customize inbox |\n| `'form'` | `formFields`, `formRequired`, `formTitle`, `formDescription`, `formSubmitLabel`, `formConditions`, `formPrefill` | none |\n| `'dashboard'` | `dashboard`: the tiles, in order | none |\n| any type | `personal`, `locked`, `lockedBy` | off while `locked` |\n\nEvery type also keeps `state` (filter, sorts, groups, hidden fields, column\norder, widths, pinning) and switches with the grid's own state, so a filter\nset on the board is still there when you come back to it. The search text is\nnot part of it. The views menu's \"Create\" offers each type `SvBase` can show\nand greys out one the table cannot: a Kanban needs a single select, a\ncalendar a date field, a chart a field to group by. A new view starts with no\nfilter, sorts or groups and every field shown.\n`viewTypes={['grid', 'kanban', 'calendar']}` on `SvBase` limits what\n\"Create\" offers.\n\nA timeline (`type: 'gantt'`) is a calendar with a task table beside the bars.\n\"Create\" offers it as Timeline in every base, and its renderer loads the first\ntime one is shown. Its bars take the view's colors, `progressField` is a percent field drawn as each\nbar's fill, and `dependencyField` is a link field to the same table naming\nwhat each record waits for, drawn as arrows.\n\nA record's dates are its own: a drop on a Sunday stays on the Sunday. By keyboard a\nbar works as a calendar event does: Space picks it up, Left and Right carry it a day\n(Shift, a week), Space puts it down and Escape puts it back. Enter opens the record,\nShift+F10 opens its menu, and Delete asks first, as a row's menu does.\n\n## When an edit changes where a record belongs\n\nA view's filter, sort and groups apply again after an edit, with one pause. In a\ngrid the row you are working in stays where it is while it is the active row, so\nit does not leap away under the cursor; it takes its place when you move to\nanother row, click outside the grid, or close its record. The record count follows\nat the same moment. An undo, a paste and a fill apply at once. A board, a calendar,\na timeline, a gallery and a list show the change as soon as it is made, or when a\nrecord dialog closes. On a table read a page at a time the database is asked again\nat that moment, once it has the edit.\n\n## The view bar, panel by panel\n\n- **Views menu** (the view's name): switch, rename, duplicate (\"Roadmap by\n quarter copy\") and delete. Delete asks in the view's own row, and nothing moves\n until the answer. A deleted view goes to the trash and is one step of undo. The\n last view cannot be deleted.\n- **Hide fields**: a switch per field, a search, hide all and show all, and\n drag to reorder. The list keeps its order while you switch fields, and hidden\n fields sit where they are in the table. The primary field is always shown.\n- **Filter**: \"Where [field] [operator] [value]\" rows joined by \"and\" or\n \"or\", and condition groups. Each change filters at once. A condition still\n missing its value filters nothing.\n- **Group**: up to three fields, each with a direction.\n- **Sort**: fields in order, each with a direction labeled for its type (\"1\n → 9\", \"A → Z\", \"first → last\" for a single select). Attachment and button\n fields are not offered.\n- **Color**: None, Select field (a single select's option colors) or\n Conditions in order.\n- **Row height**: Short 34, Medium 56, Tall 88, Extra tall 128.\n `rowHeights` on `SvBase` or `SvBaseViewBar` sets the list.\n- **Search**: in a grid view, the grid's find bar (Ctrl+F) that marks every\n match. In a board, calendar, gallery, list or inbox, a \"Find a record\" box\n that narrows the records. Switching views clears it.\n- **Search all tables**: the box in the tab bar (below).\n\nA button with something set says so and is tinted. With `fieldMenu` on, each\ncolumn's menu adds \"Filter by this field\" and \"Group by this field\", which\nopen the bar's panel with the new condition or group in it.\n\n## Filters in code\n\n`filterKindOf(columnDef)` gives the kind a field filters as, and\n`operatorsFor(kind)` the operators the panel offers for it, in menu order.\n[Filter conditions](../fields.md#filter-conditions) maps each field type to\nits kind.\n\n| `kind` | `op` values | `value` |\n|---|---|---|\n| `'text'`, `'link'` | `contains`, `notContains`, `is`, `isNot` | a string; a link compares the linked records' names |\n| `'number'` | `eq`, `neq`, `lt`, `gt`, `lte`, `gte` | a number |\n| `'select'` | `is`, `isNot` | one option value, or a person's id (or `ME`) for a user field |\n| `'select'` | `isAnyOf`, `isNoneOf` | an array of them |\n| `'multi'` | `hasAnyOf`, `hasAllOf`, `isExactly`, `hasNoneOf` | an array of option values or ids |\n| `'checkbox'` | `is` | `true` or `false` |\n| `'date'` | `is`, `isBefore`, `isAfter`, `isOnOrBefore`, `isOnOrAfter` | `{ mode: 'today' }`, with `mode` also `'tomorrow'`, `'yesterday'`, `'oneWeekAgo'`, `'oneWeekFromNow'`, `'oneMonthAgo'`, `'oneMonthFromNow'`; or `{ mode: 'exactDate', date: '2026-11-01' }` |\n| `'date'` | `isWithin` | `{ span: 'pastWeek' }`, with `span` also `'pastMonth'`, `'pastYear'`, `'nextWeek'`, `'nextMonth'`, `'nextYear'`; or `{ span: 'pastDays', days: 14 }`, `{ span: 'nextDays', days: 14 }` |\n| `'files'` | `contains` (file name contains) | a string |\n\nEvery kind except checkbox also takes `isEmpty` and `isNotEmpty`, with no\nvalue. A few rules from the code:\n\n- An empty number cell is no number: it matches only `neq`.\n- Dates compare whole days. A bare `'2026-11-01'` is that day where the user\n is, and \"within the past week\" runs from 7 days ago to today.\n- A user condition holds the person's id (`'u2'`), not the name, or `ME`\n for whoever is signed in.\n- A link condition sets `table` to the linked table's id, so it can match\n records by name.\n\nSet a filter on the shown view with `api.setAdvancedFilter(filter)`; the\nview keeps it like one set in the panel.\n\n## Row colors: by field or by conditions\n\n```ts\nimport type { ColorSetting } from '@svgrid/enterprise/base'\n\nconst byStatus: ColorSetting = { mode: 'field', field: 'status' }\nconst byRisk: ColorSetting = {\n mode: 'conditions',\n rules: [\n { color: '#f82b60', condition: { column: 'status', kind: 'select', op: 'is', value: 'Blocked' } },\n { color: '#fcb400', condition: { column: 'target', kind: 'date', op: 'isBefore', value: { mode: 'today' } } },\n ],\n}\n```\n\nBy field, each row takes its option's color, and the row header's title says\nthe option, so the color is never the only signal. A multiple select colors\nby its first value. By conditions, the first rule that matches wins: a\nblocked feature is red even when it is also late. A condition is tested as\nthe filter tests it: on the value the cell shows, so a formula, lookup,\nrollup or count field works, and a condition still missing its value colors\nnothing. A calendar colors its events by a single select only; a Gantt's bars\ntake either kind.\n\nThe color is a 3px stripe on the row header, unless the `rowColor` prop on\n`SvBase` sets a `target` (`'bar'` or `'row'`): a view's colors use it too. A\nview without `color` shows the `rowColor` prop's colors until someone picks\ncolors for it in Color. `{ mode: 'none' }` turns the colors off in that view.\n\n## Summaries in the footer and in group headers\n\nWith `summary` on, a click on a field's footer cell picks its summary: sum,\naverage, median, min, max or range for numbers; checked, unchecked or percent\nchecked for a checkbox; earliest, latest or the date range in days for dates;\nempty, filled, unique, percent empty or percent filled for any field. The\nresult is written as the field writes values. The choice is stored in the\nfield's `fieldProps.summary` and reported through `onColumnChange`; see\n[Summaries at the bottom](../fields.md#summaries-at-the-bottom).\n\nGroup headers show the same summaries over that group's records. Both leave\nout records the filter hides, even in a collapsed group.\n\n## The view bar over your own grid\n\n`SvBaseViewBar` is the bar alone, for a page that lays out its own\n`<SvGrid>`:\n\n```svelte\n<script lang=\"ts\">\n import { SvGrid } from '@svgrid/grid'\n import { SvBaseViewBar, enableBase, type FieldView } from '@svgrid/enterprise/base'\n\n enableBase()\n // Raw: the bar finds its grid by the api object itself, and a proxy is another object.\n let api = $state.raw<any>(null)\n let views = $state<FieldView[]>([\n { id: 'all', name: 'All features', type: 'grid' },\n { id: 'board', name: 'Status board', type: 'kanban', stackBy: 'status' },\n ])\n let activeView = $state('all')\n</script>\n\n{#if api}\n <SvBaseViewBar {api} bind:views bind:activeView\n rowHeights={[{ label: 'Compact', height: 32 }, { label: 'Roomy', height: 64 }]} />\n{/if}\n<SvGrid data={rows} {columns} primaryField=\"feature\" editable containerHeight={600}\n onApiReady={(a) => (api = a)} />\n```\n\nIts \"Create\" offers grid, Kanban and calendar (and Gantt once enabled). It\nloads the board, calendar and gallery renderers the first time a view of\nthat type is shown, and opens the record and the \"Delete record?\" question\nfor a card's menu. Chart and form views are drawn by `SvBase`, not by the\nbar, so leave them out of `viewTypes` here, or draw `SvBaseForm` yourself\nwhen a form view is active.\n\n## Pitfalls\n\n- **A new view from \"Create\" starts with `color: { mode: 'none' }`**, so it\n hides the `rowColor` prop's colors. A new calendar or Gantt is the\n exception: it starts colored by the first single select. Pick colors in\n Color, or declare the view in code without `color`.\n- **`color.field` is the data field; the other keys are column ids.** They\n differ only when a column sets an `id`.\n- **Summaries belong to the field, not the view.** Picking \"Average\" in one\n view changes every view's footer and group headers.\n- **Save views only after they load.** See the `loaded` check in step 12.\n\n## Next\n\n- [Linked records](./linked-records.md): link Features to Feedback and chart\n or filter by the linked names.\n- [Formulas](./formulas.md): computed fields a view can sort, group and chart\n by.\n- [Forms](./forms.md): required fields, the thank-you screen and forms\n outside the base.\n- [Collaboration](./collaboration.md): comments, people and who changed what.\n- [Automations](./automations.md): conditions written with this same filter.\n- [Keep a base in a database](./data.md): load the tables from Postgres,\n Supabase or your API and write every change back.\n"
5307
+ "markdown": "# Views for every team\n\nOne table of records, many ways to look at it. Product reads the roadmap\ngrouped by quarter, engineering works off a status board, the leads check a\ntimeline, and anyone in the company files a request through a form. Each of\nthose is a saved view: a name, a type, and the filter, sorts, groups, hidden\nfields, row height and colors it had when someone last left it.\n\n![The Features table feeds six saved views: Roadmap by quarter (a grid grouped by Quarter, sorted by Impact, Lead hidden, row height 56, colored by Status), Now (a grid filtered to active work), Status board (a Kanban stacked by Status), Timeline (bars from Start to Target), Effort by team (a bar chart) and Request a feature (a form). onViewsChange hands the whole list to the app, which stores it and loads it back into bind:views.](/docs-media/base-views.svg)\n\n<div data-docs-demo=\"513-base-product-roadmap\" data-height=\"680\"></div>\n\n> SvGrid Base is part of the **Suite edition** of `@svgrid/enterprise`. Without a key it runs with a watermark.\n\nThis page builds the Features table of that roadmap one view at a time. Each\nstep adds to the `views` list from the step before and works on its own.\n\n## 1. The table\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, setFieldUsers, type FieldView } from '@svgrid/enterprise/base'\n\n enableBase()\n setFieldUsers([\n { id: 'u1', name: 'Ana Ruiz' },\n { id: 'u2', name: 'Ben Okafor' },\n { id: 'u3', name: 'Cleo Park' },\n ])\n\n type Feature = {\n id: number; feature: string; quarter: string; status: string; teams: string[]\n lead: string; impact: number; effort: number; start: string; target: string; votes: number\n }\n\n const QUARTERS = [\n { value: 'Q4 2026', color: '#2d7ff9' },\n { value: 'Q1 2027', color: '#18bfff' },\n { value: 'Later', color: '#94a3b8' },\n ]\n const STATUS = [\n { value: 'Planned', color: '#94a3b8' },\n { value: 'In progress', color: '#2d7ff9' },\n { value: 'In beta', color: '#8b46ff' },\n { value: 'Shipped', color: '#20c933' },\n { value: 'Blocked', color: '#f82b60' },\n ]\n const TEAMS = [\n { value: 'Web', color: '#2d7ff9' },\n { value: 'Mobile', color: '#ff6f2c' },\n { value: 'Platform', color: '#8b46ff' },\n ]\n\n let rows = $state.raw<Feature[]>([\n { id: 1, feature: 'Single sign-on with SAML', quarter: 'Q4 2026', status: 'In progress', teams: ['Platform'], lead: 'u2', impact: 5, effort: 12, start: '2026-10-05', target: '2026-11-13', votes: 182 },\n { id: 2, feature: 'Offline mode for the mobile app', quarter: 'Q4 2026', status: 'Blocked', teams: ['Mobile'], lead: 'u3', impact: 4, effort: 13, start: '2026-10-12', target: '2026-12-04', votes: 96 },\n { id: 3, feature: 'Dark mode', quarter: 'Q4 2026', status: 'Shipped', teams: ['Web'], lead: 'u1', impact: 2, effort: 4, start: '2026-10-01', target: '2026-10-16', votes: 154 },\n { id: 4, feature: 'Audit log export', quarter: 'Q1 2027', status: 'Planned', teams: ['Platform', 'Web'], lead: 'u1', impact: 3, effort: 5, start: '2027-01-11', target: '2027-01-29', votes: 41 },\n ])\n\n const columns = [\n { field: 'feature', header: 'Feature', width: 260, fieldType: 'text' },\n { field: 'quarter', header: 'Quarter', fieldType: 'singleSelect', fieldProps: { options: QUARTERS } },\n { field: 'status', header: 'Status', fieldType: 'singleSelect', fieldProps: { options: STATUS } },\n { field: 'teams', header: 'Teams', fieldType: 'multiSelect', fieldProps: { options: TEAMS } },\n { field: 'lead', header: 'Lead', fieldType: 'user' },\n { field: 'impact', header: 'Impact', fieldType: 'rating', fieldProps: { max: 5 } },\n { field: 'effort', header: 'Effort (days)', fieldType: 'number', fieldProps: { format: 'integer', summary: 'sum' } },\n { field: 'start', header: 'Start', fieldType: 'date', fieldProps: { dateFormat: 'friendly' } },\n { field: 'target', header: 'Target', fieldType: 'date', fieldProps: { dateFormat: 'friendly' } },\n { field: 'votes', header: 'Votes', fieldType: 'number', fieldProps: { format: 'integer', summary: 'sum' } },\n ]\n\n let views = $state<FieldView[]>([{ id: 'all', name: 'All features', type: 'grid' }])\n</script>\n\n<SvBase data={rows} {columns} bind:views primaryField=\"feature\"\n editable sortable showRowNumbers expandRows summary containerHeight={640}\n rowAdder={{ label: 'Add feature', defaults: () => ({ id: Date.now(), status: 'Planned', quarter: 'Later', teams: [] }) }} />\n```\n\nThe view bar over the grid has \"All features\" on the left, then Hide\nfields, Filter, Group, Sort, Color and a row height button, and \"4 records\"\nwith a search button on the right. The footer row shows \"Sum 34\" under Effort\nand \"Sum 473\" under Votes, because those fields set `fieldProps.summary`.\n\nEvery view key that names a field (`stackBy`, `dateField`, a filter\ncondition's `column`, a chart's `x`) takes the column id. A column without an\n`id` uses its `field`, so in this table the two are the same.\n\n## 2. A roadmap grouped by quarter\n\nA view can start with settings. `state` is the grid's own state, the same\nobject `api.getState()` returns; `color` says how rows are colored.\n\n```ts\nconst roadmap: FieldView = {\n id: 'roadmap',\n name: 'Roadmap by quarter',\n type: 'grid',\n color: { mode: 'field', field: 'status' },\n rowHeight: 56,\n state: {\n grouping: ['quarter'],\n sorting: [{ id: 'quarter', desc: false }, { id: 'impact', desc: true }],\n hiddenColumns: ['lead'],\n },\n}\n```\n\nGroups and sorts share the grid's `sorting` list: a grouped field's\ndirection is an entry ahead of the sorts. The view bar writes them that way,\nso write them that way too. Here the bar reads \"1 hidden field\", \"Grouped by\n1 field\" and \"Sorted by 1 field\": the `quarter` entry belongs to the group,\nnot to Sort.\n\nEach group header shows \"Quarter\" over the option's pill, the number of\nrecords, the summaries the fields picked for that group (\"Effort (days) 29\",\n\"Votes 432\"), and \"+\" to add a feature that already has that quarter. Groups\ncome in the select's option order, not alphabetically. Group rows are at\nleast 52px tall so the field's name fits over the value; with\n`rowHeight: 56` they match the records. Each row has a status color stripe\non its row header. 56 is the row height menu's \"Medium\": from 48px, text and\nselect pills wrap.\n\n## 3. A view that starts filtered\n\nThe filter is the grid's advanced filter with its own kind, `fieldFilter`.\nPut it in `state.advancedFilter`:\n\n```ts\nimport { describeFieldFilter, type FieldFilter } from '@svgrid/enterprise/base'\n\nconst activeWork: FieldFilter = {\n kind: 'fieldFilter',\n conjunction: 'and',\n conditions: [\n { column: 'status', kind: 'select', op: 'isAnyOf', value: ['In progress', 'In beta', 'Blocked'] },\n {\n conjunction: 'or',\n conditions: [\n { column: 'target', kind: 'date', op: 'isWithin', value: { span: 'nextDays', days: 30 } },\n { column: 'impact', kind: 'number', op: 'gte', value: 4 },\n ],\n },\n ],\n}\n\nconst now: FieldView = {\n id: 'now',\n name: 'Now',\n type: 'grid',\n state: { sorting: [{ id: 'target', desc: false }], advancedFilter: activeWork },\n}\n\nconst header = (id: string) => columns.find((c) => c.field === id)?.header ?? id\ndescribeFieldFilter(activeWork, header)\n// \"Status is any of In progress, In beta, Blocked and (Target is within the next 30 days or Impact ≥ 4)\"\n```\n\nThe Filter button reads \"Filtered by 3 fields\" (one field reads \"Filtered by\nStatus\") and the record count drops to what the filter keeps.\n`describeFieldFilter` writes a filter in the words the panel uses, for a page\nsubtitle or an audit line. It returns `\"every record\"` for a filter with no\ncomplete conditions.\n\nA condition is a field, an operator and a value; a group is one level deep.\n[Filters in code](#filters-in-code) below lists the operators per kind.\n\nOne view can serve every lead. `ME` (the string `'@me'`) in a person\ncondition stands for whoever `setCurrentUser` names when the filter runs; the\nFilter panel lists it as \"Me\", first among the people.\n\n```ts\nimport { ME, setCurrentUser } from '@svgrid/enterprise/base'\n\nsetCurrentUser({ id: 'u2', name: 'Ben Okafor' })\n\nconst mine: FieldView = {\n id: 'mine',\n name: 'My features',\n type: 'grid',\n state: {\n advancedFilter: {\n kind: 'fieldFilter',\n conjunction: 'and',\n conditions: [{ column: 'lead', kind: 'select', op: 'is', value: ME }],\n },\n },\n}\n```\n\nBen sees \"Single sign-on with SAML\"; Ana, signed in, sees her two. With\nnobody signed in the condition matches no one. `describeFieldFilter` writes\nit as \"Lead is me\".\n\n## 4. A status board\n\n```ts\nconst board: FieldView = { id: 'board', name: 'Status board', type: 'kanban', stackBy: 'status' }\n```\n\n![The roadmap's Status board: Planned, In progress, In beta and Shipped lanes of feature cards showing Quarter, Teams and Lead, with the right-click menu open on \"Single sign-on with SAML\": Expand record, Comment, Move to, Duplicate record, Delete record.](/docs-media/base-views.png)\n\n`stackBy` takes a single select or a user field that holds one person. You\nget one lane per option, in option order and color, plus \"Uncategorized\",\nwhich shows only while a record has no status. Stack by `lead` instead and\nthe lanes are the people from `setFieldUsers`.\n\nSort the view by a number field and a card dropped elsewhere in its lane takes\na number between its neighbours', so the order is the table's and lasts.\n\nDrop a card in another lane and the status changes through the grid's edit\npath, so undo, validation and `onCellValueChange` all see it. \"+ Add a card\"\nadds a record already in that lane, named by the primary field.\n\nA card shows the record's name, then the next three fields the view shows,\nin column order (the stacked field left out), each drawn as its cell draws\nit. Hide a field in Hide fields and it leaves the cards; drag it up there to\nput it on them. Cards are read, not edited: a rating shows its stars\nbut does not take a click. Double-click a card or press F2 to open the\nrecord. Right-click it for its menu: Expand record, Comment, and, when the\nbase is `editable`, Move to (a submenu of the lanes), Duplicate record and\nDelete record, which asks \"Delete record?\" first. A card with comments shows\ntheir count; click it to open the record at its comment box.\n\n## 5. A timeline\n\n```ts\nconst timeline: FieldView = {\n id: 'timeline',\n name: 'Timeline',\n type: 'gantt',\n dateField: 'start',\n endField: 'target',\n color: { mode: 'field', field: 'status' },\n}\n```\n\nEach feature is a bar from its start to its target date, beside a table of\nnames and dates. Drag a bar to move both dates; drag its edge to change the\ntarget. One drag is one Ctrl+Z. `type: 'calendar'` with the same fields shows\nthe features on a month instead: a date field without a time shows as all-day\nevents. The end of a bar or an event is exclusive, so the record keeps the last\nday: a feature dragged to cover the 19th to the 21st gets\n`target: '2026-10-21'`.\n\nWith only `dateField`, events move but cannot be resized. Pick a day (or a\nrange) on the calendar and a new record lands on it, opened with its name\nready to type. Colored by a single select, the events take the option colors\nand the options appear above the calendar as a legend that hides and shows\nthem.\n\nMoving and resizing need `editable` on the base. Without it the calendar is\nread-only.\n\n## 6. Effort by team\n\n```ts\nconst effort: FieldView = { id: 'effort', name: 'Effort by team', type: 'chart', chart: { kind: 'barH', x: 'teams', y: 'effort' } }\nconst mix: FieldView = { id: 'mix', name: 'Features per quarter', type: 'chart', chart: { kind: 'donut', x: 'quarter' } }\n```\n\nA chart groups the records the view shows by `x` and adds up `y` (or counts\nrecords when there is no `y`). `aggregation: 'avg'` averages instead of\nsumming. Kinds are `'bar'`, `'barH'`, `'line'`, `'pie'` and `'donut'`.\n\nHow `x` groups:\n\n- A single or multiple select keeps its options' order and colors. A\n multiple select counts a record in each of its options, so \"Audit log\n export\" adds 5 days to both Platform and Web.\n- A person or a linked record shows by name.\n- A date groups by month (\"Oct 2026\").\n- Records with no value go in an \"Empty\" bar, last.\n\nValues read as the summed field writes them, and a count as a whole number.\nThe series is labeled \"Sum of Effort (days)\", \"Average of ...\" or \"Records\".\nThe bar shows Filter, which narrows what is charted, and \"Customize chart\",\nwhose Type, Group by, Value and Combine (sum or average) are kept in `chart`.\nA chart view declared without `chart` draws \"Choose a field to chart in\nCustomize chart.\" until a field is picked there.\n\n## 7. A gallery of mockups\n\nAdd an attachment field, then a gallery view with it as the cover:\n\n```ts\nconst mockupField = { field: 'mockups', header: 'Mockups', fieldType: 'attachment' }\nconst gallery: FieldView = { id: 'mockups', name: 'Mockups', type: 'gallery', coverField: 'mockups' }\n```\n\nEach card shows the first image of `mockups`, the primary field as its title,\nthen the fields the grid shows, each drawn as its cell draws it. Hide a field\nin this view and it leaves the cards; drag fields in Hide fields to reorder\nthem. A click on a card opens the record. Without `coverField` the cards have\nno cover.\n\n## 8. A request form\n\n```ts\nconst request: FieldView = {\n id: 'request',\n name: 'Request a feature',\n type: 'form',\n formFields: ['feature', 'teams', 'impact'],\n formRequired: ['feature'],\n formTitle: 'Request a feature',\n formDescription: 'What should we build, and who is it for? Product reviews requests every Friday.',\n formSubmitLabel: 'Send request',\n}\n```\n\nThe form asks for the fields in `formFields` (column ids, in order), each with\nthe editor its cell uses. Without it, the form asks for every field a person\ncan fill in, the primary field first; computed fields are left out. A field\nin `formRequired` must be filled before the form sends. `formTitle` defaults\nto the view's name, `formSubmitLabel` to \"Submit\". A sent form adds a record\nthe way \"+ Add row\" does: `rowAdder.defaults` first, so a request starts as\nPlanned in Later, then what was typed, then `onRowAdd`. The bar shows only\nthe view switcher on a form.\n\nAll the views so far, in one list:\n\n```ts\nlet views = $state<FieldView[]>([roadmap, board, timeline, effort, now, mine, gallery, request])\n```\n\n## 9. A dashboard\n\n```ts\nconst OPEN: FieldFilter = { kind: 'fieldFilter', conjunction: 'and', conditions: [{ column: 'status', kind: 'select', op: 'isNot', value: 'Shipped' }] }\nconst SHIPPED: FieldFilter = { kind: 'fieldFilter', conjunction: 'and', conditions: [{ column: 'status', kind: 'select', op: 'is', value: 'Shipped' }] }\n\nconst overview: FieldView = {\n id: 'overview',\n name: 'Overview',\n type: 'dashboard',\n dashboard: [\n { id: 'open', kind: 'number', title: 'Open features', filter: OPEN },\n { id: 'days', kind: 'number', title: 'Days of work left', aggregation: 'sum', field: 'effort', filter: OPEN, caption: 'Team capacity 40 days' },\n { id: 'shipped', kind: 'progress', title: 'Shipped', filter: SHIPPED },\n { id: 'status', kind: 'chart', chart: { kind: 'bar', x: 'status' } },\n { id: 'wanted', kind: 'list', title: 'Most wanted', sortBy: 'votes', sortDesc: true, limit: 3 },\n ],\n}\n```\n\nA dashboard is a grid of tiles. Each tile reads the records of one table,\nkeeps the ones its `filter` matches (the same `FieldFilter` as a view's\nfilter), and shows one of four things:\n\n- **`'number'`** counts the records, or with `aggregation: 'sum' | 'avg' |\n 'min' | 'max'` and a number `field`, adds them up. The number is written the\n way the field writes it, so a currency field shows \"$42,000\". Here \"Open\n features\" reads 3 and \"Days of work left\" reads 30. `caption` is a line\n under the number.\n- **`'chart'`** takes a chart view's `chart` setting (`kind`, `x`, `y`,\n `aggregation`) and groups the tile's records the same way.\n- **`'list'`** shows the first `limit` records (5 by default) sorted by\n `sortBy`, with that field's value beside each name (\"Single sign-on with\n SAML 182\"); `detail` shows another field instead. A click opens the record.\n- **`'progress'`** shows how many of all the table's records match the filter:\n \"25%, 1 of 4 records\" for Shipped.\n\n`size` is `'1x1'`, `'2x1'` or `'2x2'` (columns by rows). A number or progress\ntile is `'1x1'` by default, a chart or list `'2x2'`. Below about 450px the\ntiles stack in one column. A tile without `title` is named after what it\nshows (\"Total Effort (days)\", \"Status is Shipped\").\n\nWith several tables (`<SvBase bind:tables>`), a tile's `table` names any\ntable of the base, so a dashboard on Deals can count Accounts and list the\nActivities due this week. A tile without `table` reads the table the view is\non. Its filter and fields are that table's.\n\nA number or progress tile is a button: a click, or Enter, lists the records\nit counts (\"Open the 3 records\") in a dialog, each with its name and a few\nof its fields. On a chart, a click on a bar or a slice lists the records in\nit: the records the tile's filter keeps that fall in that bucket. A click on a record in the dialog opens it, in its\nown table's tab when the tile reads another table. While the dashboard is\nbeing edited, the tiles open nothing.\n\n\"Edit dashboard\" at the top of the view adds a tile, and on each tile sets its\nsize, moves it (drag it, or the arrows), edits it in a dialog (kind, table,\nvalue, filter, title, size) or removes it. Every change is saved in the view's\n`dashboard` at once and reaches `onViewsChange` and `onTablesChange`. A locked\nview has no \"Edit dashboard\". The view bar shows only the view switcher on a\ndashboard.\n\n<div data-docs-demo=\"721-base-dashboard\" data-height=\"680\"></div>\n\n### Text, buttons, images, dividers and a record picker\n\nA dashboard can also say, show and link to things, which makes it a page of\nan app rather than only a summary (Airtable's \"blank\" interface layout):\n\n```ts\nconst home: FieldView = {\n id: 'home',\n name: 'Home',\n type: 'dashboard',\n dashboard: [\n { id: 'hello', kind: 'text', text: '# Welcome\\nYour open requests are below. See [the guide](https://example.com/guide).' },\n { id: 'help', kind: 'button', label: 'Ask for help', url: 'https://example.com/help' },\n { id: 'logo', kind: 'image', url: '/img/office.jpg', alt: 'Our office', fit: 'cover' },\n { id: 'sec', kind: 'divider', title: 'By account' },\n { id: 'account', kind: 'picker', table: 'accounts', title: 'Account', fields: ['city', 'owner'] },\n { id: 'deals', kind: 'number', title: 'Deals', pickedBy: { tile: 'account', field: 'account' } },\n ],\n}\n```\n\n- **`'text'`**: `text` in a small part of Markdown: a line starting with `#`\n is a heading, a blank line starts a paragraph, `**bold**` and\n `[words](https://...)` links inside lines. It is drawn as text, never as\n HTML, so what anyone types shows as typed.\n- **`'button'`**: `label` (default \"Open\") and `url`. An address on another\n site opens in a new tab; a path (`/...`) in the same one.\n- **`'image'`**: `url` and `alt` (said to people who cannot see it); `fit` is\n `'contain'` (the whole picture, default) or `'cover'` (the tile filled).\n- **`'divider'`**: a line across the whole dashboard with its `title` over\n it, to start a section.\n- **`'picker'`**: the records of its `table` to choose from by name; the one\n picked shows its `fields` (default: the next four after its name) and an\n \"Open the record\" button.\n\nLinks and pictures go only to `https:`, `http:`, `mailto:` (a link) or a path\non the site: a `javascript:` or `data:` address is shown as words, not\nfollowed.\n\nA number, chart, list or progress tile with `pickedBy: { tile, field }` is\nnarrowed by that picker: once a record is picked, it counts only the records\nwhose link field `field` holds it (\"the deals of the account picked\"), and\n\"Every record\" counts them all again. The pick is the viewer's and is not\nsaved. Under the hood the tile's filter gains a condition of the kind\n`'linked'`:\n\n```ts\nconst narrowing = { column: 'account', kind: 'linked', op: 'hasAnyOf', value: ['a1'] } // the link holds record a1, by id\n```\n\nFor a table the page holds, the page applies it. For a table read from a\nserver (`serverRows`), the server must apply it too: `fieldFilterToSql` and\n`fieldFilterToPostgrest` answer that they cannot (the tile then shows what\nthe server sends), so a server of your own reads `kind: 'linked'` and matches\nthe link's record ids.\n\nIn the tile dialog, \"Shows\" offers these kinds too; a picker's tile picks its\nfields, and the tiles it can narrow offer \"Narrowed by\" with each picker the\ntile's table links to. Dividers take one row across the dashboard; the other\ntiles keep their sizes.\n\n## 10. A list for the phone\n\n```ts\nconst list: FieldView = {\n id: 'list',\n name: 'Feature list',\n type: 'list',\n listFields: ['status', 'teams', 'votes'],\n state: { sorting: [{ id: 'quarter', desc: false }, { id: 'votes', desc: true }], grouping: ['quarter'], hiddenColumns: [], advancedFilter: null },\n}\n```\n\nA list shows one record per entry: the primary field in bold, and under it\nthe fields in `listFields` (column ids, in order), each drawn as its cell\ndraws it, so Status is a colored pill and Teams are tags. An empty value is\nleft out. Without `listFields` the list shows the next three fields the grid\nshows. Long names stay readable at phone width, where a grid would cut them.\n\nThe list keeps the view's filter, sorts and groups. Here it is grouped by\nQuarter: a header per quarter with the field's name, its value and the\nnumber of records, \"Quarter Q4 2026 3\", and a click on the header folds the\ngroup. Within each quarter the most voted feature comes first. Grouped by\nmore fields (`grouping: ['quarter', 'status']`), each quarter holds a header\nper status, indented under it, with its own count and its own fold.\n\nA click on a record, or Enter, opens it. The list is one tab stop: Up and\nDown move between records and group headers at every level, Home and End\njump, Left and Right fold and unfold a group, and Left on a record or a\nfolded group moves to the group it is in. A record someone else changed and nobody here\nopened yet has a dot before its name.\n\nThe view bar has Hide fields, Filter, Group, Sort and \"Customize list\", which\nturns the fields under the name on and off and saves them in `listFields`.\n\n<div data-docs-demo=\"719-base-list-view\" data-height=\"620\"></div>\n\n## 11. An inbox of changes\n\n```ts\nconst inbox: FieldView = {\n id: 'inbox',\n name: 'Inbox',\n type: 'inbox',\n inboxDateField: 'start',\n listFields: ['status', 'lead'],\n}\n```\n\nAn inbox shows the records on the left and the picked one on the right: its\nfields, each one editable in place with the editor its cell uses, and its\nlatest changes, \"Cleo Park changed Status: Planned → In progress\". \"Expand\"\nopens the full record with its comments.\n\nThe list is newest first by `inboxDateField` (a date field's column id), so\n\"Audit log export\", which starts on January 11, is on top.\nRecords without a date go last. Without `inboxDateField` the inbox keeps the\nview's sort. The view's filter applies either way.\n\nA record someone else changed and nobody here has opened is unread: its name\nis bold and it has a dot. The changes come from a table with a `source` that\nhears of writes made elsewhere ([Changes made\nelsewhere](./data.md#changes-made-elsewhere)). Picking a record marks it read,\nas opening it in any view does, and \"Mark all read\" marks the rest. The count\nat the top says \"2 unread\" or \"All read\".\n\nDown or j picks the next record and Up or k the previous one; Enter opens the\nfull record. Narrower than 640px the inbox shows the list, and a picked record\nfills it with a Back button (Escape goes back too).\n\nThe view bar has Hide fields, Filter, Sort and \"Customize inbox\": the date\nfield it lists by and the fields under each name. A field hidden in the view\nleaves the record on the right too.\n\n<div data-docs-demo=\"720-base-inbox-view\" data-height=\"620\"></div>\n\n## 12. Keep the views\n\n`onViewsChange` fires with the whole list when a view is created, renamed,\nduplicated or deleted, and about 400ms after the shown view's settings stop\nchanging. Switching views stores the one being left first. Store the list\nwherever the app stores things.\n\nIn the browser:\n\n```svelte\n<script lang=\"ts\">\n import { SvBase, enableBase, type FieldView } from '@svgrid/enterprise/base'\n\n enableBase()\n const KEY = 'roadmap:views'\n const DEFAULT_VIEWS: FieldView[] = [{ id: 'all', name: 'All features', type: 'grid' }]\n\n function read<T>(key: string, fallback: T): T {\n try {\n const raw = localStorage.getItem(key)\n return raw ? (JSON.parse(raw) as T) : fallback\n } catch {\n return fallback\n }\n }\n function write(key: string, value: unknown) {\n try {\n localStorage.setItem(key, JSON.stringify(value))\n } catch {\n // Private mode or a full quota: the views still work, unsaved.\n }\n }\n\n let views = $state<FieldView[]>(read(KEY, DEFAULT_VIEWS))\n let activeView = $state<string>(read(`${KEY}:active`, ''))\n\n $effect(() => write(`${KEY}:active`, activeView))\n</script>\n\n<SvBase data={rows} {columns} primaryField=\"feature\" editable\n bind:views bind:activeView onViewsChange={(v: FieldView[]) => write(KEY, v)} />\n```\n\nOn a server, decide who owns a list.\n\n- **Per user.** Each person keeps their own list, so one lead's filter does\n not change another's. Seed it from the team's defaults the first time.\n- **Shared.** One list per table that everyone reads. Write it only for the\n people you allow; for everyone else, keep their changes in memory or in\n their own list. A shared \"My features\" view built on `ME` still shows each\n person their own records.\n\n```ts\ntype Scope = 'me' | 'team'\n\nasync function loadViews(scope: Scope): Promise<FieldView[]> {\n const res = await fetch(`/api/tables/features/views?scope=${scope}`)\n return res.ok ? ((await res.json()) as FieldView[]) : DEFAULT_VIEWS\n}\n\nfunction saveViews(scope: Scope, next: FieldView[]) {\n void fetch(`/api/tables/features/views?scope=${scope}`, {\n method: 'PUT',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify(next),\n })\n}\n```\n\n```svelte\n{#await loadViews('me') then loaded}\n <SvBase data={rows} {columns} primaryField=\"feature\" editable\n views={loaded} onViewsChange={(v: FieldView[]) => saveViews('me', v)} />\n{/await}\n```\n\nOr mount at once and swap the list in when it arrives. When `views` or\n`activeView` changes from outside and the shown view is not the one the grid\nhas, the bar applies it. It never stores the grid's settings into a view it\nhas not applied yet.\n\n```svelte\n<script lang=\"ts\">\n let views = $state<FieldView[]>(DEFAULT_VIEWS)\n let loaded = false\n loadViews('me').then((v) => {\n views = v\n loaded = true\n })\n</script>\n\n<SvBase data={rows} {columns} primaryField=\"feature\" editable\n bind:views onViewsChange={(v: FieldView[]) => loaded && saveViews('me', v)} />\n```\n\nThe `loaded` check matters: about 400ms after mount the bar stores the\ndefault view's settings, and that save must not replace the list on the\nserver before the real one is in.\n\nWith several tables (`<SvBase bind:tables>`), each table keeps its own\n`views`, and `onTablesChange` hands back the whole list of tables, views\nincluded.\n\n## Personal and locked views\n\nOnce several people share a table, two more keys on a view matter:\n\n```ts\nconst views: FieldView[] = [\n // Only Ana sees it, under \"My views\" in the views menu.\n { id: 'ana-open', name: 'My open tasks', type: 'grid', personal: 'u1', state: { /* ... */ } },\n // Everyone sees it; nobody changes its settings until someone unlocks it.\n { id: 'report', name: 'Leadership report', type: 'grid', locked: true, lockedBy: 'u2', state: { /* ... */ } },\n]\n```\n\n- **`personal: userId`** lists the view only for that person, read from\n `setCurrentUser`. With nobody signed in every view is listed. \"Create\" in\n the views menu has a \"Personal view: only I see it\" box once someone is\n signed in.\n- **`locked: true`** keeps the view's filter, sorts, groups, hidden fields,\n colors and row height as saved. Its settings buttons are off, nothing is\n saved into it, and the bar shows \"Locked\" with who locked it. A sort from a\n column header still works for the moment; the bar then says \"Locked, not\n saved\" with a Reset. Records stay editable. A locked view cannot be\n deleted, and a copy of it starts unlocked.\n\nThe lock button in the views menu locks the view as the grid shows it now,\nor unlocks it. `lockViews={false}` on `SvBase` removes it for people who may\nnot lock views. Both keys come back through `onViewsChange` and\n`onTablesChange`, so the app stores a personal view with its owner the same\nway as the others.\n\n<div data-docs-demo=\"710-base-personal-locked-views\" data-height=\"620\"></div>\n\n## On a phone\n\nWhen a table is narrower than 600px (its own width, not the window's), a\ngrid view shows each record as a card: the primary field, then the next four\nfields the view shows, in the view's filter and sort. A tap opens the\nrecord, \"+\" adds one, and on a phone screen the opened record fills it. The\nview bar keeps its icons and the tab bar its tabs.\n\nA card keeps what a row can do. A long press, the button on the card, a\nright-click, the Menu key or Shift+F10 opens the record's menu: expand, comment,\nselect, insert above or below, duplicate and delete. A record selected this way\nshows on its card, as a ring on its list row, and in the selection bar. Copy and\npaste work on a cell, so they stay with the table. A gallery view and a list view\nhave the same menu.\n\nA Kanban lane takes the table's width, the next lane a swipe away and\nsnapped into place. A calendar opens in its agenda, the days listed one under\nanother, since a month of seven columns has no room for a title; Month, Week\nand Day are a tap away. Gallery cards stand one per row, and a form is one\ncolumn at any width.\n\n`phoneLayout` on `SvBase` moves the breakpoint (`phoneLayout={480}`), turns\nthe cards on at every width (`true`) or keeps the grid (`false`).\n\n<div data-docs-demo=\"713-base-phone-layout\" data-height=\"720\"></div>\n\n## Search across tables\n\nThe view bar's search finds records in the view on screen. The box at the\nend of the tab bar searches every table of the base. With the focus in the\nbase, Ctrl+K puts the cursor in it, and so does \"/\" outside the grid and any\ntext box. Neither is taken from the rest of the page.\n\nIt reads a field the way its cell shows it: a single select by its option's\nlabel, a person by name, a date as the field writes it, a link by the names\nof the records it links to, a formula or lookup by its result. Numbers,\ncurrency, percents, durations, ratings, checkboxes, buttons, rollups and\ncounts are not searched. Case and accents do not count: \"tomas\" finds\n\"Tomás\".\n\nEvery word of the query has to be in the record, in one field or another.\nThe results are grouped by table, five records each with \"N more in Deals\"\nunder them, each with its name, the field the words were found in and the\nwords marked. Records rank by:\n\n1. the whole query in one field, before words spread over several;\n2. the primary field, before the others;\n3. a field that is the query, then one that starts with it, then one with a\n word that starts with it, then one that holds it anywhere.\n\nTables are listed by their best record. Up and Down move through the list,\nEnter or a click shows the record's table and opens the record, and Escape\ncloses the list, then clears the box. `search={false}` on `SvBase` leaves the\nbox out. A shared base (`readOnly`) keeps it, over the tables and fields the\nshare shows.\n\nThe same search runs without the box, for a command palette of your own or\non a server:\n\n```ts\nimport { searchBase } from '@svgrid/enterprise/base'\n\nconst groups = searchBase(tables, 'northwind proposal', { limit: 10 })\nfor (const g of groups) {\n console.log(g.name, g.total)\n for (const hit of g.hits) console.log(hit.rowId, hit.title, `${hit.fieldName}: ${hit.text}`, hit.marks)\n}\n```\n\nEach hit has the table and record ids, the record's name (`title`), the\nfield that matched (`field`, `fieldName`, `primary`), its text cut around\nthe match when it is long, and `marks`, the `[start, end)` spans of the\nwords in that text. `field: (table, column) => boolean` in the options picks\nthe fields to search. To open a hit in `SvBase`, show its table\n(`bind:activeTable`) and call `openRecord({ api, rowId })` once that table's\ngrid has called `onApiReady`; the box does this for you.\n\n<div data-docs-demo=\"718-base-search-all-tables\" data-height=\"620\"></div>\n\n## View types and the keys they read\n\n| Type | Keys | View bar buttons |\n|---|---|---|\n| `'grid'` | `state`, `rowHeight`, `color` | Hide fields, Filter, Group, Sort, Color, Row height |\n| `'kanban'` | `stackBy`: a single select, or a user field with one person | Hide fields, Filter, Sort |\n| `'calendar'` | `dateField`, `endField` (optional), `color` | Hide fields, Filter, Sort, Color |\n| `'gantt'` | `dateField`, `endField`, `progressField`, `dependencyField`, `zoom`, `color` | Hide fields, Filter, Sort, Color, Customize timeline |\n| `'gallery'` | `coverField`: an attachment field | Hide fields, Filter, Sort |\n| `'chart'` | `chart`: `kind`, `x`, `y`, `aggregation` | Filter, Customize chart |\n| `'list'` | `listFields`: the fields under each name | Hide fields, Filter, Group, Sort, Customize list |\n| `'inbox'` | `inboxDateField`: a date field, newest first; `listFields` | Hide fields, Filter, Sort, Customize inbox |\n| `'form'` | `formFields`, `formRequired`, `formTitle`, `formDescription`, `formSubmitLabel`, `formConditions`, `formPrefill` | none |\n| `'dashboard'` | `dashboard`: the tiles, in order | none |\n| any type | `personal`, `locked`, `lockedBy` | off while `locked` |\n\nEvery type also keeps `state` (filter, sorts, groups, hidden fields, column\norder, widths, pinning) and switches with the grid's own state, so a filter\nset on the board is still there when you come back to it. The search text is\nnot part of it. The views menu's \"Create\" offers each type `SvBase` can show\nand greys out one the table cannot: a Kanban needs a single select, a\ncalendar a date field, a chart a field to group by. A new view starts with no\nfilter, sorts or groups and every field shown.\n`viewTypes={['grid', 'kanban', 'calendar']}` on `SvBase` limits what\n\"Create\" offers.\n\nA timeline (`type: 'gantt'`) is a calendar with a task table beside the bars.\n\"Create\" offers it as Timeline in every base, and its renderer loads the first\ntime one is shown. Its bars take the view's colors, `progressField` is a percent field drawn as each\nbar's fill, and `dependencyField` is a link field to the same table naming\nwhat each record waits for, drawn as arrows.\n\nA record's dates are its own: a drop on a Sunday stays on the Sunday. By keyboard a\nbar works as a calendar event does: Space picks it up, Left and Right carry it a day\n(Shift, a week), Space puts it down and Escape puts it back. Enter opens the record,\nShift+F10 opens its menu, and Delete asks first, as a row's menu does.\n\n## When an edit changes where a record belongs\n\nA view's filter, sort and groups apply again after an edit, with one pause. In a\ngrid the row you are working in stays where it is while it is the active row, so\nit does not leap away under the cursor; it takes its place when you move to\nanother row, click outside the grid, or close its record. The record count follows\nat the same moment. An undo, a paste and a fill apply at once. A board, a calendar,\na timeline, a gallery and a list show the change as soon as it is made, or when a\nrecord dialog closes. On a table read a page at a time the database is asked again\nat that moment, once it has the edit.\n\n## The view bar, panel by panel\n\n- **Views menu** (the view's name): switch, rename, duplicate (\"Roadmap by\n quarter copy\") and delete. Delete asks in the view's own row, and nothing moves\n until the answer. A deleted view goes to the trash and is one step of undo. The\n last view cannot be deleted.\n- **Hide fields**: a switch per field, a search, hide all and show all, and\n drag to reorder. The list keeps its order while you switch fields, and hidden\n fields sit where they are in the table. The primary field is always shown.\n- **Filter**: \"Where [field] [operator] [value]\" rows joined by \"and\" or\n \"or\", and condition groups. Each change filters at once. A condition still\n missing its value filters nothing.\n- **Group**: up to three fields, each with a direction.\n- **Sort**: fields in order, each with a direction labeled for its type (\"1\n → 9\", \"A → Z\", \"first → last\" for a single select). Attachment and button\n fields are not offered.\n- **Color**: None, Select field (a single select's option colors) or\n Conditions in order.\n- **Row height**: Short 34, Medium 56, Tall 88, Extra tall 128.\n `rowHeights` on `SvBase` or `SvBaseViewBar` sets the list.\n- **Search**: in a grid view, the grid's find bar (Ctrl+F) that marks every\n match. In a board, calendar, gallery, list or inbox, a \"Find a record\" box\n that narrows the records. Switching views clears it.\n- **Search all tables**: the box in the tab bar (below).\n\nA button with something set says so and is tinted. With `fieldMenu` on, each\ncolumn's menu adds \"Filter by this field\" and \"Group by this field\", which\nopen the bar's panel with the new condition or group in it.\n\n## Filters in code\n\n`filterKindOf(columnDef)` gives the kind a field filters as, and\n`operatorsFor(kind)` the operators the panel offers for it, in menu order.\n[Filter conditions](../fields.md#filter-conditions) maps each field type to\nits kind.\n\n| `kind` | `op` values | `value` |\n|---|---|---|\n| `'text'`, `'link'` | `contains`, `notContains`, `is`, `isNot` | a string; a link compares the linked records' names |\n| `'number'` | `eq`, `neq`, `lt`, `gt`, `lte`, `gte` | a number |\n| `'select'` | `is`, `isNot` | one option value, or a person's id (or `ME`) for a user field |\n| `'select'` | `isAnyOf`, `isNoneOf` | an array of them |\n| `'multi'` | `hasAnyOf`, `hasAllOf`, `isExactly`, `hasNoneOf` | an array of option values or ids |\n| `'checkbox'` | `is` | `true` or `false` |\n| `'date'` | `is`, `isBefore`, `isAfter`, `isOnOrBefore`, `isOnOrAfter` | `{ mode: 'today' }`, with `mode` also `'tomorrow'`, `'yesterday'`, `'oneWeekAgo'`, `'oneWeekFromNow'`, `'oneMonthAgo'`, `'oneMonthFromNow'`; or `{ mode: 'exactDate', date: '2026-11-01' }` |\n| `'date'` | `isWithin` | `{ span: 'pastWeek' }`, with `span` also `'pastMonth'`, `'pastYear'`, `'nextWeek'`, `'nextMonth'`, `'nextYear'`; or `{ span: 'pastDays', days: 14 }`, `{ span: 'nextDays', days: 14 }` |\n| `'files'` | `contains` (file name contains) | a string |\n\nEvery kind except checkbox also takes `isEmpty` and `isNotEmpty`, with no\nvalue. A few rules from the code:\n\n- An empty number cell is no number: it matches only `neq`.\n- Dates compare whole days. A bare `'2026-11-01'` is that day where the user\n is, and \"within the past week\" runs from 7 days ago to today.\n- A user condition holds the person's id (`'u2'`), not the name, or `ME`\n for whoever is signed in.\n- A link condition sets `table` to the linked table's id, so it can match\n records by name.\n\nSet a filter on the shown view with `api.setAdvancedFilter(filter)`; the\nview keeps it like one set in the panel.\n\n## Row colors: by field or by conditions\n\n```ts\nimport type { ColorSetting } from '@svgrid/enterprise/base'\n\nconst byStatus: ColorSetting = { mode: 'field', field: 'status' }\nconst byRisk: ColorSetting = {\n mode: 'conditions',\n rules: [\n { color: '#f82b60', condition: { column: 'status', kind: 'select', op: 'is', value: 'Blocked' } },\n { color: '#fcb400', condition: { column: 'target', kind: 'date', op: 'isBefore', value: { mode: 'today' } } },\n ],\n}\n```\n\nBy field, each row takes its option's color, and the row header's title says\nthe option, so the color is never the only signal. A multiple select colors\nby its first value. By conditions, the first rule that matches wins: a\nblocked feature is red even when it is also late. A condition is tested as\nthe filter tests it: on the value the cell shows, so a formula, lookup,\nrollup or count field works, and a condition still missing its value colors\nnothing. A calendar colors its events by a single select only; a Gantt's bars\ntake either kind.\n\nThe color is a 3px stripe on the row header, unless the `rowColor` prop on\n`SvBase` sets a `target` (`'bar'` or `'row'`): a view's colors use it too. A\nview without `color` shows the `rowColor` prop's colors until someone picks\ncolors for it in Color. `{ mode: 'none' }` turns the colors off in that view.\n\n## Summaries in the footer and in group headers\n\nWith `summary` on, a click on a field's footer cell picks its summary: sum,\naverage, median, min, max or range for numbers; checked, unchecked or percent\nchecked for a checkbox; earliest, latest or the date range in days for dates;\nempty, filled, unique, percent empty or percent filled for any field. The\nresult is written as the field writes values. The choice is stored in the\nfield's `fieldProps.summary` and reported through `onColumnChange`; see\n[Summaries at the bottom](../fields.md#summaries-at-the-bottom).\n\nGroup headers show the same summaries over that group's records. Both leave\nout records the filter hides, even in a collapsed group.\n\n## The view bar over your own grid\n\n`SvBaseViewBar` is the bar alone, for a page that lays out its own\n`<SvGrid>`:\n\n```svelte\n<script lang=\"ts\">\n import { SvGrid } from '@svgrid/grid'\n import { SvBaseViewBar, enableBase, type FieldView } from '@svgrid/enterprise/base'\n\n enableBase()\n // Raw: the bar finds its grid by the api object itself, and a proxy is another object.\n let api = $state.raw<any>(null)\n let views = $state<FieldView[]>([\n { id: 'all', name: 'All features', type: 'grid' },\n { id: 'board', name: 'Status board', type: 'kanban', stackBy: 'status' },\n ])\n let activeView = $state('all')\n</script>\n\n{#if api}\n <SvBaseViewBar {api} bind:views bind:activeView\n rowHeights={[{ label: 'Compact', height: 32 }, { label: 'Roomy', height: 64 }]} />\n{/if}\n<SvGrid data={rows} {columns} primaryField=\"feature\" editable containerHeight={600}\n onApiReady={(a) => (api = a)} />\n```\n\nIts \"Create\" offers grid, Kanban and calendar (and Gantt once enabled). It\nloads the board, calendar and gallery renderers the first time a view of\nthat type is shown, and opens the record and the \"Delete record?\" question\nfor a card's menu. Chart and form views are drawn by `SvBase`, not by the\nbar, so leave them out of `viewTypes` here, or draw `SvBaseForm` yourself\nwhen a form view is active.\n\n## Pitfalls\n\n- **A new view from \"Create\" starts with `color: { mode: 'none' }`**, so it\n hides the `rowColor` prop's colors. A new calendar or Gantt is the\n exception: it starts colored by the first single select. Pick colors in\n Color, or declare the view in code without `color`.\n- **`color.field` is the data field; the other keys are column ids.** They\n differ only when a column sets an `id`.\n- **Summaries belong to the field, not the view.** Picking \"Average\" in one\n view changes every view's footer and group headers.\n- **Save views only after they load.** See the `loaded` check in step 12.\n\n## Next\n\n- [Linked records](./linked-records.md): link Features to Feedback and chart\n or filter by the linked names.\n- [Formulas](./formulas.md): computed fields a view can sort, group and chart\n by.\n- [Forms](./forms.md): required fields, the thank-you screen and forms\n outside the base.\n- [Collaboration](./collaboration.md): comments, people and who changed what.\n- [Automations](./automations.md): conditions written with this same filter.\n- [Keep a base in a database](./data.md): load the tables from Postgres,\n Supabase or your API and write every change back.\n"
5308
5308
  },
5309
5309
  {
5310
5310
  "slug": "help/benchmarks",
5311
5311
  "path": "docs/help/benchmarks.md",
5312
5312
  "title": "Performance benchmarks",
5313
5313
  "section": "Help",
5314
- "markdown": "# Performance benchmarks\n\nEvery number on this page is produced by a checked-in script:\n\n| Table | Command | Where it runs |\n| -------------------------- | --------------- | ------------- |\n| Bundle size | `pnpm size` | Node |\n| Sort / filter / group, memory | `pnpm bench` | Node |\n| First paint, scroll | `pnpm bench:dom` | Chromium |\n\nAn earlier version of this page published a table of timings that no\nscript produced, and told readers to reproduce them with a `pnpm bench`\ncommand that did not exist. Where a figure has since been re-measured and\ncame out worse than the old claim, the section says so rather than\nquietly restating it.\n\nThe one number worth trusting on any grid's benchmark page is the one you\nmeasured yourself, on your data shape, on your hardware. The\n[benchmark harness recipe](../recipes/benchmark-harness.md) is a\ncopy-paste `<SvGrid>` probe that measures time-to-first-paint across any\n(rows x columns) matrix.\n\nLive load - 100k rows x 100 columns with row + column virtualization:\n\n<div data-docs-demo=\"06-large-dataset\" data-height=\"500\"></div>\n\n## Test rig\n\nThe engine table below was measured on this configuration. Your absolute\nnumbers will differ; the ratios between rows are the transferable part.\n\n| Component | Spec |\n| --------- | ------------------------------------------------ |\n| Machine | Developer workstation, win32 x64. Not a dedicated bench rig |\n| Engine | Node v24.18.0, median of 9 runs after a warm-up |\n| Browser | Chromium via Playwright, best of 2 runs |\n| Dataset | 100,000 rows x 9 columns, seeded so runs compare |\n| Target | The built `@svgrid/grid/core`, not `src/` |\n\nMedian rather than mean, so one GC pause cannot move a number. The\n`min`/`max` columns the harness prints are worth reading: the spread on a\nsort run is around 10%, which is why **elapsed time is reported but never\ngated in CI**. What CI gates instead is described under\n[What CI enforces](#what-ci-enforces).\n\n## Bundle size\n\nProduction build, gzipped, Svelte excluded as a peer. The table is\ngenerated from `docs/_data/svgrid-size.json` by `node tools/sync-guide-facts.mjs`\n(the file `pnpm size:json` writes), and the method and the lazy-chunk\nbreakdown are on the [bundle size reference](./bundle-size.md). An earlier\nversion of this page typed these and sat a month behind the measurement.\n\n<!-- size:start -->\nRe-measured **10 Nov 2026** at `@svgrid/grid` 3.0.33 with the script that ships in the repo (`pnpm size:json`):\n\n| Target | Base JS (gzip) | CSS (gzip) | Loaded on demand |\n| --- | ---: | ---: | ---: |\n| Headless core (`createSvGrid`) | **3.2 KB** | - | - |\n| Headless subpath (`@svgrid/grid/core`) | **10.1 KB** | - | - |\n| Full render component (`<SvGrid>`) | **99.9 KB** | **11.3 KB** | **198.9 KB** |\n| Standalone chart (`<SvChart>`) | **71.9 KB** | - | **15.0 KB** |\n<!-- size:end -->\n\nThe \"loaded on demand\" column is the code reachable only through\n`import()`: charts, the date-time editor, the menus, export. Outside the\ngrid package, and not in the ledger: the `@svgrid/enterprise` core is a few\nkilobytes of export, print and import shells; `jszip` loads on the first\nxlsx export or import and `pdfmake` with its fonts on the first PDF export,\nboth as optional peers, neither in your synchronous bundle.\n\nThe AI helpers are no longer in this table: they moved into the free\n`@svgrid/grid` and tree-shake out unless you import them.\n\nTree-shaking is friendly: importing `{ SvGrid, tableFeatures }`\nwithout `rowSortingFeature` doesn't pull the sort module.\n\n## First paint\n\n**Measured.** Run `pnpm bench:dom`. Driven against demo\n`06-large-dataset` in Chromium: click a dataset size, stop the clock on\nthe first painted frame that shows the new rows.\n\n| Scenario | Total | Building the data | Grid |\n| ------------------- | -------- | ----------------- | ------- |\n| 1,000 x 29 cols | 180 ms | 3 ms | 177 ms |\n| 10,000 x 53 cols | 315 ms | 31 ms | 284 ms |\n| 50,000 x 77 cols | 652 ms | 220 ms | 432 ms |\n| 100,000 x 100 cols | 1,205 ms | 633 ms | 572 ms |\n\n\"Building the data\" is the demo generating synthetic rows - at the top\nsize that is 9.5 million cells of test data, which a real app with data\nalready in hand does not pay. \"Grid\" is the remainder.\n\n**\"Grid\" overstates the grid's own cost, and by a lot.** A CPU profile of\nthe 100k x 100 mount attributes 28% of the window to the demo's data\ngenerator and another 14% to garbage collection, most of it cleaning up\nafter that generator. The demo's self-reported build time counts only the\ngenerator's own execution, not the GC pressure it creates, so subtracting\nit leaves a good deal of the demo's cost inside the \"Grid\" column. Roughly\nhalf the total is the harness rather than the component.\n\nTwo further caveats:\n\n- These come from a developer machine, not a dedicated bench rig, and\n run-to-run variance is roughly 2x. Treat the column ratios as the\n signal and the absolute values as a ceiling.\n- **The row-pipeline optimisations above did not move these numbers**, and\n should not be read as having done so. Mounting a grid with no sort and no\n filter never enters those code paths; both row models return early.\n\nAn earlier version of this page claimed 110 ms for the 100k x 100 case,\nwhich was never measured. Getting a trustworthy figure needs a mount\nharness that feeds pre-built data, so the number is not dominated by the\ncost of inventing it.\n\n## Scroll performance\n\n**Measured.** Run `pnpm bench:dom`. Sustained vertical scroll at 60\npx/frame over 240 frames, sampling the interval between animation frames:\n\n| Scenario | p50 | p95 | max | Dropped frames |\n| ---------------------- | ------- | ------- | ------- | -------------- |\n| 100k rows x 100 cols | 16.7 ms | 17.7 ms | 18.8 ms | 0 of 240 |\n| 1,000,000 rows x 9 cols | 16.7 ms | 18.1 ms | 19.1 ms | 0 of 240 |\n\nThis is the part of the grid that holds up well: a million rows scrolls\nwithout dropping a frame, and the 100-column case is indistinguishable\nfrom it because row and column virtualization are the same machinery.\n\nRead the numbers correctly, though. The harness drives scrolling from\n`requestAnimationFrame`, so an interval can never come in below the\ndisplay refresh - 16.7 ms on a 60 Hz panel is the floor, not a\nmeasurement of how cheap a frame is. What the table shows is that the\ngrid **holds** the frame budget, with zero frames over 1.5x the median.\nIt cannot show headroom above 60 fps.\n\nAn earlier version of this page claimed a p95 of 8 ms and \"~120 fps\".\nNo harness in the repository could have produced that figure, and this\none structurally cannot.\n\n> Video: One million rows in SvGrid\n\nTranscript: This SvGrid holds one million rows, generated in the browser, with sorting, filtering and inline editing switched on. Rows are virtualized, so only the ones in view are drawn. Scroll, and the frame rate stays flat. Jump anywhere in the set. The row numbers show the scale, and the grid keeps every row reachable. Row and column virtualization are the same machinery, so wide grids get the same treatment. Nothing to configure.\n\n## Sort, filter, group\n\n**Measured.** Run `pnpm bench` to reproduce. In-memory row-pipeline\noperations on 100,000 rows x 9 columns:\n\n| Operation | Median | Was | Change |\n| --------------------------------------------- | ------- | ------- | ------ |\n| Sort by one column | 27 ms | 351 ms | 13x |\n| Sort by three columns (multi-sort) | 48 ms | 569 ms | 12x |\n| Sort by a date column | 40 ms | 773 ms | 19x |\n| Sort by a text column (collated) | 59 ms | - | new |\n| Filter, one operator | 5 ms | 19 ms | 4x |\n| Filter, five operators ANDed | 7 ms | 24 ms | 3.5x |\n| Filter via the compiled Excel filter | 3 ms | - | new |\n| ... the same, on fully accented text | 22 ms | - | new |\n| Group by two columns + three aggregators | 50 ms | 229 ms | 4.6x |\n| Toggle one row's checkbox (filtered + sorted) | 0 ms | 103 ms | gone |\n| Export CSV, 10k rows x 9 cols | 10 ms | 8 ms | - |\n\nThe last two rows are the path `<SvGrid>` actually takes. The plain \"filter,\none operator\" rows drive the headless row model, whose default match is a\nsubstring compare - but the menu and filter-row surfaces compile through\n`compileExcelFilter`, which folds every value (decompose, strip diacritics,\nlowercase) so that \"cafe\" matches \"CAFÉ\". This suite had no case touching\nthat fold until a browser profile showed it was the most expensive single\nfunction in a filter. It now takes a fast path for pure-ASCII values, where\ndecomposition and diacritic-stripping are provably no-ops: 3 ms rather than\n21 ms for the same 100,000 rows. Accented data still pays the full fold,\nwhich is the 21 ms row, and gets the same answer it always did.\n\nEach case runs in a **fresh process**. They used to share one, which made\nevery figure depend on the order cases ran in: a case running eighth\ncarried seven predecessors' heap, and grouping measured 169 ms in a full\nrun against 51 ms on its own. An earlier version of this table published\nthe contaminated 106 ms for grouping. Forcing a garbage collection between\ncases did not fix it - the cost is a warmed, fragmented heap rather than\nuncollected garbage - so the harness now pays a few seconds of process\nstartup per case and reports numbers that mean the same thing whether you\nrun one case or all of them.\n\nText sorting has no \"was\", because this suite did not measure it until a\nbrowser comparison showed it was the slowest operation for every grid\ntested. Collation is what makes it expensive, and two things help:\n\n- **Columns whose distinct values are far fewer than their rows** - a\n status, a region, an owner - collate the distinct values once and sort\n rows by rank. That is what took the three-column case to 48 ms. A stride\n sample decides whether to try, because collecting distinct values costs\n about 9 ms on a column where nearly all of them are unique and saves\n about 19 ms where they repeat, so guessing wrong either way is\n measurable.\n- **`localeCompare` rather than a hoisted `Intl.Collator`.** The\n specification defines the former as constructing a collator per call, so\n caching one looks like the obvious win - and it is measurably slower,\n because V8 fast-paths `localeCompare` and going through a collator object\n misses that path. 33 ms against 83 ms sorting 100k ASCII strings, with\n byte-identical ordering. This page briefly published the slower version;\n re-measure before \"optimising\" it again.\n\nA third thing helps every sort: **the single-clause case gets its own\ncomparator.** Most sorts are one column, and the comparator runs O(n log n)\ntimes - 1.66 million calls for 100,000 rows - so the general loop's clause\nlookup, property loads and indirect call are paid 1.66 million times for\nvalues that never change. Hoisting them into a closure, and inlining the\nsubtraction for numeric columns, is worth about 15% on a single-column sort.\n\nGrouping got faster for a different reason: a group row shows the value\nits children agree on, and finding that used to mean scanning every child\neven for a column an outer level had already grouped by, where the answer\nis fixed by construction.\n\nMedians hold steady across runs; the min/max the harness also prints do\nnot, because this is a working developer machine rather than a quiet bench\nrig. A single sample can land 3x the median when something else on the box\nwakes up. That noise is the whole reason CI gates counters instead.\n\nThe \"Was\" column is the first measurement this harness took, before any of\nit was optimised. Building the harness is what exposed the defects; each\none is now closed and has a counter that fails the build if it returns:\n\n- **Sorting resolved the clause's column inside the comparator.** A\n single-clause sort of 100k rows called `getAllColumns()` 1,528,947\n times, scanning the column array on each. It now resolves once, and the\n comparator reads a precomputed key array instead of walking the row.\n- **The date comparator allocated two `Date` objects per comparison.**\n Timestamps are now computed once per row, which is most of why sorting a\n date column went from the slowest case to a fast one.\n- **Filtering materialised every row's full `Cell[]`** to read one field,\n defeating the lazy per-row cell cache: 100,000 cell arrays built and\n thrown away for a one-operator filter. Now zero.\n- **Toggling a checkbox re-ran the whole pipeline.** `rowSelection` sat in\n the row-model cache key, so selecting a row re-filtered and re-sorted the\n entire dataset even though nothing in the pipeline reads it. Selection is\n now free.\n- **Group aggregation allocated three arrays per aggregated column per\n group**, then finished with `Math.min(...values)`. It is a single pass\n now. The spread also crashed outright on a large enough group, which is\n fixed as a side effect.\n- **Text sorting collated on every comparison.** Distinct values are now\n ranked once where a column has few enough of them.\n\nNumbers here will move; the command that produces them will not.\n\n## Streaming updates\n\n**Measured.** Run `pnpm bench --case=tick-1k-of-100k-sorted` to reproduce.\nA tick is a new data array in which K row objects were replaced with new\nprices, handed to a grid sorted by that column, so the order has to stay\ncurrent. 100,000 rows x 9 columns, same rig as above:\n\n| Operation | Median | The same tick through a full sort |\n| ---------------------------------------------------- | ------ | --------------------------------- |\n| Tick: 100 rows replaced, sorted by the ticking column | 6 ms | 31 ms |\n| Tick: 1,000 rows replaced, sorted by the ticking column | 9 ms | 31 ms |\n\nThe right-hand column is the \"Sort by one column\" row above: before this\nround, any new data array re-sorted every row, so a 100-row tick cost\nexactly what a header click did. The sorted stage now keeps its previous\noutput with each row's sort key, drops the rows that were replaced, sorts\nthe K replacements among themselves and merges them back in one linear\npass: O(n + K log K) instead of O(n log n) key builds and comparisons. The\nfiltered stage does the same for a replacement whose membership did not\nchange. The result is byte-identical to a full sort, which a test pins by\ncomparing the two on every kind of key the sort specialises\n(`core.tick-repair.test.ts`).\n\nThree conditions, all checked, none assumed: the array has the same length\n(an add or a remove is structural and takes the full path), at least one\nobject was replaced and no more than a quarter of them (a new array with\nevery object the same is the documented \"I mutated rows in place, re-read\nthem\" refresh and runs the full pipeline), and the sorting state, filters\nand columns are the same objects as last time. The rows that were kept are\nassumed unchanged in value as well as identity, which is what an immutable\nupdate guarantees and an in-place mutation does not: a feed that mutates\nrows in place must pass a new array with no replacements to refresh.\n\nThe counter CI gates for this is cell reads on the kept rows during a tick,\nbudgeted at zero: a repaired sort never re-reads a row it kept, a full sort\nreads all of them.\n\nWhat the component adds on top of the engine per tick is a rebuild of the\nbase rows (2 ms at 100k, reusing every row object that did not change) and\nthe render of the visible window. Three things on that path used to be\nO(rows) per data change and are not now: the header checkbox state walked\nevery row to conclude \"none selected\" (30 ms at 100k), the expanded-rows\nstage copied all rows into a new array with nothing expanded, and the\ndev-time config check filtered every row to pick ten samples. The browser\nharness in the [comparison](./comparison.md) measures the whole thing,\nrender included, against the other grids on the same tick; that column\nlands with its next recorded run.\n\n## Spreadsheet engine\n\n**Measured.** Run `pnpm bench` with `@svgrid/enterprise` built\n(`pnpm --filter @svgrid/enterprise build:lib`) to reproduce; the cases live\nin `tools/bench/sheet-cases.mjs`. These were taken on a CI-class Linux\ncontainer with Node v22, not the workstation the table above used, so\ncompare the rows with each other rather than with the grid's.\n\nTwo different questions, because a spreadsheet is asked both: what does\nopening a workbook full of formulas cost once, and what does one keystroke\ncost all day.\n\n| Case | Median | Cells evaluated | Retained |\n| -------------------------------------------------------- | ------- | --------------- | -------- |\n| Open 1,000 rows, one `SUM` per row | 8 ms | 1,000 | 1.6 MB |\n| Open 10,000 rows, one `SUM` per row | 91 ms | 10,000 | 16.5 MB |\n| Open 50,000 rows, one `SUM` per row | 547 ms | 50,000 | 84.2 MB |\n| Type in one cell of a 10,000-row sheet | 0 ms | 1 | - |\n| Type in a column a single `SUM` of 10,000 cells reads | 14 ms | 1 | - |\n| Type at the top of a 10,000-row dependency chain | 71 ms | 10,000 | - |\n\nRead the third column first: it is the one CI gates, and the one that says\nwhat the engine is doing. A keystroke costs **one evaluation**, not a\nsheet's worth, because the dependency graph recomputes what read the cell\nand nothing else. The column-`SUM` row is the interesting pair: one\nevaluation, 14 ms, because that single `SUM` re-reads its 10,000 cells. The\nchain row is the worst case a sheet can have, a running balance where every\nrow reads the row above it: 10,000 evaluations for one keystroke, which is\nthe true cost of that shape and not a bug.\n\nSo the practical ceiling, on this container: **a sheet of tens of thousands\nof formula rows opens in well under a second and edits instantly**, and the\nthing that changes that is not the sheet's size but how much of it one cell\nfeeds. 50,000 formula rows retain about 84 MB, roughly 1.7 KB per row for\nthe text, the cached value and the graph edges.\n\nA chain longer than about a thousand rows used to overflow the JavaScript\nstack on the first keystroke and land `#NUM!` in the cell. The workbook now\nprimes a deep chain from its far end instead of recursing through it, so\nthe only limit left is the sheet's own size.\n\n## What CI enforces\n\nElapsed time is too noisy on a shared runner to fail a build on, and a\nflaky performance gate gets switched off within a month. So `pnpm\nbench:check` gates **work counters** instead - counts of algorithmic work,\nidentical on every machine:\n\n| Counter | Budget | Actual |\n| --------------------------------------------- | ------ | ------ |\n| `getAllColumns` calls inside a 1-clause sort | 8 | 1 |\n| `getAllColumns` calls inside a 3-clause sort | 24 | 1 |\n| `getAllCells` materialisations while filtering | 0 | 0 |\n| Filter re-runs caused by a selection change | 0 | 0 |\n| Sort re-runs caused by a selection change | 0 | 0 |\n| Cells evaluated opening a 10,000-row sheet | 10,004 | 10,000 |\n| Cells evaluated by one keystroke in that sheet | 4 | 1 |\n| Cells evaluated when a 10,000-cell `SUM` reads the edit | 3 | 1 |\n\nA count is a claim about the algorithm; a millisecond is a claim about the\nhardware. Only the first belongs in CI. Budgets live beside each case in\n`tools/bench/cases.mjs`, and the spreadsheet's in\n`tools/bench/sheet-cases.mjs`.\n\nEach case also declares which counters must be *present*, not just within\nbudget. Without that, a stage that stopped running altogether would satisfy\nevery budget by doing no work, which is the failure mode a gate like this\nis most likely to develop.\n\n## Memory\n\n**Measured** (`pnpm bench`, Node heap): building the row model for\n100,000 rows x 9 columns with no sort, filter or grouping retains\n**0.8 MB**, with no renderer attached. A row object is made the first time\nsomething reads it, so a grid showing twenty rows makes a screenful of\nthem, not 100,000.\n\nThe first sort, filter or grouping reads every row and makes them all. The\nbrowser harness measures that as the \"after sorting and filtering\" row in\nthe [comparison](./comparison.md#measured-performance): 17.8 MB at\n100,000 rows, and 128 MB at 1,000,000 (27 Oct 2026,\n`BENCH_ROWS=1000000 BENCH_GRIDS=svgrid pnpm bench:compare`), so about 130\nbytes a row: the row object, its id, and the sort key it keeps so a live\nupdate can be merged in without re-reading the rows. The 1,000,000-row\nfigure was 196 MB the day before. Reading one column used to build an\narray of every column's value on every row and keep it there, which only\nthe rows on screen use; a plain `field` column is now read from the data,\nand only a computed (`fieldFn`) column still builds the array.\n\nThe fully built row model was 16.0 MB at 100,000 rows before rows became\nlazy, and 64.9 MB before that. Each row was an object literal whose seven\nmethods were closures, so a 100k-row grid allocated 700k closures and a\nclosure scope per row before painting anything. The methods are now\ndefined once and shared, reading their row through `this`: 4x less memory\nand about 7x faster to build. They are still assigned as own properties\nrather than put on a prototype, because `Row` is public and several row\nmodels legitimately do `{ ...row, depth }` - a spread copies own\nproperties but not a prototype, so a class would silently strip the\nmethods off a cloned row.\n\nAn earlier version of this page said \"~22 MB\". That figure was not\nmeasured. Node's heap accounting is not the same as a browser's, so read\nthis as the engine's own cost rather than what a tab will show.\n\nTwo claims here are structural rather than measured, and both hold:\n\n- Virtualization keeps the rendered DOM to the visible window regardless\n of dataset size, so `<td>` count does not scale with row count.\n- No retained references when the grid unmounts - the cleanup path is\n exercised by the unmount test in `svgrid.behavior.test.ts`.\n\n## Server-side / chunked loading\n\nDemo [33. Server-side infinite scroll](https://svgrid.com/demos/33-server-infinite/) covers the chunked-load path. Numbers from that demo:\n\n| Scenario | Result |\n| ------------------------------------------------ | --------------------------------------- |\n| Initial paint, sparse 100k-row dataset | 110 ms to first chunk visible |\n| Scroll 50,000 rows in 1.5 s (fast wheel-flick) | 16 chunk requests cancelled mid-flight |\n| Sort 100k server-side rows | round-trip dominated by the mock latency (50-140 ms) |\n\n### The Server-Side Row Model over one million rows\n\nDemo [467](https://svgrid.com/demos/467-server-row-model-1m/) runs the\nEnterprise row model over a columnar in-memory warehouse of a million rows,\ngrouped Region > Country. `tests/perf/server-row-model.spec.ts` drives it in\nheadless Chromium (a developer workstation, 17 September 2026; a release\nritual, not a gate):\n\n| Scenario | Result |\n| ------------------------------------------------------------- | --------------------------------------------- |\n| First paint, navigation to the first group row (includes building the warehouse in the page) | 2.4 s |\n| Scroll 60,000 px through an open country of 62,000 leaves | 17 leaf requests, 0.28 per 1,000 px (one block of 100 rows is 3,400 px) |\n| Worst frame while blocks streamed in | 83 ms (layout of the changed rows; the grid keeps the row objects a block did not touch) |\n| Heap before / after streaming 700,000 px with `maxBlocksInCache: 24` | 225 MB / 225 MB - evicted blocks are collected |\n| Warehouse: a million rows as typed arrays | 22 MB; a cold sort on a new column 0.8 to 1.1 s, cached after; a grouped level 140 ms; a pivot with a grand total 330 ms |\n\n## AI helpers\n\nEnd-to-end timings against the bundled `mockAIProvider`:\n\n| Helper | Median time (ms) |\n| --------------- | ---------------- |\n| `aiFilter` | 350-750 (mock latency dominated) |\n| `aiSmartFill` (50 rows) | 400-900 |\n| `aiSummarize` | 350-750 |\n| `aiClassify` (20 rows) | 400-750 |\n\nAgainst a real model the latency is provider-side. The grid's own\nprompt-build + result-parse work stays under ~6 ms even for 1000-row\nclassify jobs.\n\n## Import / export\n\n| Operation | Time |\n| ------------------------------------ | ------ |\n| Parse CSV, 10k rows × 9 cols | 28 ms |\n| Parse xlsx, 10k rows × 9 cols | 140 ms (jszip unzip-dominated) |\n| Export CSV, 10k rows × 9 cols | 18 ms |\n| Export xlsx, 10k rows × 9 cols | 220 ms |\n| Export PDF, 1k rows × 9 cols (pdfmake) | 700 ms |\n\n## Reproducing locally\n\n```bash\ngit clone https://github.com/sv-grid/sv-grid\ncd sv-grid\npnpm install\npnpm --filter @svgrid/grid build # the bench runs the BUILT engine\npnpm bench # engine + memory tables\npnpm size # bundle table\npnpm bench:dom # first paint + scroll, in Chromium\n```\n\nThe bench takes a few options:\n\n```bash\npnpm bench --json # machine-readable, for trend tracking\npnpm bench --baseline=tools/bench/baseline.json # adds a delta column\npnpm bench --case=sort-1col # one case\npnpm bench:check # the work-counter gate CI runs\n```\n\n`pnpm bench` runs against the built `@svgrid/grid/core` rather than\n`src/`, so it also smoke-tests the entry a consumer actually imports.\n\n`pnpm bench:dom` drives Chromium through Playwright against the examples\ngallery, which lives in this repository, so it needs no private\nsubmodule. It is a Playwright project rather than part of `pnpm test:e2e`,\nand it is deliberately not in CI - see below.\n\nFor your own workload, the [benchmark harness\nrecipe](../recipes/benchmark-harness.md) and the [FPS HUD\nrecipe](../recipes/profiling-with-fps-hud.md) are copy-paste probes you\ncan drop into your app.\n\n## What we *don't* claim\n\n- That we are faster than any other grid. Nothing on *this* page is a\n head-to-head measurement. Everything here is sv-grid measured against\n sv-grid, which supports \"this got 10x faster\" and not \"this is faster\n than X\". For a real comparison there is a separate grid-agnostic\n harness - `pnpm bench:compare`, results and caveats in\n [Comparison](./comparison.md) - and it does not say we win everything.\n- That first paint is good. The grid half of the 100k x 100 case is still\n the worst number on this page, and this round did not touch it.\n- Any scroll figure above 60 fps. The harness is rAF-driven and cannot\n observe headroom past the display refresh. \"Zero dropped frames\" is the\n claim; \"120 fps\" is not.\n- Lab-grade first-paint numbers. Those come from a developer machine with\n roughly 2x run-to-run variance, not a dedicated rig.\n- \"Smoothest grid on the market\" - that depends entirely on what your\n cells render. A sparkline + currency formatter in every cell costs\n more than a number, and we don't pretend otherwise.\n- \"Zero allocations during scroll\" - the virtualizer recycles DOM\n nodes but cell snippets still allocate. The numbers above include\n real-world snippets (status pills, mini-bars).\n- Single-thread performance > 1M rows. For >1M, do the heavy lifting\n on the server and feed chunks through the [server-side infinite\n scroll pattern](https://svgrid.com/demos/33-server-infinite/).\n\n## Frequently asked questions\n\n### How fast is SvGrid?\n\nIt virtualizes both rows and columns, so only the visible window is in the DOM,\nand a million rows scrolls without dropping a frame. On the data side, a\nsingle-column sort of 100,000 rows takes about 27 ms and a one-operator filter\nabout 5 ms. Every table on this page comes from a checked-in script you can run\nyourself, and the numbers moved because the harness exposed real defects rather\nthan because the prose was rewritten.\n\n### How many rows can SvGrid handle?\n\nClient-side, 100k+ rows scroll smoothly thanks to virtualization. For millions\nof rows, page or chunk from the server (see Server-side data). The DOM only ever\nholds the visible window regardless of total row count.\n\n### How fast is SvGrid, and how big is it?\n\nIt ships a small bundle (the measured figures for the full render component\nand for the headless core alone are in the table above) and virtualizes by\ndefault. Raw scroll\nperformance is comparable for typical workloads; the bigger practical win is\nbundle size and a Svelte-native runtime with no framework bridge.\n\n## See also\n\n- [Browser support](./browser-support.md) - the matrix the benchmarks\n ran against.\n- [Testing and quality](./testing-and-quality.md) - the coverage\n thresholds that gate every release.\n"
5314
+ "markdown": "# Performance benchmarks\n\nEvery number on this page is produced by a checked-in script:\n\n| Table | Command | Where it runs |\n| -------------------------- | --------------- | ------------- |\n| Bundle size | `pnpm size` | Node |\n| Sort / filter / group, memory | `pnpm bench` | Node |\n| First paint, scroll | `pnpm bench:dom` | Chromium |\n\nAn earlier version of this page published a table of timings that no\nscript produced, and told readers to reproduce them with a `pnpm bench`\ncommand that did not exist. Where a figure has since been re-measured and\ncame out worse than the old claim, the section says so rather than\nquietly restating it.\n\nThe one number worth trusting on any grid's benchmark page is the one you\nmeasured yourself, on your data shape, on your hardware. The\n[benchmark harness recipe](../recipes/benchmark-harness.md) is a\ncopy-paste `<SvGrid>` probe that measures time-to-first-paint across any\n(rows x columns) matrix.\n\nLive load - 100k rows x 100 columns with row + column virtualization:\n\n<div data-docs-demo=\"06-large-dataset\" data-height=\"500\"></div>\n\n## Test rig\n\nThe engine table below was measured on this configuration. Your absolute\nnumbers will differ; the ratios between rows are the transferable part.\n\n| Component | Spec |\n| --------- | ------------------------------------------------ |\n| Machine | Developer workstation, win32 x64. Not a dedicated bench rig |\n| Engine | Node v24.18.0, median of 9 runs after a warm-up |\n| Browser | Chromium via Playwright, best of 2 runs |\n| Dataset | 100,000 rows x 9 columns, seeded so runs compare |\n| Target | The built `@svgrid/grid/core`, not `src/` |\n\nMedian rather than mean, so one GC pause cannot move a number. The\n`min`/`max` columns the harness prints are worth reading: the spread on a\nsort run is around 10%, which is why **elapsed time is reported but never\ngated in CI**. What CI gates instead is described under\n[What CI enforces](#what-ci-enforces).\n\n## Bundle size\n\nProduction build, gzipped, Svelte excluded as a peer. The table is\ngenerated from `docs/_data/svgrid-size.json` by `node tools/sync-guide-facts.mjs`\n(the file `pnpm size:json` writes), and the method and the lazy-chunk\nbreakdown are on the [bundle size reference](./bundle-size.md). An earlier\nversion of this page typed these and sat a month behind the measurement.\n\n<!-- size:start -->\nRe-measured **10 Nov 2026** at `@svgrid/grid` 3.0.34 with the script that ships in the repo (`pnpm size:json`):\n\n| Target | Base JS (gzip) | CSS (gzip) | Loaded on demand |\n| --- | ---: | ---: | ---: |\n| Headless core (`createSvGrid`) | **3.2 KB** | - | - |\n| Headless subpath (`@svgrid/grid/core`) | **10.2 KB** | - | - |\n| Full render component (`<SvGrid>`) | **99.6 KB** | **11.4 KB** | **206.3 KB** |\n| Standalone chart (`<SvChart>`) | **71.9 KB** | - | **16.0 KB** |\n<!-- size:end -->\n\nThe \"loaded on demand\" column is the code reachable only through\n`import()`: charts, the date-time editor, the menus, export. Outside the\ngrid package, and not in the ledger: the `@svgrid/enterprise` core is a few\nkilobytes of export, print and import shells; `jszip` loads on the first\nxlsx export or import and `pdfmake` with its fonts on the first PDF export,\nboth as optional peers, neither in your synchronous bundle.\n\nThe AI helpers are no longer in this table: they moved into the free\n`@svgrid/grid` and tree-shake out unless you import them.\n\nTree-shaking is friendly: importing `{ SvGrid, tableFeatures }`\nwithout `rowSortingFeature` doesn't pull the sort module.\n\n## First paint\n\n**Measured.** Run `pnpm bench:dom`. Driven against demo\n`06-large-dataset` in Chromium: click a dataset size, stop the clock on\nthe first painted frame that shows the new rows.\n\n| Scenario | Total | Building the data | Grid |\n| ------------------- | -------- | ----------------- | ------- |\n| 1,000 x 29 cols | 180 ms | 3 ms | 177 ms |\n| 10,000 x 53 cols | 315 ms | 31 ms | 284 ms |\n| 50,000 x 77 cols | 652 ms | 220 ms | 432 ms |\n| 100,000 x 100 cols | 1,205 ms | 633 ms | 572 ms |\n\n\"Building the data\" is the demo generating synthetic rows - at the top\nsize that is 9.5 million cells of test data, which a real app with data\nalready in hand does not pay. \"Grid\" is the remainder.\n\n**\"Grid\" overstates the grid's own cost, and by a lot.** A CPU profile of\nthe 100k x 100 mount attributes 28% of the window to the demo's data\ngenerator and another 14% to garbage collection, most of it cleaning up\nafter that generator. The demo's self-reported build time counts only the\ngenerator's own execution, not the GC pressure it creates, so subtracting\nit leaves a good deal of the demo's cost inside the \"Grid\" column. Roughly\nhalf the total is the harness rather than the component.\n\nTwo further caveats:\n\n- These come from a developer machine, not a dedicated bench rig, and\n run-to-run variance is roughly 2x. Treat the column ratios as the\n signal and the absolute values as a ceiling.\n- **The row-pipeline optimisations above did not move these numbers**, and\n should not be read as having done so. Mounting a grid with no sort and no\n filter never enters those code paths; both row models return early.\n\nAn earlier version of this page claimed 110 ms for the 100k x 100 case,\nwhich was never measured. Getting a trustworthy figure needs a mount\nharness that feeds pre-built data, so the number is not dominated by the\ncost of inventing it.\n\n## Scroll performance\n\n**Measured.** Run `pnpm bench:dom`. Sustained vertical scroll at 60\npx/frame over 240 frames, sampling the interval between animation frames:\n\n| Scenario | p50 | p95 | max | Dropped frames |\n| ---------------------- | ------- | ------- | ------- | -------------- |\n| 100k rows x 100 cols | 16.7 ms | 17.7 ms | 18.8 ms | 0 of 240 |\n| 1,000,000 rows x 9 cols | 16.7 ms | 18.1 ms | 19.1 ms | 0 of 240 |\n\nThis is the part of the grid that holds up well: a million rows scrolls\nwithout dropping a frame, and the 100-column case is indistinguishable\nfrom it because row and column virtualization are the same machinery.\n\nRead the numbers correctly, though. The harness drives scrolling from\n`requestAnimationFrame`, so an interval can never come in below the\ndisplay refresh - 16.7 ms on a 60 Hz panel is the floor, not a\nmeasurement of how cheap a frame is. What the table shows is that the\ngrid **holds** the frame budget, with zero frames over 1.5x the median.\nIt cannot show headroom above 60 fps.\n\nAn earlier version of this page claimed a p95 of 8 ms and \"~120 fps\".\nNo harness in the repository could have produced that figure, and this\none structurally cannot.\n\n> Video: One million rows in SvGrid\n\nTranscript: This SvGrid holds one million rows, generated in the browser, with sorting, filtering and inline editing switched on. Rows are virtualized, so only the ones in view are drawn. Scroll, and the frame rate stays flat. Jump anywhere in the set. The row numbers show the scale, and the grid keeps every row reachable. Row and column virtualization are the same machinery, so wide grids get the same treatment. Nothing to configure.\n\n## Sort, filter, group\n\n**Measured.** Run `pnpm bench` to reproduce. In-memory row-pipeline\noperations on 100,000 rows x 9 columns:\n\n| Operation | Median | Was | Change |\n| --------------------------------------------- | ------- | ------- | ------ |\n| Sort by one column | 27 ms | 351 ms | 13x |\n| Sort by three columns (multi-sort) | 48 ms | 569 ms | 12x |\n| Sort by a date column | 40 ms | 773 ms | 19x |\n| Sort by a text column (collated) | 59 ms | - | new |\n| Filter, one operator | 5 ms | 19 ms | 4x |\n| Filter, five operators ANDed | 7 ms | 24 ms | 3.5x |\n| Filter via the compiled Excel filter | 3 ms | - | new |\n| ... the same, on fully accented text | 22 ms | - | new |\n| Group by two columns + three aggregators | 50 ms | 229 ms | 4.6x |\n| Toggle one row's checkbox (filtered + sorted) | 0 ms | 103 ms | gone |\n| Export CSV, 10k rows x 9 cols | 10 ms | 8 ms | - |\n\nThe last two rows are the path `<SvGrid>` actually takes. The plain \"filter,\none operator\" rows drive the headless row model, whose default match is a\nsubstring compare - but the menu and filter-row surfaces compile through\n`compileExcelFilter`, which folds every value (decompose, strip diacritics,\nlowercase) so that \"cafe\" matches \"CAFÉ\". This suite had no case touching\nthat fold until a browser profile showed it was the most expensive single\nfunction in a filter. It now takes a fast path for pure-ASCII values, where\ndecomposition and diacritic-stripping are provably no-ops: 3 ms rather than\n21 ms for the same 100,000 rows. Accented data still pays the full fold,\nwhich is the 21 ms row, and gets the same answer it always did.\n\nEach case runs in a **fresh process**. They used to share one, which made\nevery figure depend on the order cases ran in: a case running eighth\ncarried seven predecessors' heap, and grouping measured 169 ms in a full\nrun against 51 ms on its own. An earlier version of this table published\nthe contaminated 106 ms for grouping. Forcing a garbage collection between\ncases did not fix it - the cost is a warmed, fragmented heap rather than\nuncollected garbage - so the harness now pays a few seconds of process\nstartup per case and reports numbers that mean the same thing whether you\nrun one case or all of them.\n\nText sorting has no \"was\", because this suite did not measure it until a\nbrowser comparison showed it was the slowest operation for every grid\ntested. Collation is what makes it expensive, and two things help:\n\n- **Columns whose distinct values are far fewer than their rows** - a\n status, a region, an owner - collate the distinct values once and sort\n rows by rank. That is what took the three-column case to 48 ms. A stride\n sample decides whether to try, because collecting distinct values costs\n about 9 ms on a column where nearly all of them are unique and saves\n about 19 ms where they repeat, so guessing wrong either way is\n measurable.\n- **`localeCompare` rather than a hoisted `Intl.Collator`.** The\n specification defines the former as constructing a collator per call, so\n caching one looks like the obvious win - and it is measurably slower,\n because V8 fast-paths `localeCompare` and going through a collator object\n misses that path. 33 ms against 83 ms sorting 100k ASCII strings, with\n byte-identical ordering. This page briefly published the slower version;\n re-measure before \"optimising\" it again.\n\nA third thing helps every sort: **the single-clause case gets its own\ncomparator.** Most sorts are one column, and the comparator runs O(n log n)\ntimes - 1.66 million calls for 100,000 rows - so the general loop's clause\nlookup, property loads and indirect call are paid 1.66 million times for\nvalues that never change. Hoisting them into a closure, and inlining the\nsubtraction for numeric columns, is worth about 15% on a single-column sort.\n\nGrouping got faster for a different reason: a group row shows the value\nits children agree on, and finding that used to mean scanning every child\neven for a column an outer level had already grouped by, where the answer\nis fixed by construction.\n\nMedians hold steady across runs; the min/max the harness also prints do\nnot, because this is a working developer machine rather than a quiet bench\nrig. A single sample can land 3x the median when something else on the box\nwakes up. That noise is the whole reason CI gates counters instead.\n\nThe \"Was\" column is the first measurement this harness took, before any of\nit was optimised. Building the harness is what exposed the defects; each\none is now closed and has a counter that fails the build if it returns:\n\n- **Sorting resolved the clause's column inside the comparator.** A\n single-clause sort of 100k rows called `getAllColumns()` 1,528,947\n times, scanning the column array on each. It now resolves once, and the\n comparator reads a precomputed key array instead of walking the row.\n- **The date comparator allocated two `Date` objects per comparison.**\n Timestamps are now computed once per row, which is most of why sorting a\n date column went from the slowest case to a fast one.\n- **Filtering materialised every row's full `Cell[]`** to read one field,\n defeating the lazy per-row cell cache: 100,000 cell arrays built and\n thrown away for a one-operator filter. Now zero.\n- **Toggling a checkbox re-ran the whole pipeline.** `rowSelection` sat in\n the row-model cache key, so selecting a row re-filtered and re-sorted the\n entire dataset even though nothing in the pipeline reads it. Selection is\n now free.\n- **Group aggregation allocated three arrays per aggregated column per\n group**, then finished with `Math.min(...values)`. It is a single pass\n now. The spread also crashed outright on a large enough group, which is\n fixed as a side effect.\n- **Text sorting collated on every comparison.** Distinct values are now\n ranked once where a column has few enough of them.\n\nNumbers here will move; the command that produces them will not.\n\n## Streaming updates\n\n**Measured.** Run `pnpm bench --case=tick-1k-of-100k-sorted` to reproduce.\nA tick is a new data array in which K row objects were replaced with new\nprices, handed to a grid sorted by that column, so the order has to stay\ncurrent. 100,000 rows x 9 columns, same rig as above:\n\n| Operation | Median | The same tick through a full sort |\n| ---------------------------------------------------- | ------ | --------------------------------- |\n| Tick: 100 rows replaced, sorted by the ticking column | 6 ms | 31 ms |\n| Tick: 1,000 rows replaced, sorted by the ticking column | 9 ms | 31 ms |\n\nThe right-hand column is the \"Sort by one column\" row above: before this\nround, any new data array re-sorted every row, so a 100-row tick cost\nexactly what a header click did. The sorted stage now keeps its previous\noutput with each row's sort key, drops the rows that were replaced, sorts\nthe K replacements among themselves and merges them back in one linear\npass: O(n + K log K) instead of O(n log n) key builds and comparisons. The\nfiltered stage does the same for a replacement whose membership did not\nchange. The result is byte-identical to a full sort, which a test pins by\ncomparing the two on every kind of key the sort specialises\n(`core.tick-repair.test.ts`).\n\nThree conditions, all checked, none assumed: the array has the same length\n(an add or a remove is structural and takes the full path), at least one\nobject was replaced and no more than a quarter of them (a new array with\nevery object the same is the documented \"I mutated rows in place, re-read\nthem\" refresh and runs the full pipeline), and the sorting state, filters\nand columns are the same objects as last time. The rows that were kept are\nassumed unchanged in value as well as identity, which is what an immutable\nupdate guarantees and an in-place mutation does not: a feed that mutates\nrows in place must pass a new array with no replacements to refresh.\n\nThe counter CI gates for this is cell reads on the kept rows during a tick,\nbudgeted at zero: a repaired sort never re-reads a row it kept, a full sort\nreads all of them.\n\nWhat the component adds on top of the engine per tick is a rebuild of the\nbase rows (2 ms at 100k, reusing every row object that did not change) and\nthe render of the visible window. Three things on that path used to be\nO(rows) per data change and are not now: the header checkbox state walked\nevery row to conclude \"none selected\" (30 ms at 100k), the expanded-rows\nstage copied all rows into a new array with nothing expanded, and the\ndev-time config check filtered every row to pick ten samples. The browser\nharness in the [comparison](./comparison.md) measures the whole thing,\nrender included, against the other grids on the same tick; that column\nlands with its next recorded run.\n\n## Spreadsheet engine\n\n**Measured.** Run `pnpm bench` with `@svgrid/enterprise` built\n(`pnpm --filter @svgrid/enterprise build:lib`) to reproduce; the cases live\nin `tools/bench/sheet-cases.mjs`. These were taken on a CI-class Linux\ncontainer with Node v22, not the workstation the table above used, so\ncompare the rows with each other rather than with the grid's.\n\nTwo different questions, because a spreadsheet is asked both: what does\nopening a workbook full of formulas cost once, and what does one keystroke\ncost all day.\n\n| Case | Median | Cells evaluated | Retained |\n| -------------------------------------------------------- | ------- | --------------- | -------- |\n| Open 1,000 rows, one `SUM` per row | 8 ms | 1,000 | 1.6 MB |\n| Open 10,000 rows, one `SUM` per row | 91 ms | 10,000 | 16.5 MB |\n| Open 50,000 rows, one `SUM` per row | 547 ms | 50,000 | 84.2 MB |\n| Type in one cell of a 10,000-row sheet | 0 ms | 1 | - |\n| Type in a column a single `SUM` of 10,000 cells reads | 14 ms | 1 | - |\n| Type at the top of a 10,000-row dependency chain | 71 ms | 10,000 | - |\n\nRead the third column first: it is the one CI gates, and the one that says\nwhat the engine is doing. A keystroke costs **one evaluation**, not a\nsheet's worth, because the dependency graph recomputes what read the cell\nand nothing else. The column-`SUM` row is the interesting pair: one\nevaluation, 14 ms, because that single `SUM` re-reads its 10,000 cells. The\nchain row is the worst case a sheet can have, a running balance where every\nrow reads the row above it: 10,000 evaluations for one keystroke, which is\nthe true cost of that shape and not a bug.\n\nSo the practical ceiling, on this container: **a sheet of tens of thousands\nof formula rows opens in well under a second and edits instantly**, and the\nthing that changes that is not the sheet's size but how much of it one cell\nfeeds. 50,000 formula rows retain about 84 MB, roughly 1.7 KB per row for\nthe text, the cached value and the graph edges.\n\nA chain longer than about a thousand rows used to overflow the JavaScript\nstack on the first keystroke and land `#NUM!` in the cell. The workbook now\nprimes a deep chain from its far end instead of recursing through it, so\nthe only limit left is the sheet's own size.\n\n## What CI enforces\n\nElapsed time is too noisy on a shared runner to fail a build on, and a\nflaky performance gate gets switched off within a month. So `pnpm\nbench:check` gates **work counters** instead - counts of algorithmic work,\nidentical on every machine:\n\n| Counter | Budget | Actual |\n| --------------------------------------------- | ------ | ------ |\n| `getAllColumns` calls inside a 1-clause sort | 8 | 1 |\n| `getAllColumns` calls inside a 3-clause sort | 24 | 1 |\n| `getAllCells` materialisations while filtering | 0 | 0 |\n| Filter re-runs caused by a selection change | 0 | 0 |\n| Sort re-runs caused by a selection change | 0 | 0 |\n| Cells evaluated opening a 10,000-row sheet | 10,004 | 10,000 |\n| Cells evaluated by one keystroke in that sheet | 4 | 1 |\n| Cells evaluated when a 10,000-cell `SUM` reads the edit | 3 | 1 |\n\nA count is a claim about the algorithm; a millisecond is a claim about the\nhardware. Only the first belongs in CI. Budgets live beside each case in\n`tools/bench/cases.mjs`, and the spreadsheet's in\n`tools/bench/sheet-cases.mjs`.\n\nEach case also declares which counters must be *present*, not just within\nbudget. Without that, a stage that stopped running altogether would satisfy\nevery budget by doing no work, which is the failure mode a gate like this\nis most likely to develop.\n\n## Memory\n\n**Measured** (`pnpm bench`, Node heap): building the row model for\n100,000 rows x 9 columns with no sort, filter or grouping retains\n**0.8 MB**, with no renderer attached. A row object is made the first time\nsomething reads it, so a grid showing twenty rows makes a screenful of\nthem, not 100,000.\n\nThe first sort, filter or grouping reads every row and makes them all. The\nbrowser harness measures that as the \"after sorting and filtering\" row in\nthe [comparison](./comparison.md#measured-performance): 17.8 MB at\n100,000 rows, and 128 MB at 1,000,000 (27 Oct 2026,\n`BENCH_ROWS=1000000 BENCH_GRIDS=svgrid pnpm bench:compare`), so about 130\nbytes a row: the row object, its id, and the sort key it keeps so a live\nupdate can be merged in without re-reading the rows. The 1,000,000-row\nfigure was 196 MB the day before. Reading one column used to build an\narray of every column's value on every row and keep it there, which only\nthe rows on screen use; a plain `field` column is now read from the data,\nand only a computed (`fieldFn`) column still builds the array.\n\nThe fully built row model was 16.0 MB at 100,000 rows before rows became\nlazy, and 64.9 MB before that. Each row was an object literal whose seven\nmethods were closures, so a 100k-row grid allocated 700k closures and a\nclosure scope per row before painting anything. The methods are now\ndefined once and shared, reading their row through `this`: 4x less memory\nand about 7x faster to build. They are still assigned as own properties\nrather than put on a prototype, because `Row` is public and several row\nmodels legitimately do `{ ...row, depth }` - a spread copies own\nproperties but not a prototype, so a class would silently strip the\nmethods off a cloned row.\n\nAn earlier version of this page said \"~22 MB\". That figure was not\nmeasured. Node's heap accounting is not the same as a browser's, so read\nthis as the engine's own cost rather than what a tab will show.\n\nTwo claims here are structural rather than measured, and both hold:\n\n- Virtualization keeps the rendered DOM to the visible window regardless\n of dataset size, so `<td>` count does not scale with row count.\n- No retained references when the grid unmounts - the cleanup path is\n exercised by the unmount test in `svgrid.behavior.test.ts`.\n\n## Server-side / chunked loading\n\nDemo [33. Server-side infinite scroll](https://svgrid.com/demos/33-server-infinite/) covers the chunked-load path. Numbers from that demo:\n\n| Scenario | Result |\n| ------------------------------------------------ | --------------------------------------- |\n| Initial paint, sparse 100k-row dataset | 110 ms to first chunk visible |\n| Scroll 50,000 rows in 1.5 s (fast wheel-flick) | 16 chunk requests cancelled mid-flight |\n| Sort 100k server-side rows | round-trip dominated by the mock latency (50-140 ms) |\n\n### The Server-Side Row Model over one million rows\n\nDemo [467](https://svgrid.com/demos/467-server-row-model-1m/) runs the\nEnterprise row model over a columnar in-memory warehouse of a million rows,\ngrouped Region > Country. `tests/perf/server-row-model.spec.ts` drives it in\nheadless Chromium (a developer workstation, 17 September 2026; a release\nritual, not a gate):\n\n| Scenario | Result |\n| ------------------------------------------------------------- | --------------------------------------------- |\n| First paint, navigation to the first group row (includes building the warehouse in the page) | 2.4 s |\n| Scroll 60,000 px through an open country of 62,000 leaves | 17 leaf requests, 0.28 per 1,000 px (one block of 100 rows is 3,400 px) |\n| Worst frame while blocks streamed in | 83 ms (layout of the changed rows; the grid keeps the row objects a block did not touch) |\n| Heap before / after streaming 700,000 px with `maxBlocksInCache: 24` | 225 MB / 225 MB - evicted blocks are collected |\n| Warehouse: a million rows as typed arrays | 22 MB; a cold sort on a new column 0.8 to 1.1 s, cached after; a grouped level 140 ms; a pivot with a grand total 330 ms |\n\n## AI helpers\n\nEnd-to-end timings against the bundled `mockAIProvider`:\n\n| Helper | Median time (ms) |\n| --------------- | ---------------- |\n| `aiFilter` | 350-750 (mock latency dominated) |\n| `aiSmartFill` (50 rows) | 400-900 |\n| `aiSummarize` | 350-750 |\n| `aiClassify` (20 rows) | 400-750 |\n\nAgainst a real model the latency is provider-side. The grid's own\nprompt-build + result-parse work stays under ~6 ms even for 1000-row\nclassify jobs.\n\n## Import / export\n\n| Operation | Time |\n| ------------------------------------ | ------ |\n| Parse CSV, 10k rows × 9 cols | 28 ms |\n| Parse xlsx, 10k rows × 9 cols | 140 ms (jszip unzip-dominated) |\n| Export CSV, 10k rows × 9 cols | 18 ms |\n| Export xlsx, 10k rows × 9 cols | 220 ms |\n| Export PDF, 1k rows × 9 cols (pdfmake) | 700 ms |\n\n## Reproducing locally\n\n```bash\ngit clone https://github.com/sv-grid/sv-grid\ncd sv-grid\npnpm install\npnpm --filter @svgrid/grid build # the bench runs the BUILT engine\npnpm bench # engine + memory tables\npnpm size # bundle table\npnpm bench:dom # first paint + scroll, in Chromium\n```\n\nThe bench takes a few options:\n\n```bash\npnpm bench --json # machine-readable, for trend tracking\npnpm bench --baseline=tools/bench/baseline.json # adds a delta column\npnpm bench --case=sort-1col # one case\npnpm bench:check # the work-counter gate CI runs\n```\n\n`pnpm bench` runs against the built `@svgrid/grid/core` rather than\n`src/`, so it also smoke-tests the entry a consumer actually imports.\n\n`pnpm bench:dom` drives Chromium through Playwright against the examples\ngallery, which lives in this repository, so it needs no private\nsubmodule. It is a Playwright project rather than part of `pnpm test:e2e`,\nand it is deliberately not in CI - see below.\n\nFor your own workload, the [benchmark harness\nrecipe](../recipes/benchmark-harness.md) and the [FPS HUD\nrecipe](../recipes/profiling-with-fps-hud.md) are copy-paste probes you\ncan drop into your app.\n\n## What we *don't* claim\n\n- That we are faster than any other grid. Nothing on *this* page is a\n head-to-head measurement. Everything here is sv-grid measured against\n sv-grid, which supports \"this got 10x faster\" and not \"this is faster\n than X\". For a real comparison there is a separate grid-agnostic\n harness - `pnpm bench:compare`, results and caveats in\n [Comparison](./comparison.md) - and it does not say we win everything.\n- That first paint is good. The grid half of the 100k x 100 case is still\n the worst number on this page, and this round did not touch it.\n- Any scroll figure above 60 fps. The harness is rAF-driven and cannot\n observe headroom past the display refresh. \"Zero dropped frames\" is the\n claim; \"120 fps\" is not.\n- Lab-grade first-paint numbers. Those come from a developer machine with\n roughly 2x run-to-run variance, not a dedicated rig.\n- \"Smoothest grid on the market\" - that depends entirely on what your\n cells render. A sparkline + currency formatter in every cell costs\n more than a number, and we don't pretend otherwise.\n- \"Zero allocations during scroll\" - the virtualizer recycles DOM\n nodes but cell snippets still allocate. The numbers above include\n real-world snippets (status pills, mini-bars).\n- Single-thread performance > 1M rows. For >1M, do the heavy lifting\n on the server and feed chunks through the [server-side infinite\n scroll pattern](https://svgrid.com/demos/33-server-infinite/).\n\n## Frequently asked questions\n\n### How fast is SvGrid?\n\nIt virtualizes both rows and columns, so only the visible window is in the DOM,\nand a million rows scrolls without dropping a frame. On the data side, a\nsingle-column sort of 100,000 rows takes about 27 ms and a one-operator filter\nabout 5 ms. Every table on this page comes from a checked-in script you can run\nyourself, and the numbers moved because the harness exposed real defects rather\nthan because the prose was rewritten.\n\n### How many rows can SvGrid handle?\n\nClient-side, 100k+ rows scroll smoothly thanks to virtualization. For millions\nof rows, page or chunk from the server (see Server-side data). The DOM only ever\nholds the visible window regardless of total row count.\n\n### How fast is SvGrid, and how big is it?\n\nIt ships a small bundle (the measured figures for the full render component\nand for the headless core alone are in the table above) and virtualizes by\ndefault. Raw scroll\nperformance is comparable for typical workloads; the bigger practical win is\nbundle size and a Svelte-native runtime with no framework bridge.\n\n## See also\n\n- [Browser support](./browser-support.md) - the matrix the benchmarks\n ran against.\n- [Testing and quality](./testing-and-quality.md) - the coverage\n thresholds that gate every release.\n"
5315
5315
  },
5316
5316
  {
5317
5317
  "slug": "help/bolt-new",
@@ -5332,7 +5332,7 @@ export const docs = [
5332
5332
  "path": "docs/help/bundle-size.md",
5333
5333
  "title": "Bundle size",
5334
5334
  "section": "Help",
5335
- "markdown": "# Bundle size\n\nWhat SvGrid costs in your bundle, how to reproduce the number on your\nbranch, and what to do if size matters.\n\n## Measured\n\nThe table below is generated from `docs/_data/svgrid-size.json`, the file\n`pnpm size:json` writes; `node tools/sync-guide-facts.mjs` rewrites it and\n`tools/competitor-facts.test.ts` fails when it is stale or when a size is\ntyped anywhere else on this page. An earlier version of this page typed its\nnumbers and sat a month behind the measurement.\n\n<!-- size:start -->\nRe-measured **10 Nov 2026** at `@svgrid/grid` 3.0.33 with the script that ships in the repo (`pnpm size:json`):\n\n| Target | Base JS (gzip) | CSS (gzip) | Loaded on demand |\n| --- | ---: | ---: | ---: |\n| Headless core (`createSvGrid`) | **3.2 KB** | - | - |\n| Headless subpath (`@svgrid/grid/core`) | **10.1 KB** | - | - |\n| Full render component (`<SvGrid>`) | **99.9 KB** | **11.3 KB** | **198.9 KB** |\n| Standalone chart (`<SvChart>`) | **71.9 KB** | - | **15.0 KB** |\n<!-- size:end -->\n\nThe two headless rows measure different things on purpose. `createSvGrid` is\nthe engine plus the row model it needs, which is what you pay when you import\njust that symbol. The subpath row is every export on `@svgrid/grid/core`\npulled in at once with nothing tree-shaken, so it is the ceiling rather than\nthe typical cost - a real consumer importing `createSvGrid` and two row models\nsits near the lower number.\n\nSvelte is a peer dependency and is excluded from every figure. Builds are\nminified and gzipped at level 9.\n\nThe \"loaded on demand\" column is code that is reachable only through\n`import()`, so it never lands in your initial bundle. The script prints the\nper-chunk breakdown; the chunks and what triggers each of them:\n\n| Chunk | Loads when |\n| --- | --- |\n| `SvDateTimePicker` | a date / datetime / time cell editor opens |\n| `chart` (engine) | charting is enabled |\n| `SvGridChart`, `SvGridChartPanel`, `SvGridChartView` | a chart renders, the chart panel opens, the grid switches to chart view |\n| `GridMenus` | a header or context menu opens |\n| `SvGridCellEditor` | editing is enabled (loads at mount, before the first edit) |\n| `SvGridDropdown` | a list / chips cell editor or the page-size picker opens |\n| `dismissable`, `popover`, `focus-trap` | any popover, menu or dropdown layer opens |\n| `export-format` | CSV / TSV / JSON export or clipboard copy runs |\n| `date-format` | a date column formats a value |\n| `column-resize`, `row-resize`, `row-drag-touch` | the first resize drag or touch row drag |\n\nThe Kanban board and the scheduler / calendar view are not in either\nfigure: their renderers live in `@svgrid/enterprise` and register into the\nfree grid through the board and scheduler view seams.\n\n## Reproduce on your branch\n\n`measure-size.mjs` runs one isolated Vite library build per target above,\nwith Svelte marked external, then gzips each emitted chunk and classifies\nit as `base` (statically reachable from the entry) or `lazy` (reachable\nonly via `import()`).\n\n```bash\nnode packages/grid/scripts/measure-size.mjs\n# or, from the repo root:\npnpm size\n```\n\nTo see where the weight sits inside the base bundle:\n\n```bash\ncorepack pnpm --filter @svgrid/grid build\nnpx source-map-explorer packages/grid/dist/index.js\n```\n\nA treemap opens in your browser. Each block is a source file scaled by its\nbyte cost in the final bundle.\n\n## @svgrid/enterprise\n\nThe Enterprise pack is a separate install and a separate bundle. `xlsx`\nexport pulls JSZip and PDF export pulls pdfmake as optional peer\ndependencies, imported on the first `api.exportData(...)` call rather than\nat module load, so neither is in your synchronous bundle.\n\n## What to do if size matters\n\n1. **Use the headless engine for read-only views.** When you only need to\n display server-side data with no interaction, `createGrid` plus a short\n `<table>` renderer costs the headless-core row of the table above rather\n than the full-component row. See the\n [headless engine reference](../reference/headless-engine.md).\n2. **Register only the features you use.** The grid is feature-gated:\n sorting, filtering, grouping, pagination, expansion, and selection are\n each opt-in and tree-shake out when not imported. See the\n [features reference](../reference/features.md).\n3. **Let the lazy chunks stay lazy.** Charts, date/time editors, menus,\n and export already split themselves. Importing their modules directly\n at the top level pulls them back into your base bundle.\n4. **Code-split the Enterprise pack.** `installEnterprise(api)` is\n async-safe, so import it in the route that needs export rather than at\n module load:\n `const { installEnterprise } = await import('@svgrid/enterprise')`.\n\n## See also\n\n- [Features reference](../reference/features.md) - what each feature does\n- [Headless engine reference](../reference/headless-engine.md) - skip the renderer entirely\n- [Going to production guide](../getting-started/6-going-to-production.md)\n"
5335
+ "markdown": "# Bundle size\n\nWhat SvGrid costs in your bundle, how to reproduce the number on your\nbranch, and what to do if size matters.\n\n## Measured\n\nThe table below is generated from `docs/_data/svgrid-size.json`, the file\n`pnpm size:json` writes; `node tools/sync-guide-facts.mjs` rewrites it and\n`tools/competitor-facts.test.ts` fails when it is stale or when a size is\ntyped anywhere else on this page. An earlier version of this page typed its\nnumbers and sat a month behind the measurement.\n\n<!-- size:start -->\nRe-measured **10 Nov 2026** at `@svgrid/grid` 3.0.34 with the script that ships in the repo (`pnpm size:json`):\n\n| Target | Base JS (gzip) | CSS (gzip) | Loaded on demand |\n| --- | ---: | ---: | ---: |\n| Headless core (`createSvGrid`) | **3.2 KB** | - | - |\n| Headless subpath (`@svgrid/grid/core`) | **10.2 KB** | - | - |\n| Full render component (`<SvGrid>`) | **99.6 KB** | **11.4 KB** | **206.3 KB** |\n| Standalone chart (`<SvChart>`) | **71.9 KB** | - | **16.0 KB** |\n<!-- size:end -->\n\nThe two headless rows measure different things on purpose. `createSvGrid` is\nthe engine plus the row model it needs, which is what you pay when you import\njust that symbol. The subpath row is every export on `@svgrid/grid/core`\npulled in at once with nothing tree-shaken, so it is the ceiling rather than\nthe typical cost - a real consumer importing `createSvGrid` and two row models\nsits near the lower number.\n\nSvelte is a peer dependency and is excluded from every figure. Builds are\nminified and gzipped at level 9.\n\nThe \"loaded on demand\" column is code that is reachable only through\n`import()`, so it never lands in your initial bundle. The script prints the\nper-chunk breakdown; the chunks and what triggers each of them:\n\n| Chunk | Loads when |\n| --- | --- |\n| `SvDateTimePicker` | a date / datetime / time cell editor opens |\n| `chart` (engine) | charting is enabled |\n| `SvGridChart`, `SvGridChartPanel`, `SvGridChartView` | a chart renders, the chart panel opens, the grid switches to chart view |\n| `GridMenus` | a header or context menu opens |\n| `SvGridCellEditor` | editing is enabled (loads at mount, before the first edit) |\n| `SvGridDropdown` | a list / chips cell editor or the page-size picker opens |\n| `dismissable`, `popover`, `focus-trap` | any popover, menu or dropdown layer opens |\n| `export-format` | CSV / TSV / JSON export or clipboard copy runs |\n| `date-format` | a date column formats a value |\n| `column-resize`, `row-resize`, `row-drag-touch` | the first resize drag or touch row drag |\n\nThe Kanban board and the scheduler / calendar view are not in either\nfigure: their renderers live in `@svgrid/enterprise` and register into the\nfree grid through the board and scheduler view seams.\n\n## Reproduce on your branch\n\n`measure-size.mjs` runs one isolated Vite library build per target above,\nwith Svelte marked external, then gzips each emitted chunk and classifies\nit as `base` (statically reachable from the entry) or `lazy` (reachable\nonly via `import()`).\n\n```bash\nnode packages/grid/scripts/measure-size.mjs\n# or, from the repo root:\npnpm size\n```\n\nTo see where the weight sits inside the base bundle:\n\n```bash\ncorepack pnpm --filter @svgrid/grid build\nnpx source-map-explorer packages/grid/dist/index.js\n```\n\nA treemap opens in your browser. Each block is a source file scaled by its\nbyte cost in the final bundle.\n\n## @svgrid/enterprise\n\nThe Enterprise pack is a separate install and a separate bundle. `xlsx`\nexport pulls JSZip and PDF export pulls pdfmake as optional peer\ndependencies, imported on the first `api.exportData(...)` call rather than\nat module load, so neither is in your synchronous bundle.\n\n## What to do if size matters\n\n1. **Use the headless engine for read-only views.** When you only need to\n display server-side data with no interaction, `createGrid` plus a short\n `<table>` renderer costs the headless-core row of the table above rather\n than the full-component row. See the\n [headless engine reference](../reference/headless-engine.md).\n2. **Register only the features you use.** The grid is feature-gated:\n sorting, filtering, grouping, pagination, expansion, and selection are\n each opt-in and tree-shake out when not imported. See the\n [features reference](../reference/features.md).\n3. **Let the lazy chunks stay lazy.** Charts, date/time editors, menus,\n and export already split themselves. Importing their modules directly\n at the top level pulls them back into your base bundle.\n4. **Code-split the Enterprise pack.** `installEnterprise(api)` is\n async-safe, so import it in the route that needs export rather than at\n module load:\n `const { installEnterprise } = await import('@svgrid/enterprise')`.\n\n## See also\n\n- [Features reference](../reference/features.md) - what each feature does\n- [Headless engine reference](../reference/headless-engine.md) - skip the renderer entirely\n- [Going to production guide](../getting-started/6-going-to-production.md)\n"
5336
5336
  },
5337
5337
  {
5338
5338
  "slug": "help/cells/cell-components",
@@ -5647,7 +5647,7 @@ export const docs = [
5647
5647
  "path": "docs/help/comparison.md",
5648
5648
  "title": "Comparison: SvGrid vs AG Grid vs TanStack Table",
5649
5649
  "section": "Help",
5650
- "markdown": "\n# Comparison: SvGrid vs AG Grid vs TanStack Table\n\nThe three projects solve overlapping problems, and the right choice\ndepends on the framework you ship on, how much of the grid you want to\nbuild yourself, and your budget. Every number on this page is read from\nnpm or measured by a script in this repository on the date shown; the\nblock below is regenerated by `node tools/sync-guide-facts.mjs` and the\nper-competitor pages under [/compare/](https://svgrid.com/compare/) list\nthe sources for every feature claim.\n\n<!-- facts:start ag-grid,tanstack-table -->\n> **Facts, checked 9 Nov 2026.** `ag-grid-community` 36.3.0, MIT, last published 4 Nov 2026, 16,400,000 npm downloads in the 30 days to 8 Nov 2026. `@tanstack/svelte-table` 9.3.1, MIT, last published 12 Oct 2026, 539,000 npm downloads in the same window. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `ag-grid-community` 36.3.0 346.4 KB JS, no separate stylesheet (measured 9 Nov 2026); `@tanstack/svelte-table` 9.3.1 36.5 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). AG Grid pricing, as its site states it: AG Grid Community is free under MIT. AG Grid Enterprise is listed at $999 USD per developer with one year of updates and Zendesk support; the Enterprise Bundle with AG Charts Enterprise is $1,498 USD per developer (https://www.ag-grid.com/license-pricing/, read 12 Sep 2026). TanStack Table pricing, as its site states it: TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table (https://tanstack.com/table/latest, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/), [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/).\n<!-- facts:end -->\n\n## TL;DR\n\n| Project | Lives in | Ships | Licence |\n| --------------------- | ---------------------------------------- | ---------------------------------------------------------- | ----------------------------------------- |\n| **SvGrid** | Svelte 5 | Headless core + Svelte render component + Enterprise pack | MIT (Community) / commercial (Enterprise) |\n| **AG Grid Community** | React, Angular, Vue, plain JS | Full grid + renderer; Svelte through community integrations | MIT |\n| **AG Grid Enterprise**| same | Adds the set filter, range selection, master/detail, tree, grouping, pivot, charts, Excel export, server-side row model | Commercial, per developer |\n| **TanStack Table** | React, Vue, Svelte, Solid, Qwik, Lit, JS | Headless engine **only**; the Svelte adapter is Svelte 5 | MIT |\n\nThe bundle sizes and prices are in the facts block above, with the date\nthey were read and the method behind the sizes.\n\n## When SvGrid is the right choice\n\n- You're on **Svelte 5** and want a grid that uses the runtime's idioms\n (snippets for cells, `$state` for data, `$derived` for aggregates) -\n not a React-port pretending to be Svelte.\n- You want a **headless core you can render yourself** AND a\n default-styled component for the 80% case. Most \"headless\" libraries\n make you write the markup; most \"monolith\" libraries make you fight\n the markup. SvGrid does both in one package.\n- You need **clean theming via CSS custom properties** and a documented\n `--sg-*` token surface, not a hard-coded class soup.\n- You ship under **strict CSP** (no `eval`, no `new Function`, no\n inline scripts). SvGrid runs clean; AG Grid Community does too.\n TanStack Table is engine-only so the question doesn't apply.\n- You want **SSR markup that is meaningful before hydration** (good\n first paint, SEO, SvelteKit `+page.server` integration). SvGrid +\n TanStack Table both qualify. AG Grid renders client-side.\n- You want the features AG Grid sells as Enterprise - the set filter,\n range selection, master/detail, tree data, grouping with aggregation,\n integrated charts, flat server-side infinite scroll - **in the MIT\n core**. (Server-side grouping, tree, pivot and transactions are\n Enterprise in both.)\n\n## When AG Grid is the right choice\n\n- You're on **React, Angular, or Vue**, not Svelte. SvGrid is\n Svelte-only, and ag-grid.com points Svelte users to community-built\n integrations rather than an official package.\n- You need a push-based **viewport row model** for a real-time trading\n blotter, where the server streams the visible window over a socket.\n SvGrid's server-side row model covers sort, filter, grouping, tree,\n pivot, transactions and select-all across unloaded rows, and its\n real-time page merges deltas into loaded rows, but it does not have the\n viewport model.\n- You need **custom tool panels**. SvGrid's tool panel is a fixed\n Columns + Filters pair. Custom filter components are covered by a\n column's `customFilter`.\n- You need **AG Charts** for maps, network graphs, 3D or a canvas renderer.\n SvGrid's chart is SVG with decimation, and it ships 29 types, a configurable\n axis model, synchronized charts, a financial toolkit with indicator panes\n and drawing tools, a builder and PDF export in the MIT core; it does not\n draw maps.\n- You want a vendor with **a decade of enterprise procurement paperwork**\n already on file.\n\n## When TanStack Table is the right choice\n\n- You want a **rendering-framework-agnostic engine** so the same\n business logic powers React + Svelte + Solid surfaces in your\n monorepo.\n- You're already in the TanStack ecosystem (Query, Router, Form,\n Virtual) and want one mental model.\n- You're happy writing **all the markup yourself** - the row recycling,\n the keyboard map, the ARIA roles, the focus management, the\n drag-to-resize. That's the cost of \"engine only\".\n\n## Feature parity at a glance\n\nYes: shipped in that edition. Partial: needs extra work or a companion\npackage. No: not available. The AG Grid columns follow ag-grid.com's\nCommunity vs Enterprise page and its Set Filter, Tree Data and CSV export\npages; the dated list of those sources is on the\n[SvGrid vs AG Grid](https://svgrid.com/compare/ag-grid/) page, and the\nTanStack sources on [SvGrid vs TanStack Table](https://svgrid.com/compare/tanstack-table/).\n\n| | SvGrid Community | SvGrid Enterprise | AG Grid Community | AG Grid Enterprise | TanStack Table |\n| ------------------------------- | ---------------- | ----------------- | ----------------- | ------------------ | -------------- |\n| Headless core (engine only) | Yes | Yes | No | No | Yes |\n| Default render component | Yes (Svelte 5) | Yes | Yes (each framework) | Yes | No |\n| Sort (multi-column) | Yes | Yes | Yes | Yes | Yes (engine) |\n| Filter menu (operator + facet) | Yes | Yes | Partial (text and number filters; the set filter is Enterprise) | Yes | No |\n| Filter row | Yes | Yes | Yes | Yes | No |\n| Pagination | Yes | Yes | Yes | Yes | Yes (engine) |\n| Grouping + aggregation | Yes | Yes | No (Enterprise) | Yes | Yes (engine) |\n| Tree / expand-collapse rows | Yes | Yes | No (Enterprise) | Yes | Yes (engine) |\n| Master/detail | Yes | Yes | No (Enterprise) | Yes | No |\n| Cell range selection + copy/paste | Yes | Yes | No (Enterprise) | Yes | No |\n| Fill handle | Yes | Yes | No (Enterprise) | Yes | No |\n| Inline editing (typed editors) | Yes | Yes | Yes | Yes | No |\n| Column virtualization | Yes | Yes | Yes | Yes | No |\n| Row virtualization | Yes | Yes | Yes | Yes | Partial (pair with TanStack Virtual) |\n| Column pinning (left/right) | Yes | Yes | Yes | Yes | No |\n| Column tool panel | Yes | Yes | No (Enterprise) | Yes | No |\n| Context menu | Yes | Yes | No (Enterprise) | Yes | No |\n| Fit-to-width with shrink | Yes | Yes | Partial | Yes | No |\n| WAI-ARIA grid pattern | Yes | Yes | Yes | Yes | No (your markup) |\n| Server-side data, flat (sort / filter / paging / infinite scroll) | Yes (built in) | Yes (built in) | Yes (infinite row model) | Yes (built in) | Partial (manual flags) |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | No (Enterprise) | Yes (built in) | No (Enterprise) | Yes (built in) | No |\n| Global search over server-side data | Yes (`filterModel.global`) | Yes | No (the quick filter is client-side only) | No | n/a |\n| CSP-clean (no eval, no inline) | Yes | Yes | Yes | Yes | n/a |\n| Meaningful SSR markup | Yes | Yes | No | No | depends on your markup |\n| CSV / TSV / JSON export | Yes | Yes | Yes (CSV) | Yes | No |\n| Excel (.xlsx) / PDF export | No | Yes | No | Yes (Excel) | No |\n| Print | No | Yes | No | No | No |\n| Excel / CSV import | No | Yes | No | No | No |\n| In-grid AI helpers (bring your own model) | Yes | Yes | No | No | No |\n| MCP server for AI assistants | Yes (`@svgrid/mcp`) | Yes | Yes (`ag-mcp`) | Yes | No (agent skills ship in the package) |\n| Pivot table | No | Yes | No | Yes | No (custom) |\n| Integrated charts | Yes (29 types, builder, financial toolkit) | Yes | No | Yes (AG Charts) | No |\n| Support response time | GitHub (best-effort) | 1 business day; 1h sev-1 on Enterprise | GitHub (best-effort) | no number published | GitHub; paid Enterprise Support offered |\n| Theming via CSS variables | Yes (`--sg-*`) | Yes | Yes (theme builder) | Yes | n/a |\n| Source button per demo | Yes (gallery) | Yes | No | No | No |\n\n## Bundle size\n\nThe facts block at the top of the page has the gzipped size of each\npackage, measured by the same method: each package built alone with Vite\nin library mode, minified, gzip level 9, Svelte external because it is a\npeer dependency. SvGrid's own number comes from `pnpm size:json`\n(`packages/grid/scripts/measure-size.mjs`), the competitors' from\n`pnpm competitors:measure`; both write `docs/_data` and the block is\nregenerated from there, so the page cannot quote a size nobody measured.\n\nWhat the SvGrid number contains: the whole render component -\nvirtualization, Excel-style filters, inline editing, grouping, tree,\nmaster/detail, cell selection and accessibility - in one import, plus a\nseparate stylesheet. Charts, date and time editors, menus and export split\ninto `import()` chunks that load on demand rather than shipping in the\ninitial bundle; the headless core on its own is a small fraction of the\nrender component, and `@svgrid/enterprise` features are separate subpath\nimports. The per-entry breakdown, including the headless core and the\nlazy chunks, is in the [bundle size reference](./bundle-size.md).\n\nWhat the AG Grid number contains: `createGrid` with every Community\nmodule registered and the Quartz theme, whose CSS ships inside the script\nunder the theming API, so there is no separate stylesheet to add. The\nTanStack number is the whole Svelte adapter with its engine: no\nvirtualizer, filter UI or editors, because a headless library has none.\n\n## Measured performance\n\nA vendor's own benchmark is worth nothing unless you can rerun it, so the\nharness is in the repository and takes an adapter per grid. Adding another\ngrid is about forty lines in `examples/src/bench/adapters.ts` and nothing\nelse.\n\n```bash\n# headless, prints the table and writes tests/perf/.last-compare.json\nBENCH_REPEATS=7 BENCH_GRIDS=svgrid,aggrid,svar,tanstack pnpm bench:compare\n\n# move that run into docs/_data/competitors.json, then regenerate this page\nnode tools/record-benchmarks.mjs && node tools/sync-guide-facts.mjs\n\n# or drive it by hand\npnpm --filter @svgrid/grid-example-gallery dev # then open /bench.html\n```\n\nEach operation is timed until the grid's DOM reflects it, plus one frame:\n\n- **Mount** - fresh container, create the grid, until rows are painted.\n- **Sort** - change the sort on one column, until the new order is painted.\n Text sorts `name` (tens of thousands of distinct values), numeric sorts\n `amount`.\n- **Filter** - apply a `contains` filter to one column.\n- **Scroll** - 180 frames at 60 px/frame, sampling frame intervals.\n\n### Results\n\nThe table is generated from the ledger by `tools/sync-guide-facts.mjs`;\nthe versions and the date are part of it.\n\n<!-- bench:start -->\nMeasured 28 Oct 2026 with `pnpm bench:compare` (tests/perf/compare.spec.ts, recorded by tools/record-benchmarks.mjs): 100,000 rows x 9 columns; 1000 x 520 px container, 32 px rows, 140 px columns, each grid with its default theme; fastest of 3 samples per operation, each grid in a fresh page; Chromium through Playwright, headless; win32 10.0.26200 x64, Intel(R) Core(TM) Ultra 7 255U; a developer workstation, not a dedicated bench rig; tick: 1,000 of the rows replaced per tick with a new amount, the grid sorted by amount, on the update path each grid provides, timed until the changed cells paint; 180 ticks after 10 to warm up. Lower is better.\n\n| Operation | SvGrid | AG Grid Community | SVAR Svelte DataGrid | TanStack Table, minimal DOM |\n| --- | --- | --- | --- | --- |\n| Mount | 30.7 ms | 67 ms | 11.4 ms | 170.2 ms |\n| Sort, text column | 77.4 ms | 213.9 ms | 135.4 ms | 360.9 ms |\n| Sort, numeric column | 32.8 ms | 216.7 ms | 141.7 ms | 178.2 ms |\n| Filter, one column (indicative only, see below) | 34.5 ms | 148.8 ms | 26.1 ms | 51.1 ms |\n| Scroll, p95 frame | 34.3 ms | 34.5 ms | 34.6 ms | 34.3 ms |\n| Live tick, 1,000 rows replaced, sorted by that column, p95 | 48.6 ms | 259.2 ms | 152.2 ms | 563.4 ms |\n| Main-thread work per 60 px scroll step, mean | 3.5 ms | 4.5 ms | 3 ms | 2.9 ms |\n| Frames dropped while scrolling, of 180 | 0 | 0 | 0 | 0 |\n| Ticks over one 60 Hz frame, of 180 | 115 | 180 | 180 | 180 |\n| Rows kept in the DOM (virtualization on) | 36 | 36 | 19 | 27 |\n| JS heap the grid adds, mounted, after GC | 5.7 MB | 42.1 MB | 2.3 MB | 109.9 MB |\n| JS heap the grid adds, after sorting and filtering, after GC | 17.6 MB | 26.2 MB | 3.9 MB | 113.3 MB |\n| Elements created per scroll frame | 0 | 30.1 | 16.8 | 18.7 |\n| Elements re-inserted per scroll frame (rows recycled) | 37.3 | 35.8 | 0 | 0 |\n\nVersions: `@svgrid/grid` 3.0.25, `ag-grid-community` 36.1.0, `wx-svelte-grid` 2.7.0, `@tanstack/svelte-table` 9.2.4.\n<!-- bench:end -->\n\nRead that as: **SvGrid sorts and keeps a live sort current faster than the\nother three, scrolling is level across all four, and SVAR is ahead on mount\nand memory** - it paints its first rows in under half of SvGrid's time and\nholds the least heap. Every grid scrolls at the display rate here, so the\np95 frame reads the same for all four; the work row under it is what a\nslower machine or a 120 Hz display would feel. There TanStack and SVAR do\nthe least per scroll step, about half a millisecond under SvGrid, and AG\nGrid the most. The row varies from run to run, so read the order of the\nfirst three as close, not settled. SvGrid makes its row objects the first time a\nsort or filter reads every row rather than at mount, which is why its two\nheap rows differ. The TanStack renderer pays for building a row object per\nrecord before it can draw anything. Anyone claiming a single headline\nnumber for a grid comparison is selling something.\n\n**The heap rows are the JS heap after a forced garbage collection, less the\nsame page with the data built and no grid,** so the 100,000 rows themselves\ncount against no one. The two DOM rows are counted over 60 untimed scroll\nframes after the timed ones. \"Created\" is elements that did not exist\nbefore; \"re-inserted\" is elements the grid already had and moved, which is\nhow a keyed list recycles a row from one end of the window to the other.\nSvGrid creates none and moves the two rows that cross the edge each frame;\na grid that rewrote its rows in place would score 0 on both.\n\n**The filter row is not a like-for-like comparison and is shown only because\nhiding a number we win looks worse than explaining it.** The grids' single-\ncolumn filter APIs differ: the harness drives AG Grid's quick filter, which\nsearches every column and is strictly more work than the single-column\n`contains` SvGrid performs, while SVAR and TanStack run a predicate over the\none column. Do not read a ratio into that row.\n\n**The TanStack column is an engine plus the least DOM an app could draw.**\nTanStack Table renders nothing, so its adapter draws the row model through\na fixed-height windowed table with no virtualizer library, no filter UI and\nno editors. Its mount and sort numbers include creating the engine's row\nobjects for every record; the grids around it do the same work inside their\nown models.\n\nSorting is where SvGrid's distance comes from, and it is not subtle: SvGrid\nresolves each sort clause once and precomputes a key array, and for a\ncolumn whose distinct values are far fewer than its rows it collates the\ndistinct values once and sorts by rank. Mounting is the other direction -\nsee the accounting in [benchmarks](./benchmarks.md).\n\nAt 1,000,000 rows the harness runs one grid per invocation\n(`BENCH_ROWS=1000000 BENCH_GRIDS=<one> BENCH_TICKS=30`); TanStack's\nadapter does not finish at that size. Those runs are not in the table\nbecause only a run that measured every grid together is comparable.\n\nTwo methodology notes, both of which were bugs in this harness before they\nwere notes, and both of which changed the numbers by more than any code\nchange did:\n\n- **Each grid runs in a fresh page.** They used to run one after another in\n the same one, where the first warmed the JIT, fragmented the heap and\n left a hundred thousand rows of garbage for the next. Back-to-back runs\n on an idle machine disagreed with themselves by a factor of two.\n That order effect is gone, but this workstation still varies by about\n 2x between whole runs taken minutes apart, for every grid at once, so\n compare the columns of one run and not one run against another.\n- **Operations are timed until the DOM changes, not for a fixed two\n frames.** The old fixed wait put a two-frame floor under every number -\n an operation that recomputed nothing at all measured the floor, and a\n mount whose rows were already in the DOM measured pure waiting - which\n flattered whichever grid was slower.\n- **Each figure is the fastest of the samples, not the median.** Noise\n only ever adds time, so the minimum is the cleanest estimate of what the\n machine can do, and medians on this box drifted by a third between\n consecutive runs - more than most of the differences being measured.\n Applied identically to every grid.\n\nWhat this table deliberately does not include:\n\n- **Features the grids do not measure the same way.** Grouping, pivot,\n editing and export all differ enough in shape that a single timing would\n compare two different amounts of work. The filter row above is the one\n case we publish anyway, labelled, rather than quietly dropping a result\n that happens to favour us.\n- **Any grid we cannot license for this.** The harness is grid-agnostic on\n purpose, but some commercial grids licence their software in ways that\n restrict using it to produce competitive claims. Where that is the case\n we do not ship an adapter, and we would rather have an obviously\n incomplete table than a legally dubious one. Run your own shortlist.\n- **Your workload.** One dataset shape, one machine, one browser, default\n configuration for every grid. If a grid decision rides on this, clone the\n harness and put your own data in it.\n\n## Migrating from AG Grid\n\nThe most common starting point. See\n[Migrating from AG Grid](./migrating-from-ag-grid.md) for a side-by-side\ntranslation of column defs, features, filtering, editing, and the\nimperative API. The example that guide is built around runs here:\n\n<div data-docs-demo=\"139-migration-from-ag-grid\" data-height=\"560\"></div>\n\n## Migrating from TanStack Table\n\nThe map is one-to-one - SvGrid's headless core speaks the vocabulary\nTanStack Table v9 introduced. The big differences:\n\n- Replace `createTable(opts)` with `createSvGrid(opts)`. Same state\n machine, same features object.\n- Replace the row-model factories with the same names from\n `@svgrid/grid`.\n- The render layer changes - TanStack hands you `FlexRender` + the\n row model; SvGrid lets you keep that headless approach OR drop in\n the default `<SvGrid>` component.\n\nThe headless demo below drives a plain `<table>` from `createSvGrid`, the\nway a TanStack app would:\n\n<div data-docs-demo=\"186-headless-table\" data-height=\"520\"></div>\n\nAnd the same rows through the render component, which is the part TanStack\nleaves to you. The example runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Order = {\n id: number\n customer: string\n region: string\n status: string\n amount: number\n }\n\n const orders: Order[] = [\n { id: 1, customer: 'Ada Lovelace', region: 'EMEA', status: 'open', amount: 1240 },\n { id: 2, customer: 'Grace Hopper', region: 'AMER', status: 'shipped', amount: 860 },\n { id: 3, customer: 'Linus Torvalds', region: 'EMEA', status: 'closed', amount: 2210 },\n { id: 4, customer: 'Radia Perlman', region: 'APAC', status: 'open', amount: 430 },\n { id: 5, customer: 'Barbara Liskov', region: 'AMER', status: 'pending', amount: 1780 },\n ]\n\n let rows = $state<Order[]>(orders)\n\n const columns: GridColumns<Order> = [\n { field: 'customer', header: 'Customer', width: 180 },\n { field: 'region', header: 'Region', width: 110 },\n { field: 'status', header: 'Status', width: 110 },\n { field: 'amount', header: 'Amount', width: 120, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable selectable />\n```\n\n## Pricing\n\nSvGrid Community is MIT - free for commercial use, no attribution\nrequired at runtime. SvGrid Enterprise is a per-developer licence: $599\nper developer per year for the enterprise grid or $999 with the\nspreadsheet and Studio, both for unlimited applications; see\n<https://svgrid.com/pricing/> for what each tier adds.\n\nAG Grid Community is MIT. AG Grid Enterprise is priced per developer with\none year of updates; the figure ag-grid.com listed on the date we read it\nis in the facts block at the top of this page, and the\n[SvGrid vs AG Grid](https://svgrid.com/compare/ag-grid/) page carries the\nlink. Compare what each paid tier adds before comparing prices: most of\nwhat AG Grid sells as Enterprise is in SvGrid's free core.\n\nTanStack Table is MIT and sponsor-funded; tanstack.com offers paid\nEnterprise Support as private consulting.\n\n## Frequently asked questions\n\n### What is the best data grid for Svelte 5?\n\nFor a Svelte-5-native grid with a batteries-included render component, SvGrid is\nbuilt around runes and snippets. TanStack Table is a strong headless-only\nchoice if you want to build the DOM layer yourself across frameworks, and its\nv9 Svelte adapter targets Svelte 5. AG Grid is the most feature-complete but\nlives in React, Angular and Vue first and reaches Svelte through\ncommunity-built integrations.\n\n### Is SvGrid a good AG Grid alternative?\n\nYes, for Svelte projects. SvGrid is a fraction of AG Grid Community's\nmeasured size on a Svelte page (the facts block has both numbers and the\nmethod), is MIT-licensed for commercial use, and offers `@svgrid/enterprise`\nfor export, print, import and pivot at a per-developer price. On the\nbenchmark above it sorts a hundred-thousand-row column faster on both\ncolumn types and scrolls level; AG Grid mounts quicker. It does not match\nevery AG Grid Enterprise feature - see the\n[missing-features list](./missing-features.md) for the gaps.\n\n### SvGrid vs TanStack Table - which should I pick?\n\nPick SvGrid if you want virtualization, Excel-style filters, selection, and\ninline editing working out of the box on Svelte 5. Pick TanStack Table if you\nwant a framework-agnostic headless engine and are happy to build the rendering,\nvirtualization, and editing UI yourself. Both are MIT-licensed.\n\n### Does the TanStack Svelte adapter support Svelte 5?\n\nYes. The `@tanstack/svelte-table` v9 line declares a Svelte 5 peer range\nand passes data through getters rather than stores; the facts block above\nshows the version read from npm. The earlier `createSvelteTable` constructor\nis gone in v9.\n\n### Does AG Grid have an MCP server like SvGrid?\n\nYes. `ag-mcp` (MIT) gives an assistant AG Grid's docs, examples and\nversion-specific upgrade prompts. `@svgrid/mcp` does the same for SvGrid\nand adds a code checker that validates props and imports against the real\nAPI surface, a scaffold tool and a preview. Neither is an in-grid feature;\nSvGrid's natural-language filter, smart fill and summarise helpers are, and\nthey ship free in `@svgrid/grid` with a model provider you register.\n\n### How big is the SvGrid bundle?\n\nThe facts block at the top of this page has the measured gzipped size of the\nfull `<SvGrid>` render component and its stylesheet, read from\n`docs/_data/svgrid-size.json`, which `pnpm size:json` writes. The\n[bundle size reference](./bundle-size.md) breaks it down by entry:\nthe headless core on its own, the core subpath, the charts, and the lazy\nchunks that load on demand.\n\n## See also\n\n- [Why headless?](../why-headless.md) - the design decision behind\n SvGrid's two-layer architecture.\n- [Migrating from AG Grid](./migrating-from-ag-grid.md) - the\n practical recipe.\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - the\n one-to-one API map.\n- [Benchmarks](./benchmarks.md) - how the numbers are produced and what\n they leave out.\n- [Enterprise feature pack](../enterprise/README.md) - what SvGrid charges for.\n- [Missing features](./missing-features.md) - the honest gap list\n versus AG Grid Enterprise.\n- [All comparisons](https://svgrid.com/compare/) - one page per grid, with\n dated sources for every claim.\n"
5650
+ "markdown": "\n# Comparison: SvGrid vs AG Grid vs TanStack Table\n\nThe three projects solve overlapping problems, and the right choice\ndepends on the framework you ship on, how much of the grid you want to\nbuild yourself, and your budget. Every number on this page is read from\nnpm or measured by a script in this repository on the date shown; the\nblock below is regenerated by `node tools/sync-guide-facts.mjs` and the\nper-competitor pages under [/compare/](https://svgrid.com/compare/) list\nthe sources for every feature claim.\n\n<!-- facts:start ag-grid,tanstack-table -->\n> **Facts, checked 9 Nov 2026.** `ag-grid-community` 36.3.0, MIT, last published 4 Nov 2026, 16,400,000 npm downloads in the 30 days to 8 Nov 2026. `@tanstack/svelte-table` 9.3.1, MIT, last published 12 Oct 2026, 539,000 npm downloads in the same window. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `ag-grid-community` 36.3.0 346.4 KB JS, no separate stylesheet (measured 9 Nov 2026); `@tanstack/svelte-table` 9.3.1 36.5 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). AG Grid pricing, as its site states it: AG Grid Community is free under MIT. AG Grid Enterprise is listed at $999 USD per developer with one year of updates and Zendesk support; the Enterprise Bundle with AG Charts Enterprise is $1,498 USD per developer (https://www.ag-grid.com/license-pricing/, read 12 Sep 2026). TanStack Table pricing, as its site states it: TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table (https://tanstack.com/table/latest, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/), [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/).\n<!-- facts:end -->\n\n## TL;DR\n\n| Project | Lives in | Ships | Licence |\n| --------------------- | ---------------------------------------- | ---------------------------------------------------------- | ----------------------------------------- |\n| **SvGrid** | Svelte 5 | Headless core + Svelte render component + Enterprise pack | MIT (Community) / commercial (Enterprise) |\n| **AG Grid Community** | React, Angular, Vue, plain JS | Full grid + renderer; Svelte through community integrations | MIT |\n| **AG Grid Enterprise**| same | Adds the set filter, range selection, master/detail, tree, grouping, pivot, charts, Excel export, server-side row model | Commercial, per developer |\n| **TanStack Table** | React, Vue, Svelte, Solid, Qwik, Lit, JS | Headless engine **only**; the Svelte adapter is Svelte 5 | MIT |\n\nThe bundle sizes and prices are in the facts block above, with the date\nthey were read and the method behind the sizes.\n\n## When SvGrid is the right choice\n\n- You're on **Svelte 5** and want a grid that uses the runtime's idioms\n (snippets for cells, `$state` for data, `$derived` for aggregates) -\n not a React-port pretending to be Svelte.\n- You want a **headless core you can render yourself** AND a\n default-styled component for the 80% case. Most \"headless\" libraries\n make you write the markup; most \"monolith\" libraries make you fight\n the markup. SvGrid does both in one package.\n- You need **clean theming via CSS custom properties** and a documented\n `--sg-*` token surface, not a hard-coded class soup.\n- You ship under **strict CSP** (no `eval`, no `new Function`, no\n inline scripts). SvGrid runs clean; AG Grid Community does too.\n TanStack Table is engine-only so the question doesn't apply.\n- You want **SSR markup that is meaningful before hydration** (good\n first paint, SEO, SvelteKit `+page.server` integration). SvGrid +\n TanStack Table both qualify. AG Grid renders client-side.\n- You want the features AG Grid sells as Enterprise - the set filter,\n range selection, master/detail, tree data, grouping with aggregation,\n integrated charts, flat server-side infinite scroll - **in the MIT\n core**. (Server-side grouping, tree, pivot and transactions are\n Enterprise in both.)\n\n## When AG Grid is the right choice\n\n- You're on **React, Angular, or Vue**, not Svelte. SvGrid is\n Svelte-only, and ag-grid.com points Svelte users to community-built\n integrations rather than an official package.\n- You need a push-based **viewport row model** for a real-time trading\n blotter, where the server streams the visible window over a socket.\n SvGrid's server-side row model covers sort, filter, grouping, tree,\n pivot, transactions and select-all across unloaded rows, and its\n real-time page merges deltas into loaded rows, but it does not have the\n viewport model.\n- You need **custom tool panels**. SvGrid's tool panel is a fixed\n Columns + Filters pair. Custom filter components are covered by a\n column's `customFilter`.\n- You need **AG Charts** for maps, network graphs, 3D or a canvas renderer.\n SvGrid's chart is SVG with decimation, and it ships 29 types, a configurable\n axis model, synchronized charts, a financial toolkit with indicator panes\n and drawing tools, a builder and PDF export in the MIT core; it does not\n draw maps.\n- You want a vendor with **a decade of enterprise procurement paperwork**\n already on file.\n\n## When TanStack Table is the right choice\n\n- You want a **rendering-framework-agnostic engine** so the same\n business logic powers React + Svelte + Solid surfaces in your\n monorepo.\n- You're already in the TanStack ecosystem (Query, Router, Form,\n Virtual) and want one mental model.\n- You're happy writing **all the markup yourself** - the row recycling,\n the keyboard map, the ARIA roles, the focus management, the\n drag-to-resize. That's the cost of \"engine only\".\n\n## Feature parity at a glance\n\nYes: shipped in that edition. Partial: needs extra work or a companion\npackage. No: not available. The AG Grid columns follow ag-grid.com's\nCommunity vs Enterprise page and its Set Filter, Tree Data and CSV export\npages; the dated list of those sources is on the\n[SvGrid vs AG Grid](https://svgrid.com/compare/ag-grid/) page, and the\nTanStack sources on [SvGrid vs TanStack Table](https://svgrid.com/compare/tanstack-table/).\n\n| | SvGrid Community | SvGrid Enterprise | AG Grid Community | AG Grid Enterprise | TanStack Table |\n| ------------------------------- | ---------------- | ----------------- | ----------------- | ------------------ | -------------- |\n| Headless core (engine only) | Yes | Yes | No | No | Yes |\n| Default render component | Yes (Svelte 5) | Yes | Yes (each framework) | Yes | No |\n| Sort (multi-column) | Yes | Yes | Yes | Yes | Yes (engine) |\n| Filter menu (operator + facet) | Yes | Yes | Partial (text and number filters; the set filter is Enterprise) | Yes | No |\n| Filter row | Yes | Yes | Yes | Yes | No |\n| Pagination | Yes | Yes | Yes | Yes | Yes (engine) |\n| Grouping + aggregation | Yes | Yes | No (Enterprise) | Yes | Yes (engine) |\n| Tree / expand-collapse rows | Yes | Yes | No (Enterprise) | Yes | Yes (engine) |\n| Master/detail | Yes | Yes | No (Enterprise) | Yes | No |\n| Cell range selection + copy/paste | Yes | Yes | No (Enterprise) | Yes | No |\n| Fill handle | Yes | Yes | No (Enterprise) | Yes | No |\n| Inline editing (typed editors) | Yes | Yes | Yes | Yes | No |\n| Column virtualization | Yes | Yes | Yes | Yes | No |\n| Row virtualization | Yes | Yes | Yes | Yes | Partial (pair with TanStack Virtual) |\n| Column pinning (left/right) | Yes | Yes | Yes | Yes | No |\n| Column tool panel | Yes | Yes | No (Enterprise) | Yes | No |\n| Context menu | Yes | Yes | No (Enterprise) | Yes | No |\n| Fit-to-width with shrink | Yes | Yes | Partial | Yes | No |\n| WAI-ARIA grid pattern | Yes | Yes | Yes | Yes | No (your markup) |\n| Server-side data, flat (sort / filter / paging / infinite scroll) | Yes (built in) | Yes (built in) | Yes (infinite row model) | Yes (built in) | Partial (manual flags) |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | No (Enterprise) | Yes (built in) | No (Enterprise) | Yes (built in) | No |\n| Global search over server-side data | Yes (`filterModel.global`) | Yes | No (the quick filter is client-side only) | No | n/a |\n| CSP-clean (no eval, no inline) | Yes | Yes | Yes | Yes | n/a |\n| Meaningful SSR markup | Yes | Yes | No | No | depends on your markup |\n| CSV / TSV / JSON export | Yes | Yes | Yes (CSV) | Yes | No |\n| Excel (.xlsx) / PDF export | No | Yes | No | Yes (Excel) | No |\n| Print | No | Yes | No | No | No |\n| Excel / CSV import | No | Yes | No | No | No |\n| In-grid AI helpers (bring your own model) | Yes | Yes | No | No | No |\n| MCP server for AI assistants | Yes (`@svgrid/mcp`) | Yes | Yes (`ag-mcp`) | Yes | No (agent skills ship in the package) |\n| Pivot table | No | Yes | No | Yes | No (custom) |\n| Integrated charts | Yes (29 types, builder, financial toolkit) | Yes | No | Yes (AG Charts) | No |\n| Support response time | GitHub (best-effort) | 1 business day; 1h sev-1 on Enterprise | GitHub (best-effort) | no number published | GitHub; paid Enterprise Support offered |\n| Theming via CSS variables | Yes (`--sg-*`) | Yes | Yes (theme builder) | Yes | n/a |\n| Source button per demo | Yes (gallery) | Yes | No | No | No |\n\n## Bundle size\n\nThe facts block at the top of the page has the gzipped size of each\npackage, measured by the same method: each package built alone with Vite\nin library mode, minified, gzip level 9, Svelte external because it is a\npeer dependency. SvGrid's own number comes from `pnpm size:json`\n(`packages/grid/scripts/measure-size.mjs`), the competitors' from\n`pnpm competitors:measure`; both write `docs/_data` and the block is\nregenerated from there, so the page cannot quote a size nobody measured.\n\nWhat the SvGrid number contains: the whole render component -\nvirtualization, Excel-style filters, inline editing, grouping, tree,\nmaster/detail, cell selection and accessibility - in one import, plus a\nseparate stylesheet. Charts, date and time editors, menus and export split\ninto `import()` chunks that load on demand rather than shipping in the\ninitial bundle; the headless core on its own is a small fraction of the\nrender component, and `@svgrid/enterprise` features are separate subpath\nimports. The per-entry breakdown, including the headless core and the\nlazy chunks, is in the [bundle size reference](./bundle-size.md).\n\nWhat the AG Grid number contains: `createGrid` with every Community\nmodule registered and the Quartz theme, whose CSS ships inside the script\nunder the theming API, so there is no separate stylesheet to add. The\nTanStack number is the whole Svelte adapter with its engine: no\nvirtualizer, filter UI or editors, because a headless library has none.\n\n## Measured performance\n\nA vendor's own benchmark is worth nothing unless you can rerun it, so the\nharness is in the repository and takes an adapter per grid. Adding another\ngrid is about forty lines in `examples/src/bench/adapters.ts` and nothing\nelse.\n\n```bash\n# headless, prints the table and writes tests/perf/.last-compare.json\nBENCH_REPEATS=7 BENCH_GRIDS=svgrid,aggrid,svar,tanstack pnpm bench:compare\n\n# move that run into docs/_data/competitors.json, then regenerate this page\nnode tools/record-benchmarks.mjs && node tools/sync-guide-facts.mjs\n\n# or drive it by hand\npnpm --filter @svgrid/grid-example-gallery dev # then open /bench.html\n```\n\nEach operation is timed until the grid's DOM reflects it, plus one frame:\n\n- **Mount** - fresh container, create the grid, until rows are painted.\n- **Sort** - change the sort on one column, until the new order is painted.\n Text sorts `name` (tens of thousands of distinct values), numeric sorts\n `amount`.\n- **Filter** - apply a `contains` filter to one column.\n- **Scroll** - 180 frames at 60 px/frame, sampling frame intervals.\n\n### Results\n\nThe table is generated from the ledger by `tools/sync-guide-facts.mjs`;\nthe versions and the date are part of it.\n\n<!-- bench:start -->\nMeasured 28 Oct 2026 with `pnpm bench:compare` (tests/perf/compare.spec.ts, recorded by tools/record-benchmarks.mjs): 100,000 rows x 9 columns; 1000 x 520 px container, 32 px rows, 140 px columns, each grid with its default theme; fastest of 3 samples per operation, each grid in a fresh page; Chromium through Playwright, headless; win32 10.0.26200 x64, Intel(R) Core(TM) Ultra 7 255U; a developer workstation, not a dedicated bench rig; tick: 1,000 of the rows replaced per tick with a new amount, the grid sorted by amount, on the update path each grid provides, timed until the changed cells paint; 180 ticks after 10 to warm up. Lower is better.\n\n| Operation | SvGrid | AG Grid Community | SVAR Svelte DataGrid | TanStack Table, minimal DOM |\n| --- | --- | --- | --- | --- |\n| Mount | 30.7 ms | 67 ms | 11.4 ms | 170.2 ms |\n| Sort, text column | 77.4 ms | 213.9 ms | 135.4 ms | 360.9 ms |\n| Sort, numeric column | 32.8 ms | 216.7 ms | 141.7 ms | 178.2 ms |\n| Filter, one column (indicative only, see below) | 34.5 ms | 148.8 ms | 26.1 ms | 51.1 ms |\n| Scroll, p95 frame | 34.3 ms | 34.5 ms | 34.6 ms | 34.3 ms |\n| Live tick, 1,000 rows replaced, sorted by that column, p95 | 48.6 ms | 259.2 ms | 152.2 ms | 563.4 ms |\n| Main-thread work per 60 px scroll step, mean | 3.5 ms | 4.5 ms | 3 ms | 2.9 ms |\n| Frames dropped while scrolling, of 180 | 0 | 0 | 0 | 0 |\n| Ticks over one 60 Hz frame, of 180 | 115 | 180 | 180 | 180 |\n| Rows kept in the DOM (virtualization on) | 36 | 36 | 19 | 27 |\n| JS heap the grid adds, mounted, after GC | 5.7 MB | 42.1 MB | 2.3 MB | 109.9 MB |\n| JS heap the grid adds, after sorting and filtering, after GC | 17.6 MB | 26.2 MB | 3.9 MB | 113.3 MB |\n| Elements created per scroll frame | 0 | 30.1 | 16.8 | 18.7 |\n| Elements re-inserted per scroll frame (rows recycled) | 37.3 | 35.8 | 0 | 0 |\n\nVersions: `@svgrid/grid` 3.0.25, `ag-grid-community` 36.1.0, `wx-svelte-grid` 2.7.0, `@tanstack/svelte-table` 9.2.4.\n<!-- bench:end -->\n\nRead that as: **SvGrid sorts and keeps a live sort current faster than the\nother three, scrolling is level across all four, and SVAR is ahead on mount\nand memory** - it paints its first rows in under half of SvGrid's time and\nholds the least heap. Every grid scrolls at the display rate here, so the\np95 frame reads the same for all four; the work row under it is what a\nslower machine or a 120 Hz display would feel. There TanStack and SVAR do\nthe least per scroll step, about half a millisecond under SvGrid, and AG\nGrid the most. The row varies from run to run, so read the order of the\nfirst three as close, not settled. SvGrid makes its row objects the first time a\nsort or filter reads every row rather than at mount, which is why its two\nheap rows differ. The TanStack renderer pays for building a row object per\nrecord before it can draw anything. Anyone claiming a single headline\nnumber for a grid comparison is selling something.\n\n**The heap rows are the JS heap after a forced garbage collection, less the\nsame page with the data built and no grid,** so the 100,000 rows themselves\ncount against no one. The two DOM rows are counted over 60 untimed scroll\nframes after the timed ones. \"Created\" is elements that did not exist\nbefore; \"re-inserted\" is elements the grid already had and moved, which is\nhow a keyed list recycles a row from one end of the window to the other.\nSvGrid creates none and moves the two rows that cross the edge each frame;\na grid that rewrote its rows in place would score 0 on both.\n\n**The filter row is not a like-for-like comparison and is shown only because\nhiding a number we win looks worse than explaining it.** The grids' single-\ncolumn filter APIs differ: the harness drives AG Grid's quick filter, which\nsearches every column and is strictly more work than the single-column\n`contains` SvGrid performs, while SVAR and TanStack run a predicate over the\none column. Do not read a ratio into that row.\n\n**The TanStack column is an engine plus the least DOM an app could draw.**\nTanStack Table renders nothing, so its adapter draws the row model through\na fixed-height windowed table with no virtualizer library, no filter UI and\nno editors. Its mount and sort numbers include creating the engine's row\nobjects for every record; the grids around it do the same work inside their\nown models.\n\nSorting is where SvGrid's distance comes from, and it is not subtle: SvGrid\nresolves each sort clause once and precomputes a key array, and for a\ncolumn whose distinct values are far fewer than its rows it collates the\ndistinct values once and sorts by rank. Mounting is the other direction -\nsee the accounting in [benchmarks](./benchmarks.md).\n\nAt 1,000,000 rows the harness runs one grid per invocation\n(`BENCH_ROWS=1000000 BENCH_GRIDS=<one> BENCH_TICKS=30`); TanStack's\nadapter does not finish at that size. Those runs are not in the table\nbecause only a run that measured every grid together is comparable.\n\nTwo methodology notes, both of which were bugs in this harness before they\nwere notes, and both of which changed the numbers by more than any code\nchange did:\n\n- **Each grid runs in a fresh page.** They used to run one after another in\n the same one, where the first warmed the JIT, fragmented the heap and\n left a hundred thousand rows of garbage for the next. Back-to-back runs\n on an idle machine disagreed with themselves by a factor of two.\n That order effect is gone, but this workstation still varies by about\n 2x between whole runs taken minutes apart, for every grid at once, so\n compare the columns of one run and not one run against another.\n- **Operations are timed until the DOM changes, not for a fixed two\n frames.** The old fixed wait put a two-frame floor under every number -\n an operation that recomputed nothing at all measured the floor, and a\n mount whose rows were already in the DOM measured pure waiting - which\n flattered whichever grid was slower.\n- **Each figure is the fastest of the samples, not the median.** Noise\n only ever adds time, so the minimum is the cleanest estimate of what the\n machine can do, and medians on this box drifted by a third between\n consecutive runs - more than most of the differences being measured.\n Applied identically to every grid.\n\nWhat this table deliberately does not include:\n\n- **Features the grids do not measure the same way.** Grouping, pivot,\n editing and export all differ enough in shape that a single timing would\n compare two different amounts of work. The filter row above is the one\n case we publish anyway, labelled, rather than quietly dropping a result\n that happens to favour us.\n- **Any grid we cannot license for this.** The harness is grid-agnostic on\n purpose, but some commercial grids licence their software in ways that\n restrict using it to produce competitive claims. Where that is the case\n we do not ship an adapter, and we would rather have an obviously\n incomplete table than a legally dubious one. Run your own shortlist.\n- **Your workload.** One dataset shape, one machine, one browser, default\n configuration for every grid. If a grid decision rides on this, clone the\n harness and put your own data in it.\n\n## Migrating from AG Grid\n\nThe most common starting point. See\n[Migrating from AG Grid](./migrating-from-ag-grid.md) for a side-by-side\ntranslation of column defs, features, filtering, editing, and the\nimperative API. The example that guide is built around runs here:\n\n<div data-docs-demo=\"139-migration-from-ag-grid\" data-height=\"560\"></div>\n\n## Migrating from TanStack Table\n\nThe map is one-to-one - SvGrid's headless core speaks the vocabulary\nTanStack Table v9 introduced. The big differences:\n\n- Replace `createTable(opts)` with `createSvGrid(opts)`. Same state\n machine, same features object.\n- Replace the row-model factories with the same names from\n `@svgrid/grid`.\n- The render layer changes - TanStack hands you `FlexRender` + the\n row model; SvGrid lets you keep that headless approach OR drop in\n the default `<SvGrid>` component.\n\nThe headless demo below drives a plain `<table>` from `createSvGrid`, the\nway a TanStack app would:\n\n<div data-docs-demo=\"186-headless-table\" data-height=\"520\"></div>\n\nAnd the same rows through the render component, which is the part TanStack\nleaves to you. The example runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Order = {\n id: number\n customer: string\n region: string\n status: string\n amount: number\n }\n\n const orders: Order[] = [\n { id: 1, customer: 'Ada Lovelace', region: 'EMEA', status: 'open', amount: 1240 },\n { id: 2, customer: 'Grace Hopper', region: 'AMER', status: 'shipped', amount: 860 },\n { id: 3, customer: 'Linus Torvalds', region: 'EMEA', status: 'closed', amount: 2210 },\n { id: 4, customer: 'Radia Perlman', region: 'APAC', status: 'open', amount: 430 },\n { id: 5, customer: 'Barbara Liskov', region: 'AMER', status: 'pending', amount: 1780 },\n ]\n\n let rows = $state<Order[]>(orders)\n\n const columns: GridColumns<Order> = [\n { field: 'customer', header: 'Customer', width: 180 },\n { field: 'region', header: 'Region', width: 110 },\n { field: 'status', header: 'Status', width: 110 },\n { field: 'amount', header: 'Amount', width: 120, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable selectable />\n```\n\n## Pricing\n\nSvGrid Community is MIT - free for commercial use, no attribution\nrequired at runtime. SvGrid Enterprise is a per-developer licence: $599\nper developer per year for the enterprise grid or $999 with the\nspreadsheet and Studio, both for unlimited applications; see\n<https://svgrid.com/pricing/> for what each tier adds.\n\nAG Grid Community is MIT. AG Grid Enterprise is priced per developer with\none year of updates; the figure ag-grid.com listed on the date we read it\nis in the facts block at the top of this page, and the\n[SvGrid vs AG Grid](https://svgrid.com/compare/ag-grid/) page carries the\nlink. Compare what each paid tier adds before comparing prices: most of\nwhat AG Grid sells as Enterprise is in SvGrid's free core.\n\nTanStack Table is MIT and sponsor-funded; tanstack.com offers paid\nEnterprise Support as private consulting.\n\n## Frequently asked questions\n\n### What is the best data grid for Svelte 5?\n\nFor a Svelte-5-native grid with a batteries-included render component, SvGrid is\nbuilt around runes and snippets. TanStack Table is a strong headless-only\nchoice if you want to build the DOM layer yourself across frameworks, and its\nv9 Svelte adapter targets Svelte 5. AG Grid is the most feature-complete but\nlives in React, Angular and Vue first and reaches Svelte through\ncommunity-built integrations.\n\n### Is SvGrid a good AG Grid alternative?\n\nYes, for Svelte projects. SvGrid is a fraction of AG Grid Community's\nmeasured size on a Svelte page (the facts block has both numbers and the\nmethod), is MIT-licensed for commercial use, and offers `@svgrid/enterprise`\nfor export, print, import and pivot at a per-developer price. On the\nbenchmark above it sorts a hundred-thousand-row column faster on both\ncolumn types and scrolls level; AG Grid mounts quicker. It does not match\nevery AG Grid Enterprise feature - see the\n[missing-features list](./missing-features.md) for the gaps.\n\n### SvGrid vs TanStack Table - which should I pick?\n\nPick SvGrid if you want virtualization, Excel-style filters, selection, and\ninline editing working out of the box on Svelte 5. Pick TanStack Table if you\nwant a framework-agnostic headless engine and are happy to build the rendering,\nvirtualization, and editing UI yourself. Both are MIT-licensed.\n\n### Does the TanStack Svelte adapter support Svelte 5?\n\nYes. The `@tanstack/svelte-table` v9 line declares a Svelte 5 peer range\nand passes data through getters rather than stores; the facts block above\nshows the version read from npm. The earlier `createSvelteTable` constructor\nis gone in v9.\n\n### Does AG Grid have an MCP server like SvGrid?\n\nYes. `ag-mcp` (MIT) gives an assistant AG Grid's docs, examples and\nversion-specific upgrade prompts. `@svgrid/mcp` does the same for SvGrid\nand adds a code checker that validates props and imports against the real\nAPI surface, a scaffold tool and a preview. Neither is an in-grid feature;\nSvGrid's natural-language filter, smart fill and summarise helpers are, and\nthey ship free in `@svgrid/grid` with a model provider you register.\n\n### How big is the SvGrid bundle?\n\nThe facts block at the top of this page has the measured gzipped size of the\nfull `<SvGrid>` render component and its stylesheet, read from\n`docs/_data/svgrid-size.json`, which `pnpm size:json` writes. The\n[bundle size reference](./bundle-size.md) breaks it down by entry:\nthe headless core on its own, the core subpath, the charts, and the lazy\nchunks that load on demand.\n\n## See also\n\n- [Why headless?](../why-headless.md) - the design decision behind\n SvGrid's two-layer architecture.\n- [Migrating from AG Grid](./migrating-from-ag-grid.md) - the\n practical recipe.\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - the\n one-to-one API map.\n- [Benchmarks](./benchmarks.md) - how the numbers are produced and what\n they leave out.\n- [Enterprise feature pack](../enterprise/README.md) - what SvGrid charges for.\n- [Missing features](./missing-features.md) - the honest gap list\n versus AG Grid Enterprise.\n- [All comparisons](https://svgrid.com/compare/) - one page per grid, with\n dated sources for every claim.\n"
5651
5651
  },
5652
5652
  {
5653
5653
  "slug": "help/conditional-form-schema",
@@ -6060,14 +6060,14 @@ export const docs = [
6060
6060
  "path": "docs/help/mcp-server.md",
6061
6061
  "title": "MCP server",
6062
6062
  "section": "Help",
6063
- "markdown": "# MCP server\n\nThe SvGrid MCP server does two things for an AI client (Claude Code,\nClaude Desktop, Cursor, Zed, Codex, custom agents): it answers questions\nabout SvGrid from the files this repository ships, and it **checks the\ncode the model writes** against the real exported surface of the version\nyou have installed. No API key either way.\n\n![An AI coding agent calls the @svgrid/mcp server over the Model Context Protocol, which runs grid tools and returns structured JSON results back to the agent.](/docs-media/grid-mcp.svg)\n\n> **What is MCP?** Model Context Protocol is the open standard\n> ([modelcontextprotocol.io](https://modelcontextprotocol.io)) for\n> exposing tools to LLM clients. SvGrid ships an MCP server so the\n> model your team already uses can \"see\" the grid without you having\n> to copy-paste docs into prompts.\n\nIt runs either way you like: as a local process\n([`@svgrid/mcp`](https://www.npmjs.com/package/@svgrid/mcp) on npm), or as\na hosted endpoint at `https://mcp.svgrid.com/mcp` that needs no install at\nall. Both are listed in the official MCP registry as `com.svgrid/svgrid`.\n\n## Two ways to connect\n\n**Hosted (nothing to install).** Point any MCP client at the URL:\n\n```\nhttps://mcp.svgrid.com/mcp\n```\n\n```bash\nclaude mcp add --transport http svgrid https://mcp.svgrid.com/mcp\n```\n\nOne click:\n[Add to Cursor](https://cursor.com/en/install-mcp?name=svgrid&config=eyJ1cmwiOiJodHRwczovL21jcC5zdmdyaWQuY29tL21jcCJ9)\n· [Add to VS Code](https://insiders.vscode.dev/redirect/mcp/install?name=svgrid&config=%7B%22name%22%3A%22svgrid%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fmcp.svgrid.com%2Fmcp%22%7D)\n\nIt carries seven tools - `search`, `fetch`, `list_examples`,\n`get_example_source`, `get_api_reference`, `check_svgrid_code` and\n`svgrid_preview` - and needs no Node, no config file, and no key.\n\nIn an AI app builder, add the same URL as a custom MCP server with no\nauthentication: [Bolt.new](./bolt-new.md#3-connect-the-mcp-server) and\n[Lovable](./lovable.md#the-mcp-server-as-a-chat-connector) both take it.\n\n**Local (`npx @svgrid/mcp`).** Everything the hosted server has, plus\nthe Studio tools, the docs and demos as MCP resources, ready-made\nprompts, and a `svgrid_check_code` that additionally *compiles* the file\nwith the Svelte compiler rather than checking it statically. Use it when\nyou want the compile pass, the Studio tools, or no third-party endpoint\nin the loop.\n\n| | Hosted | Local |\n| --- | --- | --- |\n| Setup | a URL | `npx @svgrid/mcp` |\n| Needs Node | no | yes |\n| Code check compiles | no, static checks only | yes |\n| Resources and prompts | yes | yes |\n| Visual preview | yes | yes |\n| `studio_*` tools | no | yes, opt-in |\n| Works offline | no | yes |\n\n## Install\n\nNo install step is required - `npx` fetches it on demand:\n\n```bash\n# One-shot, from any project\nnpx -y @svgrid/mcp\n```\n\nTo pin it as a dev dependency instead:\n\n```bash\npnpm add -D @svgrid/mcp\n```\n\nThe local server is a Node binary (`svgrid-mcp`) that speaks MCP over\nstdio. There is no daemon to maintain. The hosted server needs no install\nstep at all - skip to [Wire it into your AI client](#wire-it-into-your-ai-client)\nand use the URL.\n\n## Wire it into your AI client\n\nEvery snippet below shows the local (stdio) form and the hosted (HTTP)\nform. Pick one - running both connects the same tools twice.\n\n### Claude Code\n\nThe plugin is the shortest path: it installs the\n[Agent Skill](./skill.md) and the hosted server together, so the model\ngets the house style *and* the checker in one step.\n\n```\n/plugin marketplace add sv-grid/sv-grid\n/plugin install svgrid@svgrid\n```\n\nOr add the server on its own:\n\n```bash\n# hosted, nothing to install\nclaude mcp add --transport http svgrid https://mcp.svgrid.com/mcp\n\n# or local\nclaude mcp add svgrid -- npx -y @svgrid/mcp\n```\n\nThen run `/mcp` in a session and you will see `svgrid` listed.\n\nTo share the server with your team, add `--scope project`. That writes\na `.mcp.json` at the repository root which you can commit, so everyone\nwho clones the repo gets the same tooling with no per-machine setup:\n\n```bash\nclaude mcp add svgrid --scope project -- npx -y @svgrid/mcp\n```\n\n### Claude Desktop\n\nEdit `~/Library/Application Support/Claude/claude_desktop_config.json`\n(macOS) or `%APPDATA%\\Claude\\claude_desktop_config.json` (Windows):\n\n```json\n{\n \"mcpServers\": {\n \"svgrid\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"]\n }\n }\n}\n```\n\nOr the hosted server, with nothing to install:\n\n```json\n{\n \"mcpServers\": {\n \"svgrid\": {\n \"type\": \"http\",\n \"url\": \"https://mcp.svgrid.com/mcp\"\n }\n }\n}\n```\n\nRestart Claude Desktop, then ask *\"using svgrid, build me a grid that\ngroups by department\"* to confirm the tools are exposed.\n\n### Cursor\n\nOne click:\n[Add to Cursor](https://cursor.com/en/install-mcp?name=svgrid&config=eyJ1cmwiOiJodHRwczovL21jcC5zdmdyaWQuY29tL21jcCJ9)\n\nOr by hand, in `Settings -> MCP -> Add new MCP server`:\n\n```json\n{ \"url\": \"https://mcp.svgrid.com/mcp\" }\n```\n\n```json\n{ \"command\": \"npx\", \"args\": [\"-y\", \"@svgrid/mcp\"] }\n```\n\n### Zed\n\n`~/.config/zed/settings.json`:\n\n```json\n{\n \"context_servers\": {\n \"svgrid\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"],\n \"env\": {}\n }\n }\n}\n```\n\n### VS Code\n\nOne click:\n[Add to VS Code](https://insiders.vscode.dev/redirect/mcp/install?name=svgrid&config=%7B%22name%22%3A%22svgrid%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fmcp.svgrid.com%2Fmcp%22%7D)\n\nOr create `.vscode/mcp.json` in the workspace. Note that VS Code uses\n`servers` rather than the `mcpServers` wrapper:\n\n```json\n{\n \"servers\": {\n \"svgrid\": {\n \"type\": \"http\",\n \"url\": \"https://mcp.svgrid.com/mcp\"\n }\n }\n}\n```\n\n```json\n{\n \"servers\": {\n \"svgrid\": {\n \"type\": \"stdio\",\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"]\n }\n }\n}\n```\n\n### Custom agents (OpenAI Agents SDK, Anthropic SDK, LangChain)\n\nPoint your client's MCP stdio transport at `npx -y @svgrid/mcp`, or its\nstreamable-HTTP transport at `https://mcp.svgrid.com/mcp`. Any client that\nspeaks either transport works.\n\nThe hosted server names its two retrieval tools `search` and `fetch`\nexactly, which is what a connector needs to index it.\n\n## Tools exposed\n\nFive, by default. `tools/list` is sent on **every** request, so the tool\nsurface is pure overhead on every turn - and this server used to spend\n~4,710 tokens of it on 36 tools, 79% of them Studio tools that most\nsessions never call once.\n\n| Tool | What it does |\n| ---- | ------------ |\n| `svgrid_search` | Search the docs, every demo and the API surface **in one call**. No arguments returns an index. |\n| `svgrid_get` | Read one thing in full: a doc slug, a demo id, or `api`. |\n| `svgrid_check_code` | Verify code against the real exported surface before the user sees it. |\n| `svgrid_preview` | Render a **live, interactive grid** in the conversation from columns + rows. |\n| `svgrid_scaffold` | Turn a Drizzle schema, sample rows or an EntitySchema into runnable SvelteKit files. |\n\nStudio adds four more, and only when you ask for them - set\n`SVGRID_MCP_STUDIO=1` or a licence key in the server's env. They need a\nlicence to be useful, so they are not charged to everyone else's context\nwindow.\n\n`svgrid_scaffold` and the Studio tools run on `@svgrid/enterprise`, which the\nserver loads only when it is installed beside it. `npx -y @svgrid/mcp` leaves\nit out so the server starts fast. To scaffold, install both in the project and\nrun the local server:\n\n```bash\nnpm i -D @svgrid/mcp @svgrid/enterprise\nclaude mcp add svgrid -- npx svgrid-mcp\n```\n\nWithout `-y`, `npx` runs the installed copy instead of downloading one, so\nthere is no install on the first connect.\n\nThe **hosted** server carries `search`, `fetch`, `list_examples`,\n`get_example_source`, `get_api_reference`, `check_svgrid_code` and\n`svgrid_preview`, plus the same resources and prompts. The first two are named\nthat way deliberately: those exact names are what a one-click connector needs to\nindex a remote server. It also answers to `svgrid_search`, `svgrid_get` and\n`svgrid_check_code`, so a model that learned the npm package's names is never\ntold \"unknown tool\".\n\nWhat it does **not** have is the compile pass (the Svelte compiler cannot run in\na Worker), the Studio tools (they need a filesystem), and the offline guarantee\n- it is a remote server, so by definition your query reaches it. Use the local\none when any of those matter.\n\n### Verification\n\n#### `svgrid_check_code`\n\nChecks a file **against the version you have installed** and returns\nline-numbered diagnostics with the exact replacement for each: unknown\n`<SvGrid>` props, wrong `ColumnDef` keys, api methods that do not\nexist, imports that will not resolve, and Svelte 5 runes mistakes. It\nalso compiles the component, so a syntax error surfaces here rather\nthan in the user's terminal.\n\nThis is the tool that makes the rest worth having. Retrieval alone\nstill lets a model write a confident, wrong grid; nothing else in the\nserver stops it. Run it on every file you write, fix what it reports,\nand run it again.\n\n### Finding and reading\n\n#### `svgrid_search`\n\nOne call across all three corpora, because \"how do I pin a column\" does\nnot announce whether it is answered by a doc page, a demo, or an API\nname:\n\n```json\n{ \"query\": \"pin a column\" }\n```\n\nReturns ranked doc hits with excerpts, matching demos, and matching API\nnames - each with the reference you pass to `svgrid_get`. Narrow with\n`kind` (`docs` | `examples` | `api`), trade tokens for completeness\nwith `detail` (`concise` | `full`), and cap with `limit` (max 50).\n\nCall it with **no arguments** for an index of doc sections, demo\ncategories and API groups - the cheapest way to orient before searching.\n\n#### `svgrid_get`\n\n```json\n{ \"ref\": \"help/columns/column-definitions\" }\n{ \"ref\": \"11-stock-market\" }\n{ \"ref\": \"api\" }\n```\n\nThe kind is inferred from the reference; pass `kind` to force it. A\nreference that does not resolve comes back with near matches rather\nthan a bare \"not found\".\n\n### It fixes what it finds\n\n`svgrid_check_code` does not stop at telling you what is wrong. Where the\ncorrection is exact it returns the corrected file too:\n\n```json\n{\n \"diagnostics\": [ ... ],\n \"applied\": [\"rowData -> data (line 7)\", \"accessorKey -> field (line 3)\"],\n \"fixed\": \"<the corrected source>\"\n}\n```\n\nBoth fields are omitted when there is nothing mechanically fixable, so a clean\ncheck stays small.\n\nThe rules are deliberately narrow, because this tool's whole worth is that it\nnever cries wolf - and rewriting raises the stakes from wasting a turn to\ncorrupting working code. Only exact renames are applied (a known rename, or a\nclose-enough spelling), only on word boundaries, and only on the line the\ndiagnostic reported. Advice that is ambiguous - \"this has no equivalent, remove\nit\" - stays advice. A test asserts that running auto-fix across every demo\nin this repository changes nothing at all.\n\n### It knows which version you have\n\nThe result carries the version it checked against:\n\n```json\n{ \"version\": { \"corpus\": \"3.0.0\", \"installed\": \"2.6.8\", \"warning\": \"...\" } }\n```\n\nThe server reads the `@svgrid/grid` in your project and says plainly when it\ndisagrees with the corpus it ships, so a model is told to hedge instead of\nconfidently citing an API you do not have. A server that answers from a remote\nbackend cannot do this - it has no idea what is in your `node_modules`.\n\n### Nothing leaves your machine\n\nThis server makes **no network calls**. Not for docs, not for the API surface,\nnot for verification - the whole corpus (every demo, every doc page, the full\nexported API) ships inside the package, which is why it is megabytes rather\nthan kilobytes. There is no API key, no licence check on the wire, and no telemetry.\n\nIt works on a plane, behind a corporate proxy, and inside an air-gapped\nnetwork. `tools/mcp-offline.test.ts` enforces it: the server is driven through\na full session with every network primitive booby-trapped, and the test fails if\nanything is even attempted.\n\nThe one honest exception: the **preview** loads the grid from a CDN to draw it.\nThat happens in your client's sandboxed iframe, never in this server, and only\nwhen you actually render a preview.\n\n### Seeing the grid, not just reading about it\n\n`svgrid_preview` renders a **real, interactive SvGrid inside the\nconversation** - sortable, filterable, scrollable. Not a screenshot and not a\nmock table: the same `<sv-grid>` custom element a page would use, loaded from\nthe CDN, with the columns and rows you passed.\n\n```json\n{\n \"title\": \"Team roster\",\n \"columns\": [{ \"field\": \"name\", \"header\": \"Name\" }, { \"field\": \"amount\", \"header\": \"Amount\" }],\n \"data\": [{ \"name\": \"Ada\", \"amount\": 20000 }],\n \"sortable\": true\n}\n```\n\nIt uses [MCP Apps](https://blog.modelcontextprotocol.io/posts/2026-01-26-mcp-apps/),\nthe official UI extension: the tool points at a `ui://` resource through\n`_meta`, the client loads that HTML in a sandboxed iframe, and the tool's\n`structuredContent` arrives over a postMessage bridge. Supported in Claude\n(web and desktop), VS Code and ChatGPT.\n\nClients without UI support are not left out - they ignore the `_meta` and get\na text summary of the same grid, so it is always safe to call. That is the\nextension's own rule, and it is why the text half describes the grid rather\nthan pointing at a picture the reader cannot see.\n\nPass `demo` instead of `columns` / `data` and you get the demo's source: a\ndemo is a Svelte component, not data, so there is nothing to hand the element\nand rendering an empty grid would just look broken.\n\n### Resources and prompts\n\nThe docs and demos are also exposed as **MCP resources**, so a user can\nattach a page or a demo directly instead of hoping the model thinks to\nsearch for it:\n\n```\nsvgrid://doc/help/columns/column-definitions\nsvgrid://example/11-stock-market\n```\n\nAnd three **prompts** ship ready to run: `build_grid`, `explain_api`\nand `review_grid_code`. Each is written to make the model use this\nserver rather than recall SvGrid from training data, and each ends at\n`svgrid_check_code`.\n\n### SvGrid Studio (commercial)\n\nOff unless enabled. Set `SVGRID_MCP_STUDIO=1`, or a valid\n`SVGRID_LICENSE_KEY`, in the server's env, and run the server from a project\nthat has `@svgrid/enterprise` installed beside it\n(`npm i -D @svgrid/mcp @svgrid/enterprise`, then `npx svgrid-mcp`). Without it\nthese tools answer with those steps instead of running.\n\n#### `svgrid_scaffold`\n\nPoint it at a Drizzle schema, sample JSON rows, or an `EntitySchema`\nyou already have, and it infers the schema and generates the `$lib`\nschema module, a `+server.ts` API route and a `+page.svelte` with\n`SvGrid` + `SvGridEditPanel` - in one call. Pass `schemaOnly: true` to\nstop after inference. Generated bodies carry `svgrid:managed` markers,\nso regenerating preserves your edits outside them.\n\n#### The `studio_*` tools\n\nFour tools drive the same validated project model the visual designer\nuses:\n\n| Tool | What it does |\n| ---- | ------------ |\n| `studio_project` | `new`, `load`, `describe`, `config`, `capabilities` |\n| `studio_apply` | A **batch** of model changes - entities, screens, blocks, components - applied in order |\n| `studio_configure` | Theme, auth, access, tenancy, data layer, jobs, deploy target, layouts, in one call |\n| `studio_build` | `validate`, then `generate` the full runnable SvelteKit app |\n\nCall `studio_project` with `action: \"capabilities\"` first: it reports\nexactly what the installed version supports, so the agent does not have\nto guess block kinds or component keys.\n\n`studio_apply` takes a batch on purpose. Building a five-screen app was\ntwenty-odd round trips when every block was its own tool; now a whole\nscreen is one call, and a failure reports which op failed and what\nalready applied.\n\n### The pre-3.0 names\n\nEvery tool name from before 3.0 still answers - `search_docs`,\n`get_doc`, `list_examples`, `get_example_source`, `list_docs`,\n`get_api_reference`, `check_svgrid_code`, `introspect_source`,\n`scaffold_entity` and all 27 individual `studio_*` tools. They are not\n**listed**, because listing is what costs context and answering a name\nyou did not advertise costs nothing. Saved prompts and scripts keep\nworking; nobody pays for the compatibility.\n\nTheir JSON response shape follows 3.0, not 2.x. A model reads either\nwithout trouble; a script that parsed the old shape by hand needs a look.\n\n## Licensing\n\nThe documentation and example tools are free. The Studio code\ngenerators are part of the commercial offering. You can evaluate them\nwithout a key: they run, but prepend a notice comment\nto generated files. To license them, set\nthe key in your MCP client's server config:\n\n```json\n{\n \"mcpServers\": {\n \"svgrid\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"],\n \"env\": { \"SVGRID_LICENSE_KEY\": \"YOUR-LICENSE-KEY\" }\n }\n }\n}\n```\n\n## Enterprise features and the trial\n\nMost of SvGrid is the MIT `@svgrid/grid` package. Export to Excel and PDF,\npivot tables, the Kanban board, the scheduler, the spreadsheet and the\nserver-side row model are in `@svgrid/enterprise`, which is commercial.\n\nThe server tells your assistant which is which. In `svgrid_search` results a\npaid doc or demo carries `\"enterprise\": true`, and when a question is about a\npaid feature the answer says so once, with the trial link.\n`svgrid_get` puts the same line at the top of a paid doc or demo, and\n`svgrid_check_code` adds it when the fix is to install the package. Nothing\nis sent anywhere to do this; it is part of the answer text.\n\n`@svgrid/enterprise` runs without a license key and shows a watermark until\none is set, so you can try a feature before deciding. The\n[30-day trial](https://svgrid.com/pricing/?trial=mcp-docs) adds a download of\nevery demo and 30 days of email support.\n\n## Verifying it works\n\nAsk your model: *\"What MCP tools do you have from svgrid?\"* The local\nserver lists five, including `svgrid_check_code` (four Studio tools more when\nenabled); the hosted server lists seven.\n\nThen give it something to catch:\n\n> Check this with svgrid: `<SvGrid rowData={rows} {columns} />`\n\nIt should come back with `rowData` is not a SvGrid prop, use `data`. If\nit explains the code instead of calling a tool, the server is not\nconnected.\n\nIf nothing is listed at all, check your client's MCP log. For the local\nserver the usual cause is `npx` not being on PATH - use the absolute path\nto the binary. For the hosted one, confirm the endpoint answers:\n\n```bash\ncurl https://mcp.svgrid.com/health\n```\n\n## Security model\n\nThe two ways of running it differ here, so pick deliberately.\n\n**Local (`npx @svgrid/mcp`)**\n\n- Runs on your machine over stdio. No telemetry, no outbound network\n calls, no API key.\n- Serves a documentation and example corpus bundled into the package at\n build time, so answers are pinned to the version you installed.\n- `svgrid_check_code` reads the source you pass it in-process and never\n sends it anywhere.\n\n**Hosted (`https://mcp.svgrid.com/mcp`)**\n\n- Your client sends tool arguments to a Cloudflare Worker we operate.\n Anything you pass to a tool leaves your machine, and for\n `check_svgrid_code` that means **the source you ask it to check**.\n- Each call is recorded: tool name, duration, whether it succeeded, the\n query or id it was given, how many pages matched, and whether the answer\n mentioned the Enterprise trial. For `check_svgrid_code` only the **byte\n length** of the submitted source and the names of the rules that fired\n are recorded, never the source itself. The records stay in Cloudflare\n (Workers logs, and an Analytics Engine dataset kept three months) and are\n read in aggregate to see what people ask for.\n- No account, no key, no cookies. Requests are not tied to a user.\n- If your code cannot leave the building, use the local server. That is\n what it is for.\n\n**Both**\n\n- The Studio tools return generated files as data. Writing them to disk\n is your client's decision, not the server's.\n- See [security](./security.md) for the general supply-chain posture.\n\n## Building your own MCP integrations\n\nThe same docs manifest, JSON Schemas (column definitions, the grid's\noptions, export options, and the chart spec at `schemas/chart-spec.json`,\nwhich `svgrid_get` also serves for \"chart spec schema\"), and `llms.txt`\nfiles are also served directly from the docs site:\n\n```ts\nconst docs = await fetch('https://svgrid.com/docs.json').then((r) => r.json())\nconst schemas = await fetch('https://svgrid.com/schemas/index.json').then((r) => r.json())\nconst llms = await fetch('https://svgrid.com/llms-full.txt').then((r) => r.text())\n```\n\nIf you do not want to run the MCP server, building these into your\nagent's system prompt gives most of the same grounding.\n\n## Without the MCP server\n\nAn agent that can only fetch pages gets the same material over plain HTTP,\nin pieces small enough for one fetch (fetch tools cut a response near\n100 KB):\n\n| URL | What it holds |\n|---|---|\n| `/llms.txt` | The docs index, with links to everything below. `/llms/<product>.txt` (grid, ui, enterprise, sheet, base, studio, compare) is one product's part of it |\n| `/docs/<slug>.md` | One docs page as markdown |\n| `/demos.md`, `/demos/<id>.md` | Every live demo, and one demo with what it shows and its full source |\n| `/api.json` | Every export of `@svgrid/grid` and `@svgrid/enterprise` by module, with the member names of each type |\n| `/api/<name>.json` | One symbol's doc comment and signature, by its lowercased name: `/api/svgridapi.json`, `/api/createreview.json` |\n| `/.well-known/mcp.json` | Where this server is: the remote endpoint, the npm package, the setup commands |\n\n```ts\nconst index = await fetch('https://svgrid.com/api.json').then((r) => r.json())\nconst api = await fetch('https://svgrid.com/api/svgridapi.json').then((r) => r.json())\nconst demo = await fetch('https://svgrid.com/demos/702-change-review.md').then((r) => r.text())\n```\n\nIn a browser with WebMCP (`document.modelContext`), svgrid.com registers\nfour tools for an agent driving the page: `svgrid_search`, `svgrid_get_doc`,\n`svgrid_get_demo` and `svgrid_lookup_api`. They answer from the files\nabove.\n\n## Frequently asked questions\n\n### What is the SvGrid MCP server?\n\nA Model Context Protocol server that lets AI clients (Claude Code, Claude\nDesktop, Cursor, Zed, custom agents) query SvGrid's documentation, read real\ndemo source, scaffold SvelteKit CRUD apps, and check the code they write\nagainst the real API - grounded in the files the package ships, so the model\nanswers from current facts instead of guessing.\n\n### Do I need an API key to run it?\n\nNo, on either transport. A `SVGRID_LICENSE_KEY` is optional and only affects\nthe commercial Studio code generators.\n\n### Local or hosted - which should I use?\n\nHosted (`https://mcp.svgrid.com/mcp`) if you want it working in one paste,\nwith no Node and no config file. Local (`npx @svgrid/mcp`) if you want the\nSvelte compiler pass in `svgrid_check_code`, the Studio tools, offline\nuse, or your source never leaving the machine. See\n[Security model](#security-model) for what each one sends.\n\n### How does it help AI assistants write better SvGrid code?\n\nTwo ways. It exposes the docs, example sources, and API reference as tools,\nso the assistant retrieves version-pinned answers instead of hallucinating an\nAPI from training data. And `svgrid_check_code` closes the loop: the model\nruns what it wrote past the real exported surface and gets told exactly what\nis wrong before you see it. That matters most for Svelte 5, where models\nroutinely mix in outdated Svelte 4 syntax.\n\n### Does it work with a version of SvGrid I have not upgraded to?\n\nThe local server answers for the version bundled in the `@svgrid/mcp` release\nyou install, so pin it alongside `@svgrid/grid` if you are behind. The hosted\nserver always tracks the current release. Either way, every\n`svgrid_check_code` result names the version it checked against in its\n`checkedAgainst` field.\n\n## See also\n\n- [Agent Skill](./skill.md) - the always-on house-style layer; the Claude Code plugin installs it together with this server\n- [LLM grounding](./llm-grounding.md) - the same files used by the MCP server, but documented for direct LLM consumption\n- [Agents](./agents.md) - how to build an AI agent that drives the live grid\n- [AI assistant](./ai.md) - the in-grid AI features (filter / smart-fill / classify / summarise), free in @svgrid/grid\n"
6063
+ "markdown": "# MCP server\n\nThe SvGrid MCP server does two things for an AI client (Claude Code,\nClaude Desktop, Cursor, Zed, Codex, custom agents): it answers questions\nabout SvGrid from the files this repository ships, and it **checks the\ncode the model writes** against the real exported surface of the version\nyou have installed. No API key either way.\n\n![An AI coding agent calls the @svgrid/mcp server over the Model Context Protocol, which runs grid tools and returns structured JSON results back to the agent.](/docs-media/grid-mcp.svg)\n\n> **What is MCP?** Model Context Protocol is the open standard\n> ([modelcontextprotocol.io](https://modelcontextprotocol.io)) for\n> exposing tools to LLM clients. SvGrid ships an MCP server so the\n> model your team already uses can \"see\" the grid without you having\n> to copy-paste docs into prompts.\n\nIt runs either way you like: as a local process\n([`@svgrid/mcp`](https://www.npmjs.com/package/@svgrid/mcp) on npm), or as\na hosted endpoint at `https://mcp.svgrid.com/mcp` that needs no install at\nall. Both are listed in the official MCP registry as `com.svgrid/svgrid`.\n\n## Two ways to connect\n\n**Hosted (nothing to install).** Point any MCP client at the URL:\n\n```\nhttps://mcp.svgrid.com/mcp\n```\n\n```bash\nclaude mcp add --transport http svgrid https://mcp.svgrid.com/mcp\n```\n\nOne click:\n[Add to Cursor](https://cursor.com/en/install-mcp?name=svgrid&config=eyJ1cmwiOiJodHRwczovL21jcC5zdmdyaWQuY29tL21jcCJ9)\n· [Add to VS Code](https://insiders.vscode.dev/redirect/mcp/install?name=svgrid&config=%7B%22name%22%3A%22svgrid%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fmcp.svgrid.com%2Fmcp%22%7D)\n\nIt carries seven tools - `search`, `fetch`, `list_examples`,\n`get_example_source`, `get_api_reference`, `check_svgrid_code` and\n`svgrid_preview` - and needs no Node, no config file, and no key.\n\nIn an AI app builder, add the same URL as a custom MCP server with no\nauthentication: [Bolt.new](./bolt-new.md#3-connect-the-mcp-server) and\n[Lovable](./lovable.md#the-mcp-server-as-a-chat-connector) both take it.\n\n**Local (`npx @svgrid/mcp`).** Everything the hosted server has, plus\nthe Studio tools, the docs and demos as MCP resources, ready-made\nprompts, and a `svgrid_check_code` that additionally *compiles* the file\nwith the Svelte compiler rather than checking it statically. Use it when\nyou want the compile pass, the Studio tools, or no third-party endpoint\nin the loop.\n\n| | Hosted | Local |\n| --- | --- | --- |\n| Setup | a URL | `npx @svgrid/mcp` |\n| Needs Node | no | yes |\n| Code check compiles | no, static checks only | yes, with your project's Svelte |\n| Resources and prompts | yes | yes |\n| Visual preview | yes | yes |\n| `studio_*` tools | no | yes, opt-in |\n| Works offline | no | yes |\n\n## Install\n\nNo install step is required - `npx` fetches it on demand:\n\n```bash\n# One-shot, from any project\nnpx -y @svgrid/mcp\n```\n\nTo pin it as a dev dependency instead:\n\n```bash\npnpm add -D @svgrid/mcp\n```\n\nThe local server is a Node binary (`svgrid-mcp`) that speaks MCP over\nstdio. There is no daemon to maintain. The hosted server needs no install\nstep at all - skip to [Wire it into your AI client](#wire-it-into-your-ai-client)\nand use the URL.\n\n## Wire it into your AI client\n\nEvery snippet below shows the local (stdio) form and the hosted (HTTP)\nform. Pick one - running both connects the same tools twice.\n\n### Claude Code\n\nThe plugin is the shortest path: it installs the\n[Agent Skill](./skill.md) and the hosted server together, so the model\ngets the house style *and* the checker in one step.\n\n```\n/plugin marketplace add sv-grid/sv-grid\n/plugin install svgrid@svgrid\n```\n\nOr add the server on its own:\n\n```bash\n# hosted, nothing to install\nclaude mcp add --transport http svgrid https://mcp.svgrid.com/mcp\n\n# or local\nclaude mcp add svgrid -- npx -y @svgrid/mcp\n```\n\nThen run `/mcp` in a session and you will see `svgrid` listed.\n\nTo share the server with your team, add `--scope project`. That writes\na `.mcp.json` at the repository root which you can commit, so everyone\nwho clones the repo gets the same tooling with no per-machine setup:\n\n```bash\nclaude mcp add svgrid --scope project -- npx -y @svgrid/mcp\n```\n\n### Claude Desktop\n\nEdit `~/Library/Application Support/Claude/claude_desktop_config.json`\n(macOS) or `%APPDATA%\\Claude\\claude_desktop_config.json` (Windows):\n\n```json\n{\n \"mcpServers\": {\n \"svgrid\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"]\n }\n }\n}\n```\n\nOr the hosted server, with nothing to install:\n\n```json\n{\n \"mcpServers\": {\n \"svgrid\": {\n \"type\": \"http\",\n \"url\": \"https://mcp.svgrid.com/mcp\"\n }\n }\n}\n```\n\nRestart Claude Desktop, then ask *\"using svgrid, build me a grid that\ngroups by department\"* to confirm the tools are exposed.\n\n### Cursor\n\nOne click:\n[Add to Cursor](https://cursor.com/en/install-mcp?name=svgrid&config=eyJ1cmwiOiJodHRwczovL21jcC5zdmdyaWQuY29tL21jcCJ9)\n\nOr by hand, in `Settings -> MCP -> Add new MCP server`:\n\n```json\n{ \"url\": \"https://mcp.svgrid.com/mcp\" }\n```\n\n```json\n{ \"command\": \"npx\", \"args\": [\"-y\", \"@svgrid/mcp\"] }\n```\n\n### Zed\n\n`~/.config/zed/settings.json`:\n\n```json\n{\n \"context_servers\": {\n \"svgrid\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"],\n \"env\": {}\n }\n }\n}\n```\n\n### VS Code\n\nOne click:\n[Add to VS Code](https://insiders.vscode.dev/redirect/mcp/install?name=svgrid&config=%7B%22name%22%3A%22svgrid%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fmcp.svgrid.com%2Fmcp%22%7D)\n\nOr create `.vscode/mcp.json` in the workspace. Note that VS Code uses\n`servers` rather than the `mcpServers` wrapper:\n\n```json\n{\n \"servers\": {\n \"svgrid\": {\n \"type\": \"http\",\n \"url\": \"https://mcp.svgrid.com/mcp\"\n }\n }\n}\n```\n\n```json\n{\n \"servers\": {\n \"svgrid\": {\n \"type\": \"stdio\",\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"]\n }\n }\n}\n```\n\n### Custom agents (OpenAI Agents SDK, Anthropic SDK, LangChain)\n\nPoint your client's MCP stdio transport at `npx -y @svgrid/mcp`, or its\nstreamable-HTTP transport at `https://mcp.svgrid.com/mcp`. Any client that\nspeaks either transport works.\n\nThe hosted server names its two retrieval tools `search` and `fetch`\nexactly, which is what a connector needs to index it.\n\n## Tools exposed\n\nFive, by default. `tools/list` is sent on **every** request, so the tool\nsurface is pure overhead on every turn - and this server used to spend\n~4,710 tokens of it on 36 tools, 79% of them Studio tools that most\nsessions never call once.\n\n| Tool | What it does |\n| ---- | ------------ |\n| `svgrid_search` | Search the docs, every demo and the API surface **in one call**. No arguments returns an index. |\n| `svgrid_get` | Read one thing in full: a doc slug, a demo id, or `api`. |\n| `svgrid_check_code` | Verify code against the real exported surface before the user sees it. |\n| `svgrid_preview` | Render a **live, interactive grid** in the conversation from columns + rows. |\n| `svgrid_scaffold` | Turn a Drizzle schema, sample rows or an EntitySchema into runnable SvelteKit files. |\n\nStudio adds four more, and only when you ask for them - set\n`SVGRID_MCP_STUDIO=1` or a licence key in the server's env. They need a\nlicence to be useful, so they are not charged to everyone else's context\nwindow.\n\n`svgrid_scaffold` and the Studio tools run on `@svgrid/enterprise`, which the\nserver loads only when it is installed beside it. `npx -y @svgrid/mcp` leaves\nit out so the server starts fast. To scaffold, install both in the project and\nrun the local server:\n\n```bash\nnpm i -D @svgrid/mcp @svgrid/enterprise\nclaude mcp add svgrid -- npx svgrid-mcp\n```\n\nWithout `-y`, `npx` runs the installed copy instead of downloading one, so\nthere is no install on the first connect.\n\nThe **hosted** server carries `search`, `fetch`, `list_examples`,\n`get_example_source`, `get_api_reference`, `check_svgrid_code` and\n`svgrid_preview`, plus the same resources and prompts. The first two are named\nthat way deliberately: those exact names are what a one-click connector needs to\nindex a remote server. It also answers to `svgrid_search`, `svgrid_get` and\n`svgrid_check_code`, so a model that learned the npm package's names is never\ntold \"unknown tool\".\n\nWhat it does **not** have is the compile pass (the Svelte compiler cannot run in\na Worker), the Studio tools (they need a filesystem), and the offline guarantee\n- it is a remote server, so by definition your query reaches it. Use the local\none when any of those matter.\n\n### Verification\n\n#### `svgrid_check_code`\n\nChecks a file **against the version you have installed** and returns\nline-numbered diagnostics with the exact replacement for each: unknown\n`<SvGrid>` props, wrong `ColumnDef` keys, api methods that do not\nexist, imports that will not resolve, and Svelte 5 runes mistakes. It\nalso compiles the component, so a syntax error surfaces here rather\nthan in the user's terminal.\n\nThis is the tool that makes the rest worth having. Retrieval alone\nstill lets a model write a confident, wrong grid; nothing else in the\nserver stops it. Run it on every file you write, fix what it reports,\nand run it again.\n\n### Finding and reading\n\n#### `svgrid_search`\n\nOne call across all three corpora, because \"how do I pin a column\" does\nnot announce whether it is answered by a doc page, a demo, or an API\nname:\n\n```json\n{ \"query\": \"pin a column\" }\n```\n\nReturns ranked doc hits with excerpts, matching demos, and matching API\nnames - each with the reference you pass to `svgrid_get`. Narrow with\n`kind` (`docs` | `examples` | `api`), trade tokens for completeness\nwith `detail` (`concise` | `full`), and cap with `limit` (max 50).\n\nCall it with **no arguments** for an index of doc sections, demo\ncategories and API groups - the cheapest way to orient before searching.\n\n#### `svgrid_get`\n\n```json\n{ \"ref\": \"help/columns/column-definitions\" }\n{ \"ref\": \"11-stock-market\" }\n{ \"ref\": \"api\" }\n```\n\nThe kind is inferred from the reference; pass `kind` to force it. A\nreference that does not resolve comes back with near matches rather\nthan a bare \"not found\".\n\n### It fixes what it finds\n\n`svgrid_check_code` does not stop at telling you what is wrong. Where the\ncorrection is exact it returns the corrected file too:\n\n```json\n{\n \"diagnostics\": [ ... ],\n \"applied\": [\"rowData -> data (line 7)\", \"accessorKey -> field (line 3)\"],\n \"fixed\": \"<the corrected source>\"\n}\n```\n\nBoth fields are omitted when there is nothing mechanically fixable, so a clean\ncheck stays small.\n\nThe rules are deliberately narrow, because this tool's whole worth is that it\nnever cries wolf - and rewriting raises the stakes from wasting a turn to\ncorrupting working code. Only exact renames are applied (a known rename, or a\nclose-enough spelling), only on word boundaries, and only on the line the\ndiagnostic reported. Advice that is ambiguous - \"this has no equivalent, remove\nit\" - stays advice. A test asserts that running auto-fix across every demo\nin this repository changes nothing at all.\n\n### It knows which version you have\n\nThe result carries the version it checked against:\n\n```json\n{ \"version\": { \"corpus\": \"3.0.0\", \"installed\": \"2.6.8\", \"warning\": \"...\" } }\n```\n\nThe server reads the `@svgrid/grid` in your project and says plainly when it\ndisagrees with the corpus it ships, so a model is told to hedge instead of\nconfidently citing an API you do not have. A server that answers from a remote\nbackend cannot do this - it has no idea what is in your `node_modules`.\n\n### Nothing leaves your machine\n\nThis server makes **no network calls**. Not for docs, not for the API surface,\nnot for verification - the whole corpus (every demo, every doc page, the full\nexported API) ships inside the package, which is why it is megabytes rather\nthan kilobytes. There is no API key, no licence check on the wire, and no telemetry.\n\nIt works on a plane, behind a corporate proxy, and inside an air-gapped\nnetwork. `tools/mcp-offline.test.ts` enforces it: the server is driven through\na full session with every network primitive booby-trapped, and the test fails if\nanything is even attempted.\n\nThe one honest exception: the **preview** loads the grid from a CDN to draw it.\nThat happens in your client's sandboxed iframe, never in this server, and only\nwhen you actually render a preview.\n\n### Seeing the grid, not just reading about it\n\n`svgrid_preview` renders a **real, interactive SvGrid inside the\nconversation** - sortable, filterable, scrollable. Not a screenshot and not a\nmock table: the same `<sv-grid>` custom element a page would use, loaded from\nthe CDN, with the columns and rows you passed.\n\n```json\n{\n \"title\": \"Team roster\",\n \"columns\": [{ \"field\": \"name\", \"header\": \"Name\" }, { \"field\": \"amount\", \"header\": \"Amount\" }],\n \"data\": [{ \"name\": \"Ada\", \"amount\": 20000 }],\n \"sortable\": true\n}\n```\n\nIt uses [MCP Apps](https://blog.modelcontextprotocol.io/posts/2026-01-26-mcp-apps/),\nthe official UI extension: the tool points at a `ui://` resource through\n`_meta`, the client loads that HTML in a sandboxed iframe, and the tool's\n`structuredContent` arrives over a postMessage bridge. Supported in Claude\n(web and desktop), VS Code and ChatGPT.\n\nClients without UI support are not left out - they ignore the `_meta` and get\na text summary of the same grid, so it is always safe to call. That is the\nextension's own rule, and it is why the text half describes the grid rather\nthan pointing at a picture the reader cannot see.\n\nPass `demo` instead of `columns` / `data` and you get the demo's source: a\ndemo is a Svelte component, not data, so there is nothing to hand the element\nand rendering an empty grid would just look broken.\n\n### Resources and prompts\n\nThe docs and demos are also exposed as **MCP resources**, so a user can\nattach a page or a demo directly instead of hoping the model thinks to\nsearch for it:\n\n```\nsvgrid://doc/help/columns/column-definitions\nsvgrid://example/11-stock-market\n```\n\nAnd three **prompts** ship ready to run: `build_grid`, `explain_api`\nand `review_grid_code`. Each is written to make the model use this\nserver rather than recall SvGrid from training data, and each ends at\n`svgrid_check_code`.\n\n### SvGrid Studio (commercial)\n\nOff unless enabled. Set `SVGRID_MCP_STUDIO=1`, or a valid\n`SVGRID_LICENSE_KEY`, in the server's env, and run the server from a project\nthat has `@svgrid/enterprise` installed beside it\n(`npm i -D @svgrid/mcp @svgrid/enterprise`, then `npx svgrid-mcp`). Without it\nthese tools answer with those steps instead of running.\n\n#### `svgrid_scaffold`\n\nPoint it at a Drizzle schema, sample JSON rows, or an `EntitySchema`\nyou already have, and it infers the schema and generates the `$lib`\nschema module, a `+server.ts` API route and a `+page.svelte` with\n`SvGrid` + `SvGridEditPanel` - in one call. Pass `schemaOnly: true` to\nstop after inference. Generated bodies carry `svgrid:managed` markers,\nso regenerating preserves your edits outside them.\n\n#### The `studio_*` tools\n\nFour tools drive the same validated project model the visual designer\nuses:\n\n| Tool | What it does |\n| ---- | ------------ |\n| `studio_project` | `new`, `load`, `describe`, `config`, `capabilities` |\n| `studio_apply` | A **batch** of model changes - entities, screens, blocks, components - applied in order |\n| `studio_configure` | Theme, auth, access, tenancy, data layer, jobs, deploy target, layouts, in one call |\n| `studio_build` | `validate`, then `generate` the full runnable SvelteKit app |\n\nCall `studio_project` with `action: \"capabilities\"` first: it reports\nexactly what the installed version supports, so the agent does not have\nto guess block kinds or component keys.\n\n`studio_apply` takes a batch on purpose. Building a five-screen app was\ntwenty-odd round trips when every block was its own tool; now a whole\nscreen is one call, and a failure reports which op failed and what\nalready applied.\n\n### The pre-3.0 names\n\nEvery tool name from before 3.0 still answers - `search_docs`,\n`get_doc`, `list_examples`, `get_example_source`, `list_docs`,\n`get_api_reference`, `check_svgrid_code`, `introspect_source`,\n`scaffold_entity` and all 27 individual `studio_*` tools. They are not\n**listed**, because listing is what costs context and answering a name\nyou did not advertise costs nothing. Saved prompts and scripts keep\nworking; nobody pays for the compatibility.\n\nTheir JSON response shape follows 3.0, not 2.x. A model reads either\nwithout trouble; a script that parsed the old shape by hand needs a look.\n\n## Licensing\n\nThe documentation and example tools are free. The Studio code\ngenerators are part of the commercial offering. You can evaluate them\nwithout a key: they run, but prepend a notice comment\nto generated files. To license them, set\nthe key in your MCP client's server config:\n\n```json\n{\n \"mcpServers\": {\n \"svgrid\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"@svgrid/mcp\"],\n \"env\": { \"SVGRID_LICENSE_KEY\": \"YOUR-LICENSE-KEY\" }\n }\n }\n}\n```\n\n## Enterprise features and the trial\n\nMost of SvGrid is the MIT `@svgrid/grid` package. Export to Excel and PDF,\npivot tables, the Kanban board, the scheduler, the spreadsheet and the\nserver-side row model are in `@svgrid/enterprise`, which is commercial.\n\nThe server tells your assistant which is which. In `svgrid_search` results a\npaid doc or demo carries `\"enterprise\": true`, and when a question is about a\npaid feature the answer says so once, with the trial link.\n`svgrid_get` puts the same line at the top of a paid doc or demo, and\n`svgrid_check_code` adds it when the fix is to install the package. Nothing\nis sent anywhere to do this; it is part of the answer text.\n\n`@svgrid/enterprise` runs without a license key and shows a watermark until\none is set, so you can try a feature before deciding. The\n[30-day trial](https://svgrid.com/pricing/?trial=mcp-docs) adds a download of\nevery demo and 30 days of email support.\n\n## Verifying it works\n\nAsk your model: *\"What MCP tools do you have from svgrid?\"* The local\nserver lists five, including `svgrid_check_code` (four Studio tools more when\nenabled); the hosted server lists seven.\n\nThen give it something to catch:\n\n> Check this with svgrid: `<SvGrid rowData={rows} {columns} />`\n\nIt should come back with `rowData` is not a SvGrid prop, use `data`. If\nit explains the code instead of calling a tool, the server is not\nconnected.\n\nIf nothing is listed at all, check your client's MCP log. For the local\nserver the usual cause is `npx` not being on PATH - use the absolute path\nto the binary. For the hosted one, confirm the endpoint answers:\n\n```bash\ncurl https://mcp.svgrid.com/health\n```\n\n**\"The svgrid MCP server timed out\" or \"connection timed out after\n30000ms\".** The server starts in well under a second; the wait is `npx`\ndownloading it. That happens on the first start and again after each new\n`@svgrid/mcp` release, and a slow network or a virus scanner can push it\npast the 30 seconds a client waits. Any of these fixes it:\n\n- use the hosted server, which has nothing to download:\n `claude mcp add --transport http svgrid https://mcp.svgrid.com/mcp`\n- install it once, so later starts skip the download:\n `pnpm add -D @svgrid/mcp`, then `claude mcp add svgrid -- npx svgrid-mcp`\n- give the client longer: in Claude Code, start it with `MCP_TIMEOUT=60000`\n\nThe compile pass of `svgrid_check_code` uses the Svelte in the project the\nserver starts in. Started outside a Svelte project, the check still runs\nits API rules and its `compiler` field reads `unavailable`.\n\n## Security model\n\nThe two ways of running it differ here, so pick deliberately.\n\n**Local (`npx @svgrid/mcp`)**\n\n- Runs on your machine over stdio. No telemetry, no outbound network\n calls, no API key.\n- Serves a documentation and example corpus bundled into the package at\n build time, so answers are pinned to the version you installed.\n- `svgrid_check_code` reads the source you pass it in-process and never\n sends it anywhere.\n\n**Hosted (`https://mcp.svgrid.com/mcp`)**\n\n- Your client sends tool arguments to a Cloudflare Worker we operate.\n Anything you pass to a tool leaves your machine, and for\n `check_svgrid_code` that means **the source you ask it to check**.\n- Each call is recorded: tool name, duration, whether it succeeded, the\n query or id it was given, how many pages matched, and whether the answer\n mentioned the Enterprise trial. For `check_svgrid_code` only the **byte\n length** of the submitted source and the names of the rules that fired\n are recorded, never the source itself. The records stay in Cloudflare\n (Workers logs, and an Analytics Engine dataset kept three months) and are\n read in aggregate to see what people ask for.\n- No account, no key, no cookies. Requests are not tied to a user.\n- If your code cannot leave the building, use the local server. That is\n what it is for.\n\n**Both**\n\n- The Studio tools return generated files as data. Writing them to disk\n is your client's decision, not the server's.\n- See [security](./security.md) for the general supply-chain posture.\n\n## Building your own MCP integrations\n\nThe same docs manifest, JSON Schemas (column definitions, the grid's\noptions, export options, and the chart spec at `schemas/chart-spec.json`,\nwhich `svgrid_get` also serves for \"chart spec schema\"), and `llms.txt`\nfiles are also served directly from the docs site:\n\n```ts\nconst docs = await fetch('https://svgrid.com/docs.json').then((r) => r.json())\nconst schemas = await fetch('https://svgrid.com/schemas/index.json').then((r) => r.json())\nconst llms = await fetch('https://svgrid.com/llms-full.txt').then((r) => r.text())\n```\n\nIf you do not want to run the MCP server, building these into your\nagent's system prompt gives most of the same grounding.\n\n## Without the MCP server\n\nAn agent that can only fetch pages gets the same material over plain HTTP,\nin pieces small enough for one fetch (fetch tools cut a response near\n100 KB):\n\n| URL | What it holds |\n|---|---|\n| `/llms.txt` | The docs index, with links to everything below. `/llms/<product>.txt` (grid, ui, enterprise, sheet, base, studio, compare) is one product's part of it |\n| `/docs/<slug>.md` | One docs page as markdown |\n| `/demos.md`, `/demos/<id>.md` | Every live demo, and one demo with what it shows and its full source |\n| `/api.json` | Every export of `@svgrid/grid` and `@svgrid/enterprise` by module, with the member names of each type |\n| `/api/<name>.json` | One symbol's doc comment and signature, by its lowercased name: `/api/svgridapi.json`, `/api/createreview.json` |\n| `/.well-known/mcp.json` | Where this server is: the remote endpoint, the npm package, the setup commands |\n\n```ts\nconst index = await fetch('https://svgrid.com/api.json').then((r) => r.json())\nconst api = await fetch('https://svgrid.com/api/svgridapi.json').then((r) => r.json())\nconst demo = await fetch('https://svgrid.com/demos/702-change-review.md').then((r) => r.text())\n```\n\nIn a browser with WebMCP (`document.modelContext`), svgrid.com registers\nfour tools for an agent driving the page: `svgrid_search`, `svgrid_get_doc`,\n`svgrid_get_demo` and `svgrid_lookup_api`. They answer from the files\nabove.\n\n## Frequently asked questions\n\n### What is the SvGrid MCP server?\n\nA Model Context Protocol server that lets AI clients (Claude Code, Claude\nDesktop, Cursor, Zed, custom agents) query SvGrid's documentation, read real\ndemo source, scaffold SvelteKit CRUD apps, and check the code they write\nagainst the real API - grounded in the files the package ships, so the model\nanswers from current facts instead of guessing.\n\n### Do I need an API key to run it?\n\nNo, on either transport. A `SVGRID_LICENSE_KEY` is optional and only affects\nthe commercial Studio code generators.\n\n### Local or hosted - which should I use?\n\nHosted (`https://mcp.svgrid.com/mcp`) if you want it working in one paste,\nwith no Node and no config file. Local (`npx @svgrid/mcp`) if you want the\nSvelte compiler pass in `svgrid_check_code`, the Studio tools, offline\nuse, or your source never leaving the machine. See\n[Security model](#security-model) for what each one sends.\n\n### How does it help AI assistants write better SvGrid code?\n\nTwo ways. It exposes the docs, example sources, and API reference as tools,\nso the assistant retrieves version-pinned answers instead of hallucinating an\nAPI from training data. And `svgrid_check_code` closes the loop: the model\nruns what it wrote past the real exported surface and gets told exactly what\nis wrong before you see it. That matters most for Svelte 5, where models\nroutinely mix in outdated Svelte 4 syntax.\n\n### Does it work with a version of SvGrid I have not upgraded to?\n\nThe local server answers for the version bundled in the `@svgrid/mcp` release\nyou install, so pin it alongside `@svgrid/grid` if you are behind. The hosted\nserver always tracks the current release. Either way, every\n`svgrid_check_code` result names the version it checked against in its\n`checkedAgainst` field.\n\n## See also\n\n- [Agent Skill](./skill.md) - the always-on house-style layer; the Claude Code plugin installs it together with this server\n- [LLM grounding](./llm-grounding.md) - the same files used by the MCP server, but documented for direct LLM consumption\n- [Agents](./agents.md) - how to build an AI agent that drives the live grid\n- [AI assistant](./ai.md) - the in-grid AI features (filter / smart-fill / classify / summarise), free in @svgrid/grid\n"
6064
6064
  },
6065
6065
  {
6066
6066
  "slug": "help/migrating-from-ag-grid",
6067
6067
  "path": "docs/help/migrating-from-ag-grid.md",
6068
6068
  "title": "Migrating from AG Grid to SvGrid",
6069
6069
  "section": "Help",
6070
- "markdown": "\n# Migrating from AG Grid to SvGrid\n\nIf you tried AG Grid on a Svelte 5 project - via `ag-grid-svelte`, the\nold `ag-grid-community/svelte`, or a hand-rolled wrapper - you probably\nhit the same friction everyone hits: the bridge between AG Grid's\nReact/Angular-first API and Svelte 5 runes is brittle, the bundle is\nheavy, and the Enterprise pricing only makes sense at scale.\n\n<!-- facts:start ag-grid -->\n> **Facts, checked 9 Nov 2026.** `ag-grid-community` 36.3.0, MIT, last published 4 Nov 2026, 16,400,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `ag-grid-community` 36.3.0 346.4 KB JS, no separate stylesheet (measured 9 Nov 2026). AG Grid pricing, as its site states it: AG Grid Community is free under MIT. AG Grid Enterprise is listed at $999 USD per developer with one year of updates and Zendesk support; the Enterprise Bundle with AG Charts Enterprise is $1,498 USD per developer (https://www.ag-grid.com/license-pricing/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/).\n<!-- facts:end -->\n\nThis page is a 30-minute migration recipe from AG Grid to SvGrid. It\ncovers what maps 1:1, what's different by design, and what you'll lose.\nWe tell you when **not** to switch at the bottom.\n\n## TL;DR\n\n| | AG Grid Community | AG Grid Enterprise | SvGrid Community | @svgrid/enterprise |\n| --- | --- | --- | --- | --- |\n| **Licence** | MIT | Commercial, per developer (the price is in the facts box above) | **MIT** | $599/dev/yr (grid) or $999/dev/yr (grid + spreadsheet + Studio) |\n| **Svelte 5 native** | No (community-built integration) | No (community-built integration) | Yes | Yes |\n| **Bundle (gzipped)** | measured, in the facts box above | not measured (licence) | measured, in the facts box above | lazy-loaded subpaths |\n| **Sorting / filtering** | Yes | Yes | Yes | (in Community) |\n| **Row grouping + aggregation** | No (Enterprise) | Yes | Yes (free) | (in Community) |\n| **Master/detail, tree, range select** | No (Enterprise) | Yes | Yes (free) | (in Community) |\n| **Set filter / Excel-style filter menu** | No (Enterprise) | Yes | Yes (free) | (in Community) |\n| **Server-side data, flat (infinite row model)** | Yes | Yes | Yes (free) | (in Community) |\n| **Server-side row model (grouping, tree, pivot, transactions, selection)** | No (Enterprise) | Yes | No | Yes |\n| **Integrated charts** | No (Enterprise) | Yes, AG Charts | Yes (free) | (in Community) |\n| **CSV export** | Yes (API) | Yes | Yes, with TSV and JSON | (in Community) |\n| **Excel export** | No | Yes | No | Yes |\n| **PDF / styled-HTML export, print** | No | No | No | Yes |\n| **Pivot tables** | No | Yes | No | Yes |\n| **MCP server for AI assistants** | Yes, `ag-mcp` | Yes | Yes, `@svgrid/mcp` | (in Community) |\n| **In-grid AI helpers** | No | No | Yes (bring your own model) | (in Community) |\n\n**SvGrid Community gives you most of what AG Grid sells as Enterprise for\nfree**, and `@svgrid/enterprise` adds Excel, PDF and print output, import,\npivot tables and the server-side row model per developer per year. Compare what each paid tier adds\nbefore comparing prices; both are in the facts box. The trade-offs are\nSvelte-only and a much smaller ecosystem. Every AG Grid cell above follows\nag-grid.com's own pages; the dated list is on the\n[SvGrid vs AG Grid](https://svgrid.com/compare/ag-grid/) page.\n\n## Mental model - what changes\n\nAG Grid is one big object you configure declaratively. SvGrid is a\n**headless engine** (`createSvGrid`) with an optional **render\ncomponent** (`<SvGrid>`) on top - the same split TanStack Table made\npopular. You can use either layer; most projects use the render\ncomponent.\n\nEvery example below runs against this setup:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature, type ColumnDef } from '@svgrid/grid'\n\n type Row = { id: number; make: string; model: string; price: number }\n\n const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n\n let rows = $state<Row[]>([\n { id: 1, make: 'Toyota', model: 'Celica', price: 35000 },\n { id: 2, make: 'Ford', model: 'Mondeo', price: 32000 },\n { id: 3, make: 'Porsche', model: 'Boxster', price: 72000 },\n ])\n const data = rows\n\n const columns: ColumnDef<typeof features, Row>[] = [\n { field: 'make', header: 'Make' },\n { field: 'model', header: 'Model' },\n { field: 'price', header: 'Price', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```svelte\n<!-- AG Grid (via a Svelte wrapper) -->\n<AgGridSvelte\n gridOptions={{\n rowData: rows,\n columnDefs: columns,\n onGridReady: (params) => (gridApi = params.api),\n }}\n/>\n\n<!-- SvGrid -->\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n onApiReady={(api) => (gridApi = api)}\n/>\n```\n\nThree things to note:\n\n1. **No giant `gridOptions` blob.** Each capability is a top-level prop.\n2. **Features are opt-in.** You pass a `features` object built with\n `tableFeatures({...})` - only the features you list ship JS.\n3. **`onApiReady`** gives you a typed `SvGridApi` that is roughly the\n AG Grid `gridApi` equivalent (see the API-mapping table below).\n\n## Column definitions - direct translation\n\nThe shapes are similar enough that you can usually translate by hand\nwithout thinking too hard.\n\n```ts\n// AG Grid\nconst columnDefs: ColDef[] = [\n { field: 'name', headerName: 'Name', sortable: true, filter: true, width: 200 },\n { field: 'price', headerName: 'Price', type: 'numericColumn',\n valueFormatter: ({ value }) => `$${value.toFixed(2)}` },\n { field: 'date', headerName: 'Date',\n valueGetter: ({ data }) => new Date(data.date).toISOString().slice(0, 10) },\n { field: 'status', headerName: 'Status',\n cellRenderer: StatusCellRenderer,\n cellRendererParams: { onChange: handleStatusChange } },\n]\n```\n\n```ts\n// SvGrid\nimport { renderComponent, type ColumnDef } from '@svgrid/grid'\nimport StatusCell from './StatusCell.svelte'\n\nconst columns: ColumnDef<typeof features, Row>[] = [\n { field: 'name', header: 'Name', width: 200 }, // sortable + filterable by default\n { field: 'price', header: 'Price',\n format: { type: 'currency', currency: 'USD' } },\n { field: 'date', header: 'Date',\n format: { type: 'date', pattern: 'y-m-d' } },\n { field: 'status', header: 'Status',\n cell: renderComponent(StatusCell, (ctx) => ({\n value: ctx.getValue(),\n onChange: handleStatusChange,\n })),\n },\n]\n```\n\n### Property mapping\n\n| AG Grid | SvGrid | Notes |\n| --- | --- | --- |\n| `field` | `field` | Same. |\n| `headerName` | `header` | Accepts a string or a snippet/component. |\n| `width` | `width` | Same. |\n| `minWidth` / `maxWidth` | `minWidth` / `maxWidth` | Same. |\n| `sortable: true` | (default) | Sorting is on when `rowSortingFeature` is registered. |\n| `filter: true` | (default) | Filtering is on when `columnFilteringFeature` is registered. |\n| `valueFormatter` | `format: { ... }` | Built-in types: `number`, `currency`, `percent`, `date`. For custom, use `cell`. |\n| `valueGetter` | `fieldFn` | Returns the value for sorting/filtering. |\n| `cellRenderer` + `cellRendererParams` | `cell: renderComponent(C, ctx => props)` | One call, type-checked. |\n| `cellEditor: 'agTextCellEditor'` | `editorType: 'text'` | Built-in: `text`, `number`, `checkbox`, `date`. |\n| `editable: true` | `enableInlineEditing` prop on `<SvGrid>` | Per-grid, not per-column. (Per-column control on the roadmap.) |\n| `pinned: 'left'` / `'right'` | Right-click column menu → Pin | Set programmatically via the api. |\n| `rowGroup: true` | Via `setGroupBy([colId])` | See Grouping below. |\n| `aggFunc: 'sum'` | `aggregate: 'sum'` | Built-in: `sum`, `avg`, `min`, `max`, `count`. |\n\n## Feature registration - the one new thing\n\nAG Grid auto-enables most features; you turn them off. SvGrid is the\nopposite - features are opt-in. The result is a smaller bundle.\n\n```ts\nimport {\n tableFeatures,\n rowSortingFeature,\n columnFilteringFeature,\n columnGroupingFeature,\n rowExpandingFeature,\n rowPaginationFeature,\n rowSelectionFeature,\n} from '@svgrid/grid'\n\nconst features = tableFeatures({\n rowSortingFeature,\n columnFilteringFeature,\n rowSelectionFeature,\n // omit any you don't need - their code won't ship\n})\n```\n\nPass `features` to `<SvGrid>` once. From then on the grid behaves like\nAG Grid's `enableSorting`, `enableFilter`, `rowSelection`, etc. are all\non for the registered features.\n\n## API mapping (`gridApi` → `SvGridApi`)\n\nYou get the SvGrid API from `onApiReady` (equivalent to AG Grid's\n`onGridReady`).\n\n| AG Grid `gridApi.X()` | SvGrid `api.X()` |\n| --- | --- |\n| `setRowData(rows)` | (declarative - just update `data` prop) |\n| `addRow(row)` / `applyTransaction({ add: [row] })` | `api.addRow(row)` / `api.addRows(rows)` |\n| `applyTransaction({ remove: [row] })` | `api.removeRow(rowIndex)` |\n| `getValue(colId, rowNode)` | `api.getCellValue(rowIndex, columnId)` |\n| `setValue(...)` | `api.setCellValue(rowIndex, columnId, value)` |\n| `setColumnVisible(colId, visible)` | `api.setColumnVisible(columnId, visible)` |\n| `getSortModel()` / `setSortModel()` | `api.setSort(columnId, 'asc'\\|'desc'\\|null)` |\n| `setFilterModel({...})` | `api.setFilter(columnId, { operator, value })` |\n| `getDisplayedRowAtIndex(i)` / `forEachNodeAfterFilterAndSort(...)` | `api.getDisplayedRows()` |\n| `getModel()` (raw rows) | `api.getData()` |\n\n## Common patterns\n\n### Sorting + filtering + pagination (the 80% case)\n\n```svelte\n<script lang=\"ts\">\n import {\n SvGrid,\n tableFeatures,\n rowSortingFeature,\n columnFilteringFeature,\n rowPaginationFeature,\n } from '@svgrid/grid'\n\n const features = tableFeatures({\n rowSortingFeature,\n columnFilteringFeature,\n rowPaginationFeature,\n })\n</script>\n\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n showPagination\n showColumnFilters\n/>\n```\n\nAG Grid equivalent: `gridOptions: { defaultColDef: { sortable: true, filter: true }, pagination: true }`.\n\n### Cell editing with persistence\n\n```svelte\n<script lang=\"ts\">\n function onCellValueChange(e: { rowIndex: number; columnId: string; value: unknown }) {\n // Persist however you like (fetch to backend, optimistic local update, etc.)\n console.log('cell changed', e)\n }\n</script>\n\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n enableInlineEditing\n onCellValueChange={onCellValueChange}\n/>\n```\n\nAG Grid equivalent: `onCellValueChanged: ({ data, colDef, newValue, oldValue }) => ...`.\nSvGrid's event payload is column-id + row-index based rather than node-based; the row data is yours\nto mutate (or not) on the `rows` array you passed in.\n\n### Grouping + aggregation\n\n```svelte\n<script lang=\"ts\">\n import {\n SvGrid,\n tableFeatures,\n columnGroupingFeature,\n rowSortingFeature,\n rowExpandingFeature,\n } from '@svgrid/grid'\n\n const features = tableFeatures({\n rowSortingFeature,\n columnGroupingFeature,\n rowExpandingFeature,\n })\n\n const columns = [\n { field: 'department', header: 'Department' },\n { field: 'team', header: 'Team' },\n { field: 'salary', header: 'Salary', aggregate: 'sum',\n format: { type: 'currency', currency: 'USD' } },\n ]\n\n function setGroup(api) {\n api.setGroupBy(['department', 'team'])\n }\n</script>\n\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n showGroupingControls\n onApiReady={setGroup}\n/>\n```\n\nAG Grid Enterprise's `rowGroupPanelShow: 'always'` + `aggFunc: 'sum'` translates 1:1.\n\n### Master / detail\n\nAG Grid Enterprise feature; **free in SvGrid Community**. See\n[demo 08](https://svgrid.com/demos/08-tree-and-master-detail/) for the exact pattern.\n\n### Server-side data\n\nAG Grid's Infinite Row Model (Community) and Server-Side Row Model\n(Enterprise) both take an `IServerSideDatasource` with a `getRows(params)`\nthat answers through `params.success({ rowData, rowCount })`. SvGrid has one\n`ServerDataSource` whose `getRows(request)` returns `{ rows, rowCount }`,\nand two controllers over it: `createServerDataSource` (free: paging or\ninfinite block scroll, sort, filter, CRUD) and `createServerRowModel`\n(Enterprise: lazy grouping, tree, pivot, transactions, selection). Both\nmount through one prop, `<SvGrid rowModel={ctl} />`.\n\nAn existing datasource written for AG Grid keeps working behind one call:\n\n```ts\nimport { adaptCallbackDatasource, createServerRowModel } from '@svgrid/enterprise'\n\nconst ctl = createServerRowModel(adaptCallbackDatasource(myServerSideDatasource), {\n groupBy: ['country'],\n aggregations: [{ col: 'amount', fn: 'sum' }],\n})\n```\n\n`adaptCallbackDatasource` maps the request both ways (`rowGroupCols` ->\n`groupBy`, `valueCols` -> `aggregations`, `pivotCols` -> `pivotBy`, the\nsort and filter models, `context`, `parentNode`) and turns `success` /\n`fail` into a resolved or rejected promise.\n\nMore often the datasource is a thin `fetch` and the real investment is the\nendpoint behind it - a Java or Node service that parses AG Grid's\n`IServerSideGetRowsRequest` JSON. That endpoint stays as it is: post\n`toCallbackRequest(request)` and read the answer back.\n\n```ts\nimport { toCallbackRequest } from '@svgrid/enterprise'\n\nconst source = {\n async getRows(request) {\n const res = await fetch('/api/sales', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify(toCallbackRequest(request)),\n })\n const { rowData, rowCount, pivotResultFields } = await res.json()\n return { rows: rowData, rowCount: rowCount ?? -1, pivotResultFields }\n },\n}\n```\n\n`fromCallbackRequest` is the other direction, for a SvelteKit endpoint that\nhas to keep serving an AG Grid client during the switch. A saved selection\nor a bulk endpoint written against `getServerSideSelectionState()` keeps\nits shape the same way: `toCallbackSelectionState(ctl.getSelectionState())`\nproduces `{ selectAll, toggledNodes }` (or the `nodeId` /\n`selectAllChildren` / `toggledNodes` tree), `ctl.setSelectionState` takes\nthat shape as it is when your group node ids are the group keys, and\n`fromCallbackSelectionState(state, { groupKey, isGroup })` maps them when\na `getRowId` gave group rows ids of their own. Option by option:\n\n| AG Grid | SvGrid |\n| --- | --- |\n| `rowModelType: 'infinite'` + `datasource` | `createServerDataSource(source, { mode: 'infinite' })` + `rowModel={ctl}` |\n| `rowModelType: 'serverSide'` + `serverSideDatasource` | `createServerRowModel(source, options)` + `rowModel={ctl}` |\n| `getRows(params)` + `params.success({ rowData, rowCount })` / `params.fail()` | `getRows(request)` returning `{ rows, rowCount }` / throwing |\n| `cacheBlockSize`, `maxBlocksInCache`, `maxConcurrentDatasourceRequests`, `blockLoadDebounceMillis` | `blockSize`, `maxBlocksInCache`, `maxConcurrentRequests`, `blockLoadDebounceMs` |\n| `infiniteInitialRowCount` / `serverSideInitialRowCount` | `initialRowCount` |\n| `rowCount: -1` (unknown) | `rowCount: -1` |\n| `refreshInfiniteCache()` / `purgeInfiniteCache()` | `ctl.refresh()` / `ctl.purge()` |\n| `request.rowGroupCols`, `groupKeys`, `valueCols`, `pivotCols`, `pivotMode` | `request.groupBy`, `groupKeys`, `aggregations`, `pivotBy`, `pivotMode` |\n| `getChildCount(data)` | `childCount: (row) => row.childCount` |\n| `isServerSideGroupOpenByDefault` | `isGroupOpenByDefault(route, row)` |\n| `getServerSideGroupLevelParams` | `levelParams(level, route)` |\n| `isServerSideGroup` / `getServerSideGroupKey` (tree) | `treeData: true` + `hasChildren` / `getRowId` |\n| `grandTotalRow` / `groupTotalRow` | `grandTotalRow` / `groupFooters` |\n| `refreshServerSide({ route, purge })` / `retryServerSideLoads()` | `ctl.refresh({ route, purge })` / `ctl.retryLoads()` |\n| `applyServerSideTransaction` / `applyServerSideTransactionAsync` / `flushServerSideAsyncTransactions` | `ctl.applyTransaction` / `applyTransactionAsync` / `flushAsyncTransactions` |\n| `applyServerSideRowData` | `ctl.applyRowData` |\n| `serverSideSortAllLevels`, `serverSideEnableClientSideSort`, `serverSideOnlyRefreshFilteredGroups` | `sortAllLevels`, `clientSideSort`, `onlyRefreshFilteredGroups` |\n| `getServerSideSelectionState` / `setServerSideSelectionState` | `ctl.getSelectionState()` / `setSelectionState()` (the `toggledNodes` shape is accepted; `toCallbackSelectionState` / `fromCallbackSelectionState` convert) |\n| `ServerSideTransactionResult.status`: `Applied`, `StoreNotFound`, `StoreLoading`, `StoreWaitingToLoad`, `StoreLoadingFailed`, `Cancelled` | `ServerTransactionResult.status`: the same six, lower camel case (`applied`, `storeNotFound`, ...) - a `switch` on the old spelling falls through silently |\n| `groupSelects: 'descendants'` | `selection: { groupSelects: 'descendants' }` |\n| `pivotResultFields` in `success`, `setPivotResultColumns`, `processPivotResultColDef` | `pivotResultFields` / `pivotResultColumns` in the result, `pivotResultColumn(field, def)` |\n| `pagination` + `paginateChildRows` | `pagination: { pageSize, paginateChildRows }` |\n| `getCacheBlockState()` / `getServerSideGroupLevelState()` | `ctl.getCacheState()` / `ctl.levelStates()` |\n| `serverSideDatasource.destroy()` | `source.destroy()` |\n\nTwo differences worth knowing before you start: SvGrid sends the global\nsearch box to the server as `filterModel.global` (AG Grid's quick filter is\nclient-side only and unsupported under its infinite and server-side models),\nand SvGrid has no viewport row model. See\n[Server grouping](./server/server-grouping.md) and\n[demo 467](https://svgrid.com/demos/467-server-row-model-1m/).\n\n### Excel / PDF export\n\nAG Grid: `gridApi.exportDataAsExcel({...})` (Enterprise-only).\nSvGrid: install `@svgrid/enterprise`, call `api.exportData({ format: 'xlsx', ... })`. See [Data export and printing](./export.md).\n\n```ts\nimport { installEnterprise, setLicenseKey } from '@svgrid/enterprise'\nsetLicenseKey('YOUR-LICENSE-KEY') // your Enterprise key\n\n// inside onApiReady:\nconst pro = installEnterprise(api)\nawait pro.exportData({ format: 'xlsx', filename: 'orders' })\n```\n\n## Gotchas - things that don't translate directly\n\n### 1. Per-column `editable: true`\nEditing is switched on at the grid level (`enableInlineEditing`) and\nnarrowed per column with `editable: boolean | (ctx) => boolean` on the\ncolumn definition, so AG Grid's per-column flag and its callback form both\nmap directly. What does not map is AG Grid's `editable` on a column\ngroup: set it on each child column.\n\n### 2. Column drag-to-reorder\nHeader drag is `enableColumnReorder` on `<SvGrid>`; `setColumnOrder`\non the API does the same from code. See\n[column moving](./columns/column-moving.md) for the events.\n\n### 3. AG Grid `valueGetter` chains\nAG Grid's `valueGetter` can read other column values via the API. In\nSvGrid, `fieldFn` only receives the row; if you need cross-column\ncomputed values, do it in the cell renderer with `ctx.row.original` or\ncompute the derived value upstream and store it in the row.\n\n### 4. `cellClass` / `rowClass` callbacks\nOn the roadmap. For now, render a wrapper element in your `cell` snippet\nwith the conditional class.\n\n### 5. The Status Bar / Side Bar / Tool Panels\nAG Grid's chrome (status bar with row count, side bar with filters and\ncolumns panels) doesn't exist in SvGrid - build it as plain Svelte\nmarkup around the grid. Most teams build their own anyway because\nAG Grid's defaults rarely match a polished design system.\n\n### 6. Set filter (the Excel-style funnel popup)\nSvGrid ships an Excel-style filter menu (free in Community). API surface\nis similar but not identical - see [Set filter](./filtering/set-filter.md).\n\n## When NOT to migrate\n\nBe honest. Stay on AG Grid if you:\n\n- **Use multiple frameworks** - AG Grid has React, Angular, Vue, Solid, Qwik, vanilla adapters. SvGrid is Svelte-only.\n- **Need a push-based viewport row model** - SvGrid's server-side row model covers grouping, tree, pivot, transactions and select-all across unloaded rows, and its real-time page merges deltas into loaded rows, but the server cannot push the visible window over a socket the way the viewport model does.\n- **Need custom tool panels** - SvGrid's tool panel is a fixed Columns + Filters pair. (Custom filter components are covered: a column's `customFilter` renders your own control in the filter menu, the floating filter row and the Filters tab.)\n- **Are mid-project and shipping in <2 weeks** - the migration is a few hours per grid, but only do it when you have buffer.\n- **Have a Svelte 4 codebase you can't upgrade** - SvGrid requires Svelte 5 runes. (Consider [htmlelements.com](https://www.htmlelements.com) for vanilla / multi-framework.)\n\nIf none of those apply: switching drops a per-developer Enterprise licence\nfor features that are in SvGrid's MIT core, cuts the bundle by the\ndifference in the facts box above, and gives you a Svelte-native API that\nplays well with runes.\n\n## Step-by-step migration\n\nA typical migration of a single grid takes 1-3 hours:\n\n1. **Install** - `pnpm add @svgrid/grid` (and `@svgrid/enterprise` if you need export).\n2. **Translate columnDefs** - use the mapping table above. Most columns are 1:1.\n3. **Wrap features** - figure out which AG Grid features you actually use; register only those in `tableFeatures({...})`.\n4. **Swap the component** - `<AgGridSvelte gridOptions={...}>` → `<SvGrid data={rows} columns={columns} features={features}>`.\n5. **Move event handlers** - AG Grid `onCellValueChanged` → SvGrid `onCellValueChange` (signature differs slightly, see above).\n6. **Move API calls** - AG Grid `gridApi.X()` → SvGrid `api.X()` per the API table.\n7. **Test interactions** - sort, filter, edit, select. Most \"just works.\"\n8. **Remove `ag-grid-*` packages** - `pnpm remove ag-grid-community ag-grid-svelte` etc. Inspect your bundle to confirm the drop; the facts box above has both measured sizes.\n\n## Need help migrating?\n\nEnterprise customers get **migration help included** with the support plan\n(architecture review, port one grid for you as a reference). Email\n`support@jqwidgets.com` after purchase, or `sales@jqwidgets.com` for\npre-sales questions.\n\n## Frequently asked questions\n\n### Is SvGrid a drop-in replacement for AG Grid in Svelte?\n\nNot a literal drop-in - there is no `ag-grid-svelte` shim to swap. But the\nconcepts map closely: column definitions, row models, sorting, filtering,\ngrouping, and an imperative API all have direct SvGrid equivalents, so most\nteams port a grid in 30 minutes to a day. It is a configuration translation,\nnot a rewrite.\n\n### What is the SvGrid equivalent of AG Grid Enterprise?\n\n`@svgrid/enterprise`. It adds Excel/PDF/styled-HTML export (CSV, TSV and JSON export are already free in the grid), a printable view, pivot\ntables and data import. The in-grid AI helpers are free in `@svgrid/grid`.\nIt is licensed per developer ($599 for the grid, $999 with the spreadsheet\nand Studio, both covering unlimited apps), and the\nCommunity package is MIT-licensed and free for commercial use.\n\n### Does SvGrid use Svelte 5 runes, or is it a wrapper?\n\nIt is Svelte-5-native. State is `$state` / `$derived` / `$effect` and cells\nrender through Svelte snippets - there is no React or Angular core underneath\nand no framework bridge to keep in sync.\n\n### Will my AG Grid bundle size shrink?\n\nAlmost always. The facts box at the top of this page has both packages\nmeasured the same way on the same date: the full SvGrid render component\nwith its stylesheet against `createGrid` with every AG Grid Community\nmodule registered. You add `@svgrid/enterprise` features only where you\nimport them, and the headless core on its own is a small fraction of the\nrender component.\n\n## More examples\n\n### AG Grid ↔ sv-grid side-by-side\n\nTwo real grids over the same dataset: the installed AG Grid Community on the left, sv-grid on the right. Same global filter drives both. Source code panels for either side.\n\n<div data-docs-demo=\"139-migration-from-ag-grid\" data-height=\"460\"></div>\n\n## See also\n\n- [Getting started](../getting-started.md) - full SvGrid walkthrough\n- [Why headless?](../why-headless.md) - the headless / render-component split\n- [Data export and printing](./export.md) - the `@svgrid/enterprise` feature pack\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](./comparison.md) - the\n measured benchmark and the feature matrix\n- [SvGrid vs AG Grid comparison page](https://svgrid.com/compare/ag-grid/) -\n every claim with its source and date\n"
6070
+ "markdown": "\n# Migrating from AG Grid to SvGrid\n\nIf you tried AG Grid on a Svelte 5 project - via `ag-grid-svelte`, the\nold `ag-grid-community/svelte`, or a hand-rolled wrapper - you probably\nhit the same friction everyone hits: the bridge between AG Grid's\nReact/Angular-first API and Svelte 5 runes is brittle, the bundle is\nheavy, and the Enterprise pricing only makes sense at scale.\n\n<!-- facts:start ag-grid -->\n> **Facts, checked 9 Nov 2026.** `ag-grid-community` 36.3.0, MIT, last published 4 Nov 2026, 16,400,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `ag-grid-community` 36.3.0 346.4 KB JS, no separate stylesheet (measured 9 Nov 2026). AG Grid pricing, as its site states it: AG Grid Community is free under MIT. AG Grid Enterprise is listed at $999 USD per developer with one year of updates and Zendesk support; the Enterprise Bundle with AG Charts Enterprise is $1,498 USD per developer (https://www.ag-grid.com/license-pricing/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/).\n<!-- facts:end -->\n\nThis page is a 30-minute migration recipe from AG Grid to SvGrid. It\ncovers what maps 1:1, what's different by design, and what you'll lose.\nWe tell you when **not** to switch at the bottom.\n\n## TL;DR\n\n| | AG Grid Community | AG Grid Enterprise | SvGrid Community | @svgrid/enterprise |\n| --- | --- | --- | --- | --- |\n| **Licence** | MIT | Commercial, per developer (the price is in the facts box above) | **MIT** | $599/dev/yr (grid) or $999/dev/yr (grid + spreadsheet + Studio) |\n| **Svelte 5 native** | No (community-built integration) | No (community-built integration) | Yes | Yes |\n| **Bundle (gzipped)** | measured, in the facts box above | not measured (licence) | measured, in the facts box above | lazy-loaded subpaths |\n| **Sorting / filtering** | Yes | Yes | Yes | (in Community) |\n| **Row grouping + aggregation** | No (Enterprise) | Yes | Yes (free) | (in Community) |\n| **Master/detail, tree, range select** | No (Enterprise) | Yes | Yes (free) | (in Community) |\n| **Set filter / Excel-style filter menu** | No (Enterprise) | Yes | Yes (free) | (in Community) |\n| **Server-side data, flat (infinite row model)** | Yes | Yes | Yes (free) | (in Community) |\n| **Server-side row model (grouping, tree, pivot, transactions, selection)** | No (Enterprise) | Yes | No | Yes |\n| **Integrated charts** | No (Enterprise) | Yes, AG Charts | Yes (free) | (in Community) |\n| **CSV export** | Yes (API) | Yes | Yes, with TSV and JSON | (in Community) |\n| **Excel export** | No | Yes | No | Yes |\n| **PDF / styled-HTML export, print** | No | No | No | Yes |\n| **Pivot tables** | No | Yes | No | Yes |\n| **MCP server for AI assistants** | Yes, `ag-mcp` | Yes | Yes, `@svgrid/mcp` | (in Community) |\n| **In-grid AI helpers** | No | No | Yes (bring your own model) | (in Community) |\n\n**SvGrid Community gives you most of what AG Grid sells as Enterprise for\nfree**, and `@svgrid/enterprise` adds Excel, PDF and print output, import,\npivot tables and the server-side row model per developer per year. Compare what each paid tier adds\nbefore comparing prices; both are in the facts box. The trade-offs are\nSvelte-only and a much smaller ecosystem. Every AG Grid cell above follows\nag-grid.com's own pages; the dated list is on the\n[SvGrid vs AG Grid](https://svgrid.com/compare/ag-grid/) page.\n\n## Mental model - what changes\n\nAG Grid is one big object you configure declaratively. SvGrid is a\n**headless engine** (`createSvGrid`) with an optional **render\ncomponent** (`<SvGrid>`) on top - the same split TanStack Table made\npopular. You can use either layer; most projects use the render\ncomponent.\n\nEvery example below runs against this setup:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature, type ColumnDef } from '@svgrid/grid'\n\n type Row = { id: number; make: string; model: string; price: number }\n\n const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n\n let rows = $state<Row[]>([\n { id: 1, make: 'Toyota', model: 'Celica', price: 35000 },\n { id: 2, make: 'Ford', model: 'Mondeo', price: 32000 },\n { id: 3, make: 'Porsche', model: 'Boxster', price: 72000 },\n ])\n const data = rows\n\n const columns: ColumnDef<typeof features, Row>[] = [\n { field: 'make', header: 'Make' },\n { field: 'model', header: 'Model' },\n { field: 'price', header: 'Price', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```svelte\n<!-- AG Grid (via a Svelte wrapper) -->\n<AgGridSvelte\n gridOptions={{\n rowData: rows,\n columnDefs: columns,\n onGridReady: (params) => (gridApi = params.api),\n }}\n/>\n\n<!-- SvGrid -->\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n onApiReady={(api) => (gridApi = api)}\n/>\n```\n\nThree things to note:\n\n1. **No giant `gridOptions` blob.** Each capability is a top-level prop.\n2. **Features are opt-in.** You pass a `features` object built with\n `tableFeatures({...})` - only the features you list ship JS.\n3. **`onApiReady`** gives you a typed `SvGridApi` that is roughly the\n AG Grid `gridApi` equivalent (see the API-mapping table below).\n\n## Column definitions - direct translation\n\nThe shapes are similar enough that you can usually translate by hand\nwithout thinking too hard.\n\n```ts\n// AG Grid\nconst columnDefs: ColDef[] = [\n { field: 'name', headerName: 'Name', sortable: true, filter: true, width: 200 },\n { field: 'price', headerName: 'Price', type: 'numericColumn',\n valueFormatter: ({ value }) => `$${value.toFixed(2)}` },\n { field: 'date', headerName: 'Date',\n valueGetter: ({ data }) => new Date(data.date).toISOString().slice(0, 10) },\n { field: 'status', headerName: 'Status',\n cellRenderer: StatusCellRenderer,\n cellRendererParams: { onChange: handleStatusChange } },\n]\n```\n\n```ts\n// SvGrid\nimport { renderComponent, type ColumnDef } from '@svgrid/grid'\nimport StatusCell from './StatusCell.svelte'\n\nconst columns: ColumnDef<typeof features, Row>[] = [\n { field: 'name', header: 'Name', width: 200 }, // sortable + filterable by default\n { field: 'price', header: 'Price',\n format: { type: 'currency', currency: 'USD' } },\n { field: 'date', header: 'Date',\n format: { type: 'date', pattern: 'y-m-d' } },\n { field: 'status', header: 'Status',\n cell: renderComponent(StatusCell, (ctx) => ({\n value: ctx.getValue(),\n onChange: handleStatusChange,\n })),\n },\n]\n```\n\n### Property mapping\n\n| AG Grid | SvGrid | Notes |\n| --- | --- | --- |\n| `field` | `field` | Same. |\n| `headerName` | `header` | Accepts a string or a snippet/component. |\n| `width` | `width` | Same. |\n| `minWidth` / `maxWidth` | `minWidth` / `maxWidth` | Same. |\n| `sortable: true` | (default) | Sorting is on when `rowSortingFeature` is registered. |\n| `filter: true` | (default) | Filtering is on when `columnFilteringFeature` is registered. |\n| `valueFormatter` | `format: { ... }` | Built-in types: `number`, `currency`, `percent`, `date`. For custom, use `cell`. |\n| `valueGetter` | `fieldFn` | Returns the value for sorting/filtering. |\n| `cellRenderer` + `cellRendererParams` | `cell: renderComponent(C, ctx => props)` | One call, type-checked. |\n| `cellEditor: 'agTextCellEditor'` | `editorType: 'text'` | Built-in: `text`, `number`, `checkbox`, `date`. |\n| `editable: true` | `enableInlineEditing` prop on `<SvGrid>` | Per-grid, not per-column. (Per-column control on the roadmap.) |\n| `pinned: 'left'` / `'right'` | Right-click column menu → Pin | Set programmatically via the api. |\n| `rowGroup: true` | Via `setGroupBy([colId])` | See Grouping below. |\n| `aggFunc: 'sum'` | `aggregate: 'sum'` | Built-in: `sum`, `avg`, `min`, `max`, `count`. |\n\n## Feature registration - the one new thing\n\nAG Grid auto-enables most features; you turn them off. SvGrid is the\nopposite - features are opt-in. The result is a smaller bundle.\n\n```ts\nimport {\n tableFeatures,\n rowSortingFeature,\n columnFilteringFeature,\n columnGroupingFeature,\n rowExpandingFeature,\n rowPaginationFeature,\n rowSelectionFeature,\n} from '@svgrid/grid'\n\nconst features = tableFeatures({\n rowSortingFeature,\n columnFilteringFeature,\n rowSelectionFeature,\n // omit any you don't need - their code won't ship\n})\n```\n\nPass `features` to `<SvGrid>` once. From then on the grid behaves like\nAG Grid's `enableSorting`, `enableFilter`, `rowSelection`, etc. are all\non for the registered features.\n\n## API mapping (`gridApi` → `SvGridApi`)\n\nYou get the SvGrid API from `onApiReady` (equivalent to AG Grid's\n`onGridReady`).\n\n| AG Grid `gridApi.X()` | SvGrid `api.X()` |\n| --- | --- |\n| `setRowData(rows)` | (declarative - just update `data` prop) |\n| `addRow(row)` / `applyTransaction({ add: [row] })` | `api.addRow(row)` / `api.addRows(rows)` |\n| `applyTransaction({ remove: [row] })` | `api.removeRow(rowIndex)` |\n| `getValue(colId, rowNode)` | `api.getCellValue(rowIndex, columnId)` |\n| `setValue(...)` | `api.setCellValue(rowIndex, columnId, value)` |\n| `setColumnVisible(colId, visible)` | `api.setColumnVisible(columnId, visible)` |\n| `getSortModel()` / `setSortModel()` | `api.setSort(columnId, 'asc'\\|'desc'\\|null)` |\n| `setFilterModel({...})` | `api.setFilter(columnId, { operator, value })` |\n| `getDisplayedRowAtIndex(i)` / `forEachNodeAfterFilterAndSort(...)` | `api.getDisplayedRows()` |\n| `getModel()` (raw rows) | `api.getData()` |\n\n## Common patterns\n\n### Sorting + filtering + pagination (the 80% case)\n\n```svelte\n<script lang=\"ts\">\n import {\n SvGrid,\n tableFeatures,\n rowSortingFeature,\n columnFilteringFeature,\n rowPaginationFeature,\n } from '@svgrid/grid'\n\n const features = tableFeatures({\n rowSortingFeature,\n columnFilteringFeature,\n rowPaginationFeature,\n })\n</script>\n\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n showPagination\n showColumnFilters\n/>\n```\n\nAG Grid equivalent: `gridOptions: { defaultColDef: { sortable: true, filter: true }, pagination: true }`.\n\n### Cell editing with persistence\n\n```svelte\n<script lang=\"ts\">\n function onCellValueChange(e: { rowIndex: number; columnId: string; value: unknown }) {\n // Persist however you like (fetch to backend, optimistic local update, etc.)\n console.log('cell changed', e)\n }\n</script>\n\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n enableInlineEditing\n onCellValueChange={onCellValueChange}\n/>\n```\n\nAG Grid equivalent: `onCellValueChanged: ({ data, colDef, newValue, oldValue }) => ...`.\nSvGrid's event payload is column-id + row-index based rather than node-based; the row data is yours\nto mutate (or not) on the `rows` array you passed in.\n\n### Grouping + aggregation\n\n```svelte\n<script lang=\"ts\">\n import {\n SvGrid,\n tableFeatures,\n columnGroupingFeature,\n rowSortingFeature,\n rowExpandingFeature,\n } from '@svgrid/grid'\n\n const features = tableFeatures({\n rowSortingFeature,\n columnGroupingFeature,\n rowExpandingFeature,\n })\n\n const columns = [\n { field: 'department', header: 'Department' },\n { field: 'team', header: 'Team' },\n { field: 'salary', header: 'Salary', aggregate: 'sum',\n format: { type: 'currency', currency: 'USD' } },\n ]\n\n function setGroup(api) {\n api.setGroupBy(['department', 'team'])\n }\n</script>\n\n<SvGrid\n data={rows}\n columns={columns}\n features={features}\n showGroupingControls\n onApiReady={setGroup}\n/>\n```\n\nAG Grid Enterprise's `rowGroupPanelShow: 'always'` + `aggFunc: 'sum'` translates 1:1.\n\n### Master / detail\n\nAG Grid Enterprise feature; **free in SvGrid Community**. See\n[demo 08](https://svgrid.com/demos/08-tree-and-master-detail/) for the exact pattern.\n\n### Server-side data\n\nAG Grid's Infinite Row Model (Community) and Server-Side Row Model\n(Enterprise) both take an `IServerSideDatasource` with a `getRows(params)`\nthat answers through `params.success({ rowData, rowCount })`. SvGrid has one\n`ServerDataSource` whose `getRows(request)` returns `{ rows, rowCount }`,\nand two controllers over it: `createServerDataSource` (free: paging or\ninfinite block scroll, sort, filter, CRUD) and `createServerRowModel`\n(Enterprise: lazy grouping, tree, pivot, transactions, selection). Both\nmount through one prop, `<SvGrid rowModel={ctl} />`.\n\nAn existing datasource written for AG Grid keeps working behind one call:\n\n```ts\nimport { adaptCallbackDatasource, createServerRowModel } from '@svgrid/enterprise'\n\nconst ctl = createServerRowModel(adaptCallbackDatasource(myServerSideDatasource), {\n groupBy: ['country'],\n aggregations: [{ col: 'amount', fn: 'sum' }],\n})\n```\n\n`adaptCallbackDatasource` maps the request both ways (`rowGroupCols` ->\n`groupBy`, `valueCols` -> `aggregations`, `pivotCols` -> `pivotBy`, the\nsort and filter models, `context`, `parentNode`) and turns `success` /\n`fail` into a resolved or rejected promise.\n\nMore often the datasource is a thin `fetch` and the real investment is the\nendpoint behind it - a Java or Node service that parses AG Grid's\n`IServerSideGetRowsRequest` JSON. That endpoint stays as it is: post\n`toCallbackRequest(request)` and read the answer back.\n\n```ts\nimport { toCallbackRequest } from '@svgrid/enterprise'\n\nconst source = {\n async getRows(request) {\n const res = await fetch('/api/sales', {\n method: 'POST',\n headers: { 'content-type': 'application/json' },\n body: JSON.stringify(toCallbackRequest(request)),\n })\n const { rowData, rowCount, pivotResultFields } = await res.json()\n return { rows: rowData, rowCount: rowCount ?? -1, pivotResultFields }\n },\n}\n```\n\n`fromCallbackRequest` is the other direction, for a SvelteKit endpoint that\nhas to keep serving an AG Grid client during the switch. A saved selection\nor a bulk endpoint written against `getServerSideSelectionState()` keeps\nits shape the same way: `toCallbackSelectionState(ctl.getSelectionState())`\nproduces `{ selectAll, toggledNodes }` (or the `nodeId` /\n`selectAllChildren` / `toggledNodes` tree), `ctl.setSelectionState` takes\nthat shape as it is when your group node ids are the group keys, and\n`fromCallbackSelectionState(state, { groupKey, isGroup })` maps them when\na `getRowId` gave group rows ids of their own. Option by option:\n\n| AG Grid | SvGrid |\n| --- | --- |\n| `rowModelType: 'infinite'` + `datasource` | `createServerDataSource(source, { mode: 'infinite' })` + `rowModel={ctl}` |\n| `rowModelType: 'serverSide'` + `serverSideDatasource` | `createServerRowModel(source, options)` + `rowModel={ctl}` |\n| `getRows(params)` + `params.success({ rowData, rowCount })` / `params.fail()` | `getRows(request)` returning `{ rows, rowCount }` / throwing |\n| `cacheBlockSize`, `maxBlocksInCache`, `maxConcurrentDatasourceRequests`, `blockLoadDebounceMillis` | `blockSize`, `maxBlocksInCache`, `maxConcurrentRequests`, `blockLoadDebounceMs` |\n| `infiniteInitialRowCount` / `serverSideInitialRowCount` | `initialRowCount` |\n| `rowCount: -1` (unknown) | `rowCount: -1` |\n| `refreshInfiniteCache()` / `purgeInfiniteCache()` | `ctl.refresh()` / `ctl.purge()` |\n| `request.rowGroupCols`, `groupKeys`, `valueCols`, `pivotCols`, `pivotMode` | `request.groupBy`, `groupKeys`, `aggregations`, `pivotBy`, `pivotMode` |\n| `getChildCount(data)` | `childCount: (row) => row.childCount` |\n| `isServerSideGroupOpenByDefault` | `isGroupOpenByDefault(route, row)` |\n| `getServerSideGroupLevelParams` | `levelParams(level, route)` |\n| `isServerSideGroup` / `getServerSideGroupKey` (tree) | `treeData: true` + `hasChildren` / `getRowId` |\n| `grandTotalRow` / `groupTotalRow` | `grandTotalRow` / `groupFooters` |\n| `refreshServerSide({ route, purge })` / `retryServerSideLoads()` | `ctl.refresh({ route, purge })` / `ctl.retryLoads()` |\n| `applyServerSideTransaction` / `applyServerSideTransactionAsync` / `flushServerSideAsyncTransactions` | `ctl.applyTransaction` / `applyTransactionAsync` / `flushAsyncTransactions` |\n| `applyServerSideRowData` | `ctl.applyRowData` |\n| `serverSideSortAllLevels`, `serverSideEnableClientSideSort`, `serverSideOnlyRefreshFilteredGroups` | `sortAllLevels`, `clientSideSort`, `onlyRefreshFilteredGroups` |\n| `getServerSideSelectionState` / `setServerSideSelectionState` | `ctl.getSelectionState()` / `setSelectionState()` (the `toggledNodes` shape is accepted; `toCallbackSelectionState` / `fromCallbackSelectionState` convert) |\n| `ServerSideTransactionResult.status`: `Applied`, `StoreNotFound`, `StoreLoading`, `StoreWaitingToLoad`, `StoreLoadingFailed`, `Cancelled` | `ServerTransactionResult.status`: the same six, lower camel case (`applied`, `storeNotFound`, ...) - a `switch` on the old spelling falls through silently |\n| `groupSelects: 'descendants'` | `selection: { groupSelects: 'descendants' }` |\n| `pivotResultFields` in `success`, `setPivotResultColumns`, `processPivotResultColDef` | `pivotResultFields` / `pivotResultColumns` in the result, `pivotResultColumn(field, def)` |\n| `pagination` + `paginateChildRows` | `pagination: { pageSize, paginateChildRows }` |\n| `getCacheBlockState()` / `getServerSideGroupLevelState()` | `ctl.getCacheState()` / `ctl.levelStates()` |\n| `serverSideDatasource.destroy()` | `source.destroy()` |\n\nTwo differences worth knowing before you start: SvGrid sends the global\nsearch box to the server as `filterModel.global` (AG Grid's quick filter is\nclient-side only and unsupported under its infinite and server-side models),\nand SvGrid has no viewport row model. See\n[Server grouping](./server/server-grouping.md) and\n[demo 467](https://svgrid.com/demos/467-server-row-model-1m/).\n\n### Excel / PDF export\n\nAG Grid: `gridApi.exportDataAsExcel({...})` (Enterprise-only).\nSvGrid: install `@svgrid/enterprise`, call `api.exportData({ format: 'xlsx', ... })`. See [Data export and printing](./export.md).\n\n```ts\nimport { installEnterprise, setLicenseKey } from '@svgrid/enterprise'\nsetLicenseKey('YOUR-LICENSE-KEY') // your Enterprise key\n\n// inside onApiReady:\nconst pro = installEnterprise(api)\nawait pro.exportData({ format: 'xlsx', filename: 'orders' })\n```\n\n## Gotchas - things that don't translate directly\n\n### 1. Per-column `editable: true`\nEditing is switched on at the grid level (`enableInlineEditing`) and\nnarrowed per column with `editable: boolean | (ctx) => boolean` on the\ncolumn definition, so AG Grid's per-column flag and its callback form both\nmap directly. What does not map is AG Grid's `editable` on a column\ngroup: set it on each child column.\n\n### 2. Column drag-to-reorder\nHeader drag is `enableColumnReorder` on `<SvGrid>`; `setColumnOrder`\non the API does the same from code. See\n[column moving](./columns/column-moving.md) for the events.\n\n### 3. AG Grid `valueGetter` chains\nAG Grid's `valueGetter` can read other column values via the API. In\nSvGrid, `fieldFn` only receives the row; if you need cross-column\ncomputed values, do it in the cell renderer with `ctx.row.original` or\ncompute the derived value upstream and store it in the row.\n\n### 4. `cellClass` / `rowClass` callbacks\nOn the roadmap. For now, render a wrapper element in your `cell` snippet\nwith the conditional class.\n\n### 5. The Status Bar / Side Bar / Tool Panels\nAG Grid's chrome (status bar with row count, side bar with filters and\ncolumns panels) doesn't exist in SvGrid - build it as plain Svelte\nmarkup around the grid. Most teams build their own anyway because\nAG Grid's defaults rarely match a polished design system.\n\n### 6. Set filter (the Excel-style funnel popup)\nSvGrid ships an Excel-style filter menu (free in Community). API surface\nis similar but not identical - see [Set filter](./filtering/set-filter.md).\n\n## When NOT to migrate\n\nBe honest. Stay on AG Grid if you:\n\n- **Use multiple frameworks** - AG Grid has React, Angular, Vue, Solid, Qwik, vanilla adapters. SvGrid is Svelte-only.\n- **Need a push-based viewport row model** - SvGrid's server-side row model covers grouping, tree, pivot, transactions and select-all across unloaded rows, and its real-time page merges deltas into loaded rows, but the server cannot push the visible window over a socket the way the viewport model does.\n- **Need custom tool panels** - SvGrid's tool panel is a fixed Columns + Filters pair. (Custom filter components are covered: a column's `customFilter` renders your own control in the filter menu, the floating filter row and the Filters tab.)\n- **Are mid-project and shipping in <2 weeks** - the migration is a few hours per grid, but only do it when you have buffer.\n- **Have a Svelte 4 codebase you can't upgrade** - SvGrid requires Svelte 5 runes. (Consider [htmlelements.com](https://www.htmlelements.com) for vanilla / multi-framework.)\n\nIf none of those apply: switching drops a per-developer Enterprise licence\nfor features that are in SvGrid's MIT core, cuts the bundle by the\ndifference in the facts box above, and gives you a Svelte-native API that\nplays well with runes.\n\n## Step-by-step migration\n\nA typical migration of a single grid takes 1-3 hours:\n\n1. **Install** - `pnpm add @svgrid/grid` (and `@svgrid/enterprise` if you need export).\n2. **Translate columnDefs** - use the mapping table above. Most columns are 1:1.\n3. **Wrap features** - figure out which AG Grid features you actually use; register only those in `tableFeatures({...})`.\n4. **Swap the component** - `<AgGridSvelte gridOptions={...}>` → `<SvGrid data={rows} columns={columns} features={features}>`.\n5. **Move event handlers** - AG Grid `onCellValueChanged` → SvGrid `onCellValueChange` (signature differs slightly, see above).\n6. **Move API calls** - AG Grid `gridApi.X()` → SvGrid `api.X()` per the API table.\n7. **Test interactions** - sort, filter, edit, select. Most \"just works.\"\n8. **Remove `ag-grid-*` packages** - `pnpm remove ag-grid-community ag-grid-svelte` etc. Inspect your bundle to confirm the drop; the facts box above has both measured sizes.\n\n## Need help migrating?\n\nEnterprise customers get **migration help included** with the support plan\n(architecture review, port one grid for you as a reference). Email\n`support@jqwidgets.com` after purchase, or `sales@jqwidgets.com` for\npre-sales questions.\n\n## Frequently asked questions\n\n### Is SvGrid a drop-in replacement for AG Grid in Svelte?\n\nNot a literal drop-in - there is no `ag-grid-svelte` shim to swap. But the\nconcepts map closely: column definitions, row models, sorting, filtering,\ngrouping, and an imperative API all have direct SvGrid equivalents, so most\nteams port a grid in 30 minutes to a day. It is a configuration translation,\nnot a rewrite.\n\n### What is the SvGrid equivalent of AG Grid Enterprise?\n\n`@svgrid/enterprise`. It adds Excel/PDF/styled-HTML export (CSV, TSV and JSON export are already free in the grid), a printable view, pivot\ntables and data import. The in-grid AI helpers are free in `@svgrid/grid`.\nIt is licensed per developer ($599 for the grid, $999 with the spreadsheet\nand Studio, both covering unlimited apps), and the\nCommunity package is MIT-licensed and free for commercial use.\n\n### Does SvGrid use Svelte 5 runes, or is it a wrapper?\n\nIt is Svelte-5-native. State is `$state` / `$derived` / `$effect` and cells\nrender through Svelte snippets - there is no React or Angular core underneath\nand no framework bridge to keep in sync.\n\n### Will my AG Grid bundle size shrink?\n\nAlmost always. The facts box at the top of this page has both packages\nmeasured the same way on the same date: the full SvGrid render component\nwith its stylesheet against `createGrid` with every AG Grid Community\nmodule registered. You add `@svgrid/enterprise` features only where you\nimport them, and the headless core on its own is a small fraction of the\nrender component.\n\n## More examples\n\n### AG Grid ↔ sv-grid side-by-side\n\nTwo real grids over the same dataset: the installed AG Grid Community on the left, sv-grid on the right. Same global filter drives both. Source code panels for either side.\n\n<div data-docs-demo=\"139-migration-from-ag-grid\" data-height=\"460\"></div>\n\n## See also\n\n- [Getting started](../getting-started.md) - full SvGrid walkthrough\n- [Why headless?](../why-headless.md) - the headless / render-component split\n- [Data export and printing](./export.md) - the `@svgrid/enterprise` feature pack\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](./comparison.md) - the\n measured benchmark and the feature matrix\n- [SvGrid vs AG Grid comparison page](https://svgrid.com/compare/ag-grid/) -\n every claim with its source and date\n"
6071
6071
  },
6072
6072
  {
6073
6073
  "slug": "help/migrating-from-airtable",
@@ -6088,49 +6088,49 @@ export const docs = [
6088
6088
  "path": "docs/help/migrating-from-devextreme.md",
6089
6089
  "title": "Migrating from DevExtreme DataGrid (DevExpress)",
6090
6090
  "section": "Help",
6091
- "markdown": "# Migrating from DevExtreme DataGrid (DevExpress)\n\nDevExtreme DataGrid is a deep, commercial multi-framework grid from\nDevExpress (jQuery, React, Vue, Angular). Moving to SvGrid trades the\ncommercial suite for a Svelte-5-native grid with an MIT core and a far\nlower-cost paid tier.\n\n> Estimated effort: **3-6 hours** per grid.\n>\n> DevExtreme's option names vary by framework flavour; map your version's\n> columns and events onto the SvGrid equivalents below.\n\n<!-- facts:start devextreme-datagrid -->\n> **Facts, checked 9 Nov 2026.** `devextreme` 26.1.6, Commercial (see licence file), last published 23 Oct 2026, 1,180,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). DevExtreme DataGrid pricing, as its site states it: js.devexpress.com lists DevExtreme Complete at $899.99 per developer for a 12-month subscription with updates and support, renewing at $449.99; no free or non-commercial tier is shown (https://js.devexpress.com/Buy/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs DevExtreme DataGrid (DevExpress)](https://svgrid.com/compare/devextreme-datagrid/).\n<!-- facts:end -->\n\n## Concept map\n\n| DevExtreme DataGrid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `columns: [{ dataField, caption }]` | `columns: [{ field, header }]` |\n| `cellTemplate` | `cell: (c) => renderSnippet(...)` |\n| `editing: { mode, allowUpdating }` | `enableInlineEditing` + `editorType` |\n| `sorting` / `filterRow` / `headerFilter` | `rowSortingFeature` / `columnFilteringFeature` |\n| `grouping` + `summary` | `columnGroupingFeature` + aggregates |\n| `masterDetail` | `rowExpandingFeature` / master-detail rows |\n| `dataSource` (store / remote) | `data` array or `externalSort/Filter` |\n| `paging` + `pager` | `showPagination` |\n| `export` | `@svgrid/enterprise` export pack |\n| `stateStoring` | persist column state yourself (saved views) |\n\n## Shape of the change\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- $('#grid').dxDataGrid({\n- dataSource: rows,\n- paging: { pageSize: 25 },\n- sorting: { mode: 'multiple' }, filterRow: { visible: true },\n- editing: { mode: 'cell', allowUpdating: true },\n- columns: [\n- { dataField: 'name', caption: 'Name' },\n- { dataField: 'amount', caption: 'Amount', format: 'currency' },\n- ],\n- })\n\n+ <SvGrid\n+ data={rows}\n+ columns={[\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]}\n+ features={tableFeatures({ rowSortingFeature, columnFilteringFeature })}\n+ showPagination enableInlineEditing />\n```\n\n## Why teams switch\n\n- **Svelte 5 native** instead of a multi-framework commercial library.\n- **MIT community core** plus an optional Enterprise pack for export / import /\n pivot / AI.\n- **Much lower-cost paid tier.**\n\n## What to weigh\n\n- DevExtreme has a **larger built-in feature surface** (state storing,\n broad export, a full suite). If you depend on those, factor them in -\n `stateStoring` in particular becomes a saved-views pattern you wire up.\n\n## Frequently asked questions\n\n### How do DevExtreme columns map to SvGrid?\n\n`dataField` becomes `field`, `caption` becomes `header`, and `cellTemplate`\nbecomes a `renderSnippet` cell. Sorting / filtering / grouping move from grid\noptions to registered features.\n\n### Is SvGrid cheaper than DevExtreme?\n\nFor a Svelte stack, yes - the core is MIT and the paid tier is far below\ncommercial-suite pricing. You give up DevExtreme's multi-framework reach and\nsome built-in features like `stateStoring`.\n\n### Does SvGrid export like DevExtreme?\n\nYes, via the `@svgrid/enterprise` pack: Excel, PDF, CSV, TSV, and HTML export plus a\nprintable view.\n\n## What you end up with\n\nGrouping with totals plus the filter row, the DevExtreme default layout.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} summary groupable sortable filterMode=\"row\" />\n```\n\n## See also\n\n- [SvGrid vs DevExtreme DataGrid](https://svgrid.com/compare/devextreme-datagrid/) - the side-by-side comparison\n- [Saved views](./saved-views.md) - the state-persistence pattern\n- [Data export and printing - Enterprise](./export.md) - the Enterprise export pack\n"
6091
+ "markdown": "# Migrating from DevExtreme DataGrid (DevExpress)\n\nDevExtreme DataGrid is a deep, commercial multi-framework grid from\nDevExpress (jQuery, React, Vue, Angular). Moving to SvGrid trades the\ncommercial suite for a Svelte-5-native grid with an MIT core and a far\nlower-cost paid tier.\n\n> Estimated effort: **3-6 hours** per grid.\n>\n> DevExtreme's option names vary by framework flavour; map your version's\n> columns and events onto the SvGrid equivalents below.\n\n<!-- facts:start devextreme-datagrid -->\n> **Facts, checked 9 Nov 2026.** `devextreme` 26.1.6, Commercial (see licence file), last published 23 Oct 2026, 1,180,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). DevExtreme DataGrid pricing, as its site states it: js.devexpress.com lists DevExtreme Complete at $899.99 per developer for a 12-month subscription with updates and support, renewing at $449.99; no free or non-commercial tier is shown (https://js.devexpress.com/Buy/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs DevExtreme DataGrid (DevExpress)](https://svgrid.com/compare/devextreme-datagrid/).\n<!-- facts:end -->\n\n## Concept map\n\n| DevExtreme DataGrid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `columns: [{ dataField, caption }]` | `columns: [{ field, header }]` |\n| `cellTemplate` | `cell: (c) => renderSnippet(...)` |\n| `editing: { mode, allowUpdating }` | `enableInlineEditing` + `editorType` |\n| `sorting` / `filterRow` / `headerFilter` | `rowSortingFeature` / `columnFilteringFeature` |\n| `grouping` + `summary` | `columnGroupingFeature` + aggregates |\n| `masterDetail` | `rowExpandingFeature` / master-detail rows |\n| `dataSource` (store / remote) | `data` array or `externalSort/Filter` |\n| `paging` + `pager` | `showPagination` |\n| `export` | `@svgrid/enterprise` export pack |\n| `stateStoring` | persist column state yourself (saved views) |\n\n## Shape of the change\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- $('#grid').dxDataGrid({\n- dataSource: rows,\n- paging: { pageSize: 25 },\n- sorting: { mode: 'multiple' }, filterRow: { visible: true },\n- editing: { mode: 'cell', allowUpdating: true },\n- columns: [\n- { dataField: 'name', caption: 'Name' },\n- { dataField: 'amount', caption: 'Amount', format: 'currency' },\n- ],\n- })\n\n+ <SvGrid\n+ data={rows}\n+ columns={[\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]}\n+ features={tableFeatures({ rowSortingFeature, columnFilteringFeature })}\n+ showPagination enableInlineEditing />\n```\n\n## Why teams switch\n\n- **Svelte 5 native** instead of a multi-framework commercial library.\n- **MIT community core** plus an optional Enterprise pack for export / import /\n pivot / AI.\n- **Much lower-cost paid tier.**\n\n## What to weigh\n\n- DevExtreme has a **larger built-in feature surface** (state storing,\n broad export, a full suite). If you depend on those, factor them in -\n `stateStoring` in particular becomes a saved-views pattern you wire up.\n\n## Frequently asked questions\n\n### How do DevExtreme columns map to SvGrid?\n\n`dataField` becomes `field`, `caption` becomes `header`, and `cellTemplate`\nbecomes a `renderSnippet` cell. Sorting / filtering / grouping move from grid\noptions to registered features.\n\n### Is SvGrid cheaper than DevExtreme?\n\nFor a Svelte stack, yes - the core is MIT and the paid tier is far below\ncommercial-suite pricing. You give up DevExtreme's multi-framework reach and\nsome built-in features like `stateStoring`.\n\n### Does SvGrid export like DevExtreme?\n\nYes, via the `@svgrid/enterprise` pack: Excel, PDF, CSV, TSV, and HTML export plus a\nprintable view.\n\n## What you end up with\n\nGrouping with totals plus the filter row, the DevExtreme default layout.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} summary groupable sortable filterMode=\"row\" />\n```\n\n## See also\n\n- [SvGrid vs DevExtreme DataGrid](https://svgrid.com/compare/devextreme-datagrid/) - the side-by-side comparison\n- [Saved views](./saved-views.md) - the state-persistence pattern\n- [Data export and printing - Enterprise](./export.md) - the Enterprise export pack\n"
6092
6092
  },
6093
6093
  {
6094
6094
  "slug": "help/migrating-from-glide",
6095
6095
  "path": "docs/help/migrating-from-glide.md",
6096
6096
  "title": "Migrating from Glide Data Grid",
6097
6097
  "section": "Help",
6098
- "markdown": "# Migrating from Glide Data Grid\n\nGlide Data Grid is the canvas-rendered, ultra-high-performance choice\nfor very wide datasets (100k+ rows, 100+ columns). Sv-grid renders to\nDOM with virtualization, which trades peak FPS for\naccessibility + standards-compliant HTML.\n\n> Estimated effort: **3-6 hours** per grid. Glide's getCellContent /\n> onCellEdited cell-data model is the opposite of sv-grid's\n> data-array model; the mental flip takes the most time.\n\n<!-- facts:start glide-data-grid -->\n> **Facts, checked 9 Nov 2026.** `@glideapps/glide-data-grid` 6.0.3, MIT, last published 3 Feb 2024, 2,010,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `@glideapps/glide-data-grid` 6.0.3 71.1 KB JS + 0.2 KB CSS, lodash + marked + react + react-dom + react-responsive-carousel external (measured 9 Nov 2026). Glide Data Grid pricing, as its site states it: Glide Data Grid is MIT and, in its own words, fully free and open source; no licence is sold for the grid (https://grid.glideapps.com/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/).\n<!-- facts:end -->\n\n## Mental-model flip\n\nGlide gives you a **canvas + a `getCellContent` callback**: the grid\nasks you for each visible cell on demand. Sv-grid wants the\n**whole data array up front** (`<SvGrid data={rows}>`); the engine\nslices for virtualization.\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- function getCellContent([col, row]) {\n- const r = rows[row]\n- if (col === 0) return { kind: 'text', data: r.name, displayData: r.name, allowOverlay: true }\n- if (col === 1) return { kind: 'number', data: r.amount, displayData: fmt(r.amount), allowOverlay: true }\n- }\n- <DataEditor columns={[...]} getCellContent={getCellContent} rows={rows.length} />\n\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name', editorType: 'text' },\n+ { field: 'amount', header: 'Amount', editorType: 'number',\n+ format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ <SvGrid data={rows} columns={columns} features={features} />\n```\n\nThis change is **good** for 99% of apps: it's less code, less risk\nof typos in cell coordinates, full TypeScript type checking on the\nrow shape. For the 1% of apps with rows that don't fit in memory,\nsee \"Out-of-memory data\" below.\n\n## Cell editing\n\n```diff\n- function onCellEdited([col, row], newValue) {\n- rows[row] = { ...rows[row], [colToField(col)]: newValue.data }\n- }\n- <DataEditor onCellEdited={onCellEdited} />\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ onCellValueChange={(e) => {\n+ rows[e.rowIndex] = { ...rows[e.rowIndex], [e.columnId]: e.newValue }\n+ }}\n+ />\n```\n\nYou can also rely on sv-grid's built-in write-back if `data` is a\nSvelte 5 `$state` array - mutations propagate automatically.\n\n## Sorting + filtering\n\nGlide doesn't ship sort / filter UI; you wrap your own. Sv-grid\nships them:\n\n```diff\n- // Build your own sort buttons in the toolbar; sort rows yourself.\n\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ <SvGrid data={rows} columns={columns} features={features} filterMode=\"menu\" />\n```\n\n## Frozen columns / row markers\n\n| Glide | sv-grid |\n| -------------------------------------- | --------------------------------------------- |\n| `freezeColumns={1}` | `api.setColumnPinning({ left: ['firstColId'] })` |\n| `rowMarkers=\"number\"` | `showRowNumbers={true}` prop |\n| `rowMarkers=\"checkbox\"` | `selectionMode=\"row\"` prop |\n\n## Custom drawing → custom snippets\n\nGlide's `drawCustomCell` lets you render anything via canvas. The\nsv-grid equivalent is a Svelte snippet (DOM):\n\n```diff\n- drawCustomCell(args) {\n- const ctx = args.ctx\n- ctx.fillStyle = args.cell.data.color\n- ctx.fillRect(args.rect.x, args.rect.y, args.rect.width, args.rect.height)\n- return true\n- }\n\n+ {\n+ field: 'color',\n+ cell: (ctx) => renderSnippet(ColorChip, { color: ctx.row.original.color }),\n+ }\n```\n\n```svelte\n{#snippet ColorChip(props)}\n <span style=\"display: inline-block; width: 18px; height: 18px; background: {props.color};\"></span>\n{/snippet}\n```\n\n## Out-of-memory data\n\nFor \"10M rows, fetched on scroll\" workloads (the only case Glide\ngenuinely beats sv-grid on), use sv-grid's\n[server-side / infinite-scroll pattern](./server-side-data.md):\n\n```svelte\n<SvGrid\n data={loadedSlice}\n columns={columns} features={features}\n externalSort={true}\n externalFilter={true}\n onSortingChange={async (s) => { loadedSlice = await fetchPage({ sort: s, offset: 0, limit: 200 }) }}\n/>\n```\n\nYou'll cap practical perf at ~100k rendered rows; for true 10M+\ndatasets Glide's canvas approach has a perf edge sv-grid won't\nmatch. Most apps don't need 10M rows on screen at once - they need\nfast server-side filtering, which both libraries support equally.\n\n## Theming\n\nGlide: huge `theme` prop object. Sv-grid: CSS custom properties (`--sg-*`).\n\n```diff\n- <DataEditor theme={{\n- accentColor: '#4F5DFF',\n- accentLight: 'rgba(62, 116, 253, 0.2)',\n- bgCell: '#1c1c20',\n- /* …40 more properties… */\n- }} />\n\n+ <div style=\"--sg-accent: #4F5DFF; --sg-bg: #1c1c20;\">\n+ <SvGrid {data} {columns} features={features} />\n+ </div>\n```\n\nSee [design tokens](./tokens.md) for the full list (≤ 30 tokens\ncovers everything; Glide's theme has ~40).\n\n## What you get for free vs Glide\n\n- **Accessibility.** Glide's canvas is invisible to screen readers\n unless you build an off-screen ARIA buffer. Sv-grid is WAI-ARIA\n 1.2 grid pattern out of the box; demo 17 covers axe-core CI.\n- **Standard HTML.** SEO crawlers, browser Find-in-Page, browser\n zoom, OS-level text scaling - all work because the grid emits\n actual `<table><tr><td>`.\n- **Familiar editing.** Browser-native `<input>` / `<select>` for\n cells, with autofill / clipboard / autocorrect intact.\n- **Built-in sort / filter / pagination UI.**\n\n## What you give up\n\n- **Peak FPS on >200k visible rows.** Canvas wins here. For most\n apps the gap is invisible; if you've actually measured it on real\n hardware and need the FPS, Glide remains the right call.\n- **Per-pixel cell drawing.** DOM cells are sub-pixel-uniform but\n not arbitrarily drawn.\n\n## Frequently asked questions\n\n### Should I move from Glide Data Grid to SvGrid?\n\nMove if you need accessibility, standards-compliant HTML, SEO-friendly markup,\nor DOM-based custom cells. Stay on Glide only if you have measured a need for\npeak canvas FPS on 200k+ simultaneously visible rows. For most apps the\nperformance gap is invisible, and SvGrid virtualizes comfortably to 100k+ rows.\n\n### How long does a Glide migration take?\n\nAbout 3-6 hours per grid. The biggest shift is conceptual: Glide's\n`getCellContent` / `onCellEdited` cell-data model is the inverse of SvGrid's\nplain data-array model, so the mental flip takes the most time, not the code.\n\n### Does SvGrid render to canvas or DOM?\n\nDOM, with row and column virtualization. That is what gives you WAI-ARIA roles,\nkeyboard navigation, screen-reader support, and selectable/indexable text -\nthings a canvas grid like Glide cannot offer.\n\n## What you end up with\n\nInline editing over a virtualized body, without a canvas to draw into.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection sortable />\n```\n\n## See also\n\n- [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/) - the side-by-side comparison\n- [Performance benchmarks](./benchmarks.md) - measured numbers\n- [Server-side data](./server-side-data.md) - the >100k-row pattern\n- [Accessibility](./accessibility.md) - the standards compliance Glide doesn't ship\n"
6098
+ "markdown": "# Migrating from Glide Data Grid\n\nGlide Data Grid is the canvas-rendered, ultra-high-performance choice\nfor very wide datasets (100k+ rows, 100+ columns). Sv-grid renders to\nDOM with virtualization, which trades peak FPS for\naccessibility + standards-compliant HTML.\n\n> Estimated effort: **3-6 hours** per grid. Glide's getCellContent /\n> onCellEdited cell-data model is the opposite of sv-grid's\n> data-array model; the mental flip takes the most time.\n\n<!-- facts:start glide-data-grid -->\n> **Facts, checked 9 Nov 2026.** `@glideapps/glide-data-grid` 6.0.3, MIT, last published 3 Feb 2024, 2,010,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `@glideapps/glide-data-grid` 6.0.3 71.1 KB JS + 0.2 KB CSS, lodash + marked + react + react-dom + react-responsive-carousel external (measured 9 Nov 2026). Glide Data Grid pricing, as its site states it: Glide Data Grid is MIT and, in its own words, fully free and open source; no licence is sold for the grid (https://grid.glideapps.com/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/).\n<!-- facts:end -->\n\n## Mental-model flip\n\nGlide gives you a **canvas + a `getCellContent` callback**: the grid\nasks you for each visible cell on demand. Sv-grid wants the\n**whole data array up front** (`<SvGrid data={rows}>`); the engine\nslices for virtualization.\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- function getCellContent([col, row]) {\n- const r = rows[row]\n- if (col === 0) return { kind: 'text', data: r.name, displayData: r.name, allowOverlay: true }\n- if (col === 1) return { kind: 'number', data: r.amount, displayData: fmt(r.amount), allowOverlay: true }\n- }\n- <DataEditor columns={[...]} getCellContent={getCellContent} rows={rows.length} />\n\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name', editorType: 'text' },\n+ { field: 'amount', header: 'Amount', editorType: 'number',\n+ format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ <SvGrid data={rows} columns={columns} features={features} />\n```\n\nThis change is **good** for 99% of apps: it's less code, less risk\nof typos in cell coordinates, full TypeScript type checking on the\nrow shape. For the 1% of apps with rows that don't fit in memory,\nsee \"Out-of-memory data\" below.\n\n## Cell editing\n\n```diff\n- function onCellEdited([col, row], newValue) {\n- rows[row] = { ...rows[row], [colToField(col)]: newValue.data }\n- }\n- <DataEditor onCellEdited={onCellEdited} />\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ onCellValueChange={(e) => {\n+ rows[e.rowIndex] = { ...rows[e.rowIndex], [e.columnId]: e.newValue }\n+ }}\n+ />\n```\n\nYou can also rely on sv-grid's built-in write-back if `data` is a\nSvelte 5 `$state` array - mutations propagate automatically.\n\n## Sorting + filtering\n\nGlide doesn't ship sort / filter UI; you wrap your own. Sv-grid\nships them:\n\n```diff\n- // Build your own sort buttons in the toolbar; sort rows yourself.\n\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ <SvGrid data={rows} columns={columns} features={features} filterMode=\"menu\" />\n```\n\n## Frozen columns / row markers\n\n| Glide | sv-grid |\n| -------------------------------------- | --------------------------------------------- |\n| `freezeColumns={1}` | `api.setColumnPinning({ left: ['firstColId'] })` |\n| `rowMarkers=\"number\"` | `showRowNumbers={true}` prop |\n| `rowMarkers=\"checkbox\"` | `selectionMode=\"row\"` prop |\n\n## Custom drawing → custom snippets\n\nGlide's `drawCustomCell` lets you render anything via canvas. The\nsv-grid equivalent is a Svelte snippet (DOM):\n\n```diff\n- drawCustomCell(args) {\n- const ctx = args.ctx\n- ctx.fillStyle = args.cell.data.color\n- ctx.fillRect(args.rect.x, args.rect.y, args.rect.width, args.rect.height)\n- return true\n- }\n\n+ {\n+ field: 'color',\n+ cell: (ctx) => renderSnippet(ColorChip, { color: ctx.row.original.color }),\n+ }\n```\n\n```svelte\n{#snippet ColorChip(props)}\n <span style=\"display: inline-block; width: 18px; height: 18px; background: {props.color};\"></span>\n{/snippet}\n```\n\n## Out-of-memory data\n\nFor \"10M rows, fetched on scroll\" workloads (the only case Glide\ngenuinely beats sv-grid on), use sv-grid's\n[server-side / infinite-scroll pattern](./server-side-data.md):\n\n```svelte\n<SvGrid\n data={loadedSlice}\n columns={columns} features={features}\n externalSort={true}\n externalFilter={true}\n onSortingChange={async (s) => { loadedSlice = await fetchPage({ sort: s, offset: 0, limit: 200 }) }}\n/>\n```\n\nYou'll cap practical perf at ~100k rendered rows; for true 10M+\ndatasets Glide's canvas approach has a perf edge sv-grid won't\nmatch. Most apps don't need 10M rows on screen at once - they need\nfast server-side filtering, which both libraries support equally.\n\n## Theming\n\nGlide: huge `theme` prop object. Sv-grid: CSS custom properties (`--sg-*`).\n\n```diff\n- <DataEditor theme={{\n- accentColor: '#4F5DFF',\n- accentLight: 'rgba(62, 116, 253, 0.2)',\n- bgCell: '#1c1c20',\n- /* …40 more properties… */\n- }} />\n\n+ <div style=\"--sg-accent: #4F5DFF; --sg-bg: #1c1c20;\">\n+ <SvGrid {data} {columns} features={features} />\n+ </div>\n```\n\nSee [design tokens](./tokens.md) for the full list (≤ 30 tokens\ncovers everything; Glide's theme has ~40).\n\n## What you get for free vs Glide\n\n- **Accessibility.** Glide's canvas is invisible to screen readers\n unless you build an off-screen ARIA buffer. Sv-grid is WAI-ARIA\n 1.2 grid pattern out of the box; demo 17 covers axe-core CI.\n- **Standard HTML.** SEO crawlers, browser Find-in-Page, browser\n zoom, OS-level text scaling - all work because the grid emits\n actual `<table><tr><td>`.\n- **Familiar editing.** Browser-native `<input>` / `<select>` for\n cells, with autofill / clipboard / autocorrect intact.\n- **Built-in sort / filter / pagination UI.**\n\n## What you give up\n\n- **Peak FPS on >200k visible rows.** Canvas wins here. For most\n apps the gap is invisible; if you've actually measured it on real\n hardware and need the FPS, Glide remains the right call.\n- **Per-pixel cell drawing.** DOM cells are sub-pixel-uniform but\n not arbitrarily drawn.\n\n## Frequently asked questions\n\n### Should I move from Glide Data Grid to SvGrid?\n\nMove if you need accessibility, standards-compliant HTML, SEO-friendly markup,\nor DOM-based custom cells. Stay on Glide only if you have measured a need for\npeak canvas FPS on 200k+ simultaneously visible rows. For most apps the\nperformance gap is invisible, and SvGrid virtualizes comfortably to 100k+ rows.\n\n### How long does a Glide migration take?\n\nAbout 3-6 hours per grid. The biggest shift is conceptual: Glide's\n`getCellContent` / `onCellEdited` cell-data model is the inverse of SvGrid's\nplain data-array model, so the mental flip takes the most time, not the code.\n\n### Does SvGrid render to canvas or DOM?\n\nDOM, with row and column virtualization. That is what gives you WAI-ARIA roles,\nkeyboard navigation, screen-reader support, and selectable/indexable text -\nthings a canvas grid like Glide cannot offer.\n\n## What you end up with\n\nInline editing over a virtualized body, without a canvas to draw into.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection sortable />\n```\n\n## See also\n\n- [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/) - the side-by-side comparison\n- [Performance benchmarks](./benchmarks.md) - measured numbers\n- [Server-side data](./server-side-data.md) - the >100k-row pattern\n- [Accessibility](./accessibility.md) - the standards compliance Glide doesn't ship\n"
6099
6099
  },
6100
6100
  {
6101
6101
  "slug": "help/migrating-from-gridjs",
6102
6102
  "path": "docs/help/migrating-from-gridjs.md",
6103
6103
  "title": "Migrating from Grid.js",
6104
6104
  "section": "Help",
6105
- "markdown": "# Migrating from Grid.js\n\nGrid.js is a small vanilla-JS table for search, sort, and pagination\nthat you render into a DOM element. SvGrid does the same in a\nSvelte-5-native component and keeps going - virtualization, Excel-style\nfilters, editing, grouping - when you outgrow the basics.\n\n> Estimated effort: **30 min - 1 hour** per table.\n\n<!-- facts:start gridjs -->\n> **Facts, checked 9 Nov 2026.** `gridjs` 6.2.0, MIT, last published 3 Mar 2024, 393,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `gridjs` 6.2.0 17.8 KB JS + 2.4 KB CSS (measured 9 Nov 2026). Grid.js pricing, as its site states it: Grid.js is MIT, free and open source; no licence is sold (https://gridjs.io/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Grid.js](https://svgrid.com/compare/gridjs/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Grid.js | sv-grid |\n| ----------------------------------------- | ----------------------------------------- |\n| `new Grid({ ... }).render(el)` | `<SvGrid ... />` |\n| `columns: ['Name', 'Amount']` | `columns: [{ field, header }]` |\n| `columns: [{ name, formatter }]` | `{ header, cell: (c) => renderSnippet(...) }` |\n| `sort: true` | `rowSortingFeature` |\n| `search: true` | `api.setState({ globalFilter: query })` / filter feature |\n| `pagination: { limit: 25 }` | `showPagination` (+ page size) |\n| `server: { url, then }` | `externalSort` / `externalFilter` + refetch |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import { Grid } from 'gridjs'\n- import 'gridjs/dist/theme/mermaid.css'\n-\n- new Grid({\n- data: rows,\n- columns: ['Name', 'Amount'],\n- sort: true, search: true,\n- pagination: { limit: 25 },\n- }).render(document.getElementById('grid'))\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## What you gain\n\n- **Virtualization** for large datasets (Grid.js renders the page).\n- **Excel-style filter menu**, **inline editing**, **grouping**, and\n **tree / master-detail** when you need them.\n- **Reactive data** - no `.render(el)` re-instantiation.\n- A **headless engine** and an **imperative API** plus `@svgrid/mcp`.\n\n## Frequently asked questions\n\n### When should I move from Grid.js to SvGrid?\n\nWhen you need more than search / sort / paginate - virtualization for big\ndatasets, inline editing, Excel-style filters, grouping, or tree data. For a\nsmall table, Grid.js stays lighter.\n\n### Is SvGrid MIT-licensed like Grid.js?\n\nYes. `@svgrid/grid` is MIT. Only the optional `@svgrid/enterprise` add-on is paid.\n\n### Does SvGrid support server-side data like Grid.js?\n\nYes. Set `externalSort` / `externalFilter` and refetch on the\n`onSortingChange` / `onFiltersChange` events.\n\n## What you end up with\n\nSearch, sort and pagination - the whole Grid.js surface, plus editing.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} editable />\n```\n\n## See also\n\n- [SvGrid vs Grid.js](https://svgrid.com/compare/gridjs/) - the side-by-side comparison\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n- [Server-side data](./server-side-data.md) - the external-data pattern\n"
6105
+ "markdown": "# Migrating from Grid.js\n\nGrid.js is a small vanilla-JS table for search, sort, and pagination\nthat you render into a DOM element. SvGrid does the same in a\nSvelte-5-native component and keeps going - virtualization, Excel-style\nfilters, editing, grouping - when you outgrow the basics.\n\n> Estimated effort: **30 min - 1 hour** per table.\n\n<!-- facts:start gridjs -->\n> **Facts, checked 9 Nov 2026.** `gridjs` 6.2.0, MIT, last published 3 Mar 2024, 393,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `gridjs` 6.2.0 17.8 KB JS + 2.4 KB CSS (measured 9 Nov 2026). Grid.js pricing, as its site states it: Grid.js is MIT, free and open source; no licence is sold (https://gridjs.io/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Grid.js](https://svgrid.com/compare/gridjs/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Grid.js | sv-grid |\n| ----------------------------------------- | ----------------------------------------- |\n| `new Grid({ ... }).render(el)` | `<SvGrid ... />` |\n| `columns: ['Name', 'Amount']` | `columns: [{ field, header }]` |\n| `columns: [{ name, formatter }]` | `{ header, cell: (c) => renderSnippet(...) }` |\n| `sort: true` | `rowSortingFeature` |\n| `search: true` | `api.setState({ globalFilter: query })` / filter feature |\n| `pagination: { limit: 25 }` | `showPagination` (+ page size) |\n| `server: { url, then }` | `externalSort` / `externalFilter` + refetch |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import { Grid } from 'gridjs'\n- import 'gridjs/dist/theme/mermaid.css'\n-\n- new Grid({\n- data: rows,\n- columns: ['Name', 'Amount'],\n- sort: true, search: true,\n- pagination: { limit: 25 },\n- }).render(document.getElementById('grid'))\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## What you gain\n\n- **Virtualization** for large datasets (Grid.js renders the page).\n- **Excel-style filter menu**, **inline editing**, **grouping**, and\n **tree / master-detail** when you need them.\n- **Reactive data** - no `.render(el)` re-instantiation.\n- A **headless engine** and an **imperative API** plus `@svgrid/mcp`.\n\n## Frequently asked questions\n\n### When should I move from Grid.js to SvGrid?\n\nWhen you need more than search / sort / paginate - virtualization for big\ndatasets, inline editing, Excel-style filters, grouping, or tree data. For a\nsmall table, Grid.js stays lighter.\n\n### Is SvGrid MIT-licensed like Grid.js?\n\nYes. `@svgrid/grid` is MIT. Only the optional `@svgrid/enterprise` add-on is paid.\n\n### Does SvGrid support server-side data like Grid.js?\n\nYes. Set `externalSort` / `externalFilter` and refetch on the\n`onSortingChange` / `onFiltersChange` events.\n\n## What you end up with\n\nSearch, sort and pagination - the whole Grid.js surface, plus editing.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} editable />\n```\n\n## See also\n\n- [SvGrid vs Grid.js](https://svgrid.com/compare/gridjs/) - the side-by-side comparison\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n- [Server-side data](./server-side-data.md) - the external-data pattern\n"
6106
6106
  },
6107
6107
  {
6108
6108
  "slug": "help/migrating-from-handsontable",
6109
6109
  "path": "docs/help/migrating-from-handsontable.md",
6110
6110
  "title": "Migrating from Handsontable",
6111
6111
  "section": "Help",
6112
- "markdown": "# Migrating from Handsontable\n\nHandsontable's mental model is \"spreadsheet-in-a-page\": every cell\neditable, formulas, copy/paste with the OS clipboard, frozen rows\nand columns. Sv-grid covers ~90% of the same surface; the port is\nmostly a configuration translation, except for HyperFormula (see\nthe Formulas section below).\n\n> Estimated effort: **2-4 hours** per grid for typical spreadsheet\n> workflows. **Add 1-2 days** if you depend heavily on\n> HyperFormula features sv-grid doesn't ship.\n\n<!-- facts:start handsontable -->\n> **Facts, checked 9 Nov 2026.** `handsontable` 18.1.2, Commercial (see licence file), last published 7 Oct 2026, 1,250,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). Handsontable pricing, as its site states it: Handsontable's Hobby licence is free for personal, exploratory projects and cannot be used in commercial settings; handsontable.com lists Standard from $999 per developer and Priority from $1,299 per developer, with Enterprise on custom terms (https://handsontable.com/pricing, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Handsontable](https://svgrid.com/compare/handsontable/).\n<!-- facts:end -->\n\n## Imports\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import Handsontable from 'handsontable'\n- import 'handsontable/dist/handsontable.full.min.css'\n\n+ import {\n+ SvGrid, tableFeatures,\n+ rowSortingFeature, columnFilteringFeature, rowSelectionFeature,\n+ + } from '@svgrid/grid'\n```\n\n## Schema → columns\n\nHandsontable accepts a 2D array OR an array of objects. Sv-grid\nexpects the array-of-objects form:\n\n```diff\n- const hot = new Handsontable(container, {\n- data: [\n- ['Tesla', 2017, 'black', false],\n- ['Nissan', 2018, 'blue', true ],\n- ],\n- colHeaders: ['Make', 'Year', 'Color', 'In stock'],\n- columns: [\n- {},\n- { type: 'numeric' },\n- {},\n- { type: 'checkbox' },\n- ],\n- })\n\n+ type Car = { make: string; year: number; color: string; inStock: boolean }\n+ const rows: Car[] = [\n+ { make: 'Tesla', year: 2017, color: 'black', inStock: false },\n+ { make: 'Nissan', year: 2018, color: 'blue', inStock: true },\n+ ]\n+ const columns: ColumnDef<typeof features, Car>[] = [\n+ { field: 'make', header: 'Make' },\n+ { field: 'year', header: 'Year', editorType: 'number' },\n+ { field: 'color', header: 'Color' },\n+ { field: 'inStock', header: 'In stock', editorType: 'checkbox' },\n+ ]\n+ <SvGrid data={rows} columns={columns} features={features} />\n```\n\nIf your existing data is a 2D array, transform once at boot:\n\n```ts\nconst headerRow = raw[0]\nconst rows = raw.slice(1).map((cells) =>\n Object.fromEntries(cells.map((v, i) => [headerRow[i], v])),\n)\n```\n\n## Editor types\n\n| Handsontable | sv-grid |\n| ---------------------------------------------- | ------------------------------------------------ |\n| `type: 'text'` | `editorType: 'text'` |\n| `type: 'numeric'` | `editorType: 'number'` |\n| `type: 'date'` | `editorType: 'date'` |\n| `type: 'checkbox'` | `editorType: 'checkbox'` |\n| `type: 'dropdown', source: [...]` | `editorType: 'list', editorOptions: [...]` |\n| `type: 'autocomplete'` | `editorType: 'list'` (combobox UI ships) |\n| `type: 'time'` | `editorType: 'time'` |\n| `type: 'password'` | Use `editorType: 'text'` + a custom snippet |\n\n## Selection + clipboard\n\n| Handsontable | sv-grid |\n| --------------------------------------- | ------------------------------------------------------- |\n| `selectCells([[0,0,2,5]])` | `api.selectCells(ranges)`, with `selectionMode='cell'` |\n| `getSelected()` | `api.getSelected()`: the same `[[startRow, startCol, endRow, endCol], ...]`, normalised to top-left / bottom-right and `[]` rather than `undefined` when nothing is selected |\n| `getSelectedLast()` | `api.getSelectedLast()`: `[startRow, startCol, endRow, endCol]` with the orientation kept (a range dragged upwards has `startRow > endRow`), or `undefined` |\n| `getSelectedRange()` / `getSelectedRangeLast()` | `api.getSelectedRange()` / `api.getSelectedRangeLast()`: `{ from, to, highlight }` per rectangle, the same three fields as `CellRange`; plain data rather than a class, so use `Math.min` / `Math.max` for the corners |\n| `copyPaste: true` (default) | Ships built-in; `enableCellSelection={true}`. The spreadsheet shell also exchanges formats, number formats and formulas with Excel and Google Sheets through the HTML clipboard |\n| `fillHandle: true` | Ships built-in (drag-to-fill on selected range) |\n| `moveCells: true` | `moveCells` - drag the range border to move it, Ctrl / Cmd to copy. On by default; see [move or copy a range](./editing/move-cells.md) |\n\n## Frozen rows / columns\n\n| Handsontable | sv-grid |\n| ---------------------------------- | ------------------------------------------------------ |\n| `fixedColumnsStart: 2` | `api.setColumnPinning({ left: ['firstColId', 'secondColId'] })` |\n| `fixedRowsTop: 1` | `pinnedTopRows` (and `pinnedBottomRows`) |\n| `manualColumnFreeze: true` | Ships via the column-header right-click menu |\n| `mergeCells: [{ row, col, rowspan, colspan }]` | `mergedCells={[{ rowIndex, colIndex, rowSpan, colSpan }]}` on `<SvGrid>`; the spreadsheet shell adds Merge & Center, Merge Across, Merge Cells and Unmerge on the ribbon and the cell menu |\n\nFreeze panes on a plain `<SvGrid>`: two pinned columns and the sticky\nletter and number headers, with formulas underneath.\n\n<div data-docs-demo=\"208-freeze-panes\" data-height=\"520\"></div>\n\n## Hooks → callbacks\n\n| Handsontable hook | sv-grid |\n| -------------------------------------- | ------------------------------------------------------- |\n| `afterChange((changes, src) => {})` | `onCellValueChange((e) => {...})` |\n| `afterSelection(...)` | `onActiveCellChange((cell) => {...})` |\n| `afterFilter(...)` | `onFiltersChange((f) => {...})` |\n| `afterColumnSort(...)` | `onSortingChange((s) => {...})` |\n| `beforeChange((changes) => false)` | Throw / return `false` from your inline validator; see [validation](./editing/validation.md) |\n| `afterCreateRow / afterRemoveRow` | Wrap `api.addRow / removeRow` in your own emitter; on the spreadsheet shell, `onChange` reports a `structure` reason with the edit |\n| `afterMergeCells / afterUnmergeCells` | On the spreadsheet shell, `onChange` reports `merges`; `api.getMergedCells()` reads the grid's list |\n| `afterSetCellMeta` (comments, validation) | `onChange` reports `comments`, `validation`, `conditional-formats` and `protection` on the shell |\n| `afterFilter` (the Filters plugin) | `onChange` reports `filter` on the shell; `getState().sheets[name].autoFilter` is the state |\n\nOn the spreadsheet shell (`<SvSheet>` in `@svgrid/enterprise`) one callback\ncovers the lot: `onChange(reasons)` fires once per tick with every kind of\nchange since the last call (`cells`, `formats`, `sizes`, `hidden`,\n`freeze`, `sheets`, `comments`, `protection`, `validation`,\n`conditional-formats`, `merges`, `filter`, `structure` with the insert or\ndelete, `restore`), which is where Handsontable's `afterChange` autosave\ngoes. `getState()` gives the document to save and `setState()` puts it back.\n\n## Cell meta → the sheet document\n\nHandsontable keeps comments, validators, read-only flags and merges as cell\nmeta on the instance. On the spreadsheet shell they are parts of the\ndocument, per sheet, saved with it and moved by inserts and deletes:\n\n| Handsontable | sv-grid spreadsheet shell |\n| --------------------------------------------------- | --------------------------------------------------------------- |\n| `comments: true`, `cell: [{ row, col, comment }]` | Review > New Comment, Shift+F2, the cell menu; `getState().sheets[name].comments` |\n| `validator: 'numeric'`, `allowInvalid: false` | Data > Data Validation (Whole number, Decimal, List, Date, Text length, Custom), Stop or Warning; `validation` in the state |\n| `readOnly: true` on a cell | Format Cells > Protection unlocks the cells that may change, Review > Protect Sheet locks the rest; `locked: false` in the format entry, `protected` per sheet |\n| `mergeCells` | Merge & Center on the ribbon and the cell menu; `merges` in the state |\n| `filters: true` + `dropdownMenu` | Ctrl+Shift+L or Data > Filter: Excel's AutoFilter arrows and menus; `autoFilter` in the state |\n| conditional formatting (a custom renderer) | Home > Styles > Conditional Formatting: Highlight Cells, Top/Bottom, Data Bars, Color Scales, Icon Sets, Manage Rules; `conditionalFormats` in the state |\n| `autofill` rules | The fill handle's own series: numbers, dates by day or by the set gap, `Item 1`, weekdays and months |\n\nOn a plain `<SvGrid>` the free grid has the primitives instead: `notes`\nand `editableComments` for comments, `editable` and `valueParser` on a\ncolumn for read-only cells and validation, `mergedCells` for merges, the\nfilter row and its Excel-style menu for filtering, and value-driven\n`conditionalFormats` by column.\n\n## Formulas (HyperFormula)\n\nSv-grid ships a focused subset under\n[`spreadsheet-formulas`](./spreadsheet-formulas.md) (SUM, IF, COUNTIF\n+ cell refs + ranges + cycle detection). It does NOT ship the full\nHyperFormula library.\n\nIf you need HyperFormula's full surface (~400 functions, complex\ndate math, R1C1 refs), the integration pattern is:\n\n```ts\nimport { HyperFormula } from 'hyperformula'\n\nconst hf = HyperFormula.buildFromArray(rowsAsMatrix, { licenseKey: 'gpl-v3' })\n\n// On every cell edit, push to HF + read back the computed value:\nfunction onCellValueChange(e) {\n hf.setCellContents({ sheet: 0, row: e.rowIndex, col: columnIndex }, e.newValue)\n // ...read computed cells you depend on + push back via api.setCellValue\n}\n```\n\nThis is the only piece where the swap isn't 1:1. Most teams find\nsv-grid's shipped subset covers 80% of real-world formula usage; if\nyours is in the 20%, HyperFormula plugs in with the pattern above.\n\n## Licence\n\nHandsontable's grid needs a commercial licence for commercial use; its free\nHobby licence is for personal, exploratory projects, as its pricing page\nstates (the wording and the date are in the facts box at the top of this\npage). Sv-grid's `@svgrid/grid` is MIT with no licence key, and the\n[Enterprise tier](../enterprise/licensing.md) needs no key to evaluate:\nit works without one and displays a watermark and a console\nmessage until you add one, so there is no cutoff to plan around.\n\n## What you get for free vs Handsontable\n\n- **No commercial licence for the grid.** @svgrid/grid is MIT.\n- **Modern Svelte 5 ergonomics.** `$state` arrays beat\n `loadData(...)`.\n- **Grid features a spreadsheet does not have.** Row grouping with\n aggregation, master/detail, server-side data (grouping, tree, pivot and transactions on the server are Enterprise) and integrated\n charts in the same package.\n- **CSP-clean.** Handsontable's HyperFormula path needs CSP `eval`\n exceptions in some configurations.\n\n## What you give up\n\n- **Full HyperFormula surface as a supported feature.** Sv-grid ships a\n subset and lets you wire HyperFormula yourself; see above.\n- **Nested headers.** Handsontable's `nestedHeaders` have no\n counterpart beyond column groups.\n- **Threaded comments and a validation Input Message.** Comments on the\n shell are Excel's notes, one text per cell; validation shows its message\n on a refusal, not on arrival.\n- **Excel export without a paid pack.** Handsontable's export plugin writes\n XLSX through its asynchronous method; on sv-grid, CSV, TSV and JSON export\n are free and Excel, PDF and print output are in `@svgrid/enterprise`.\n\n## Frequently asked questions\n\n### Can SvGrid replace Handsontable's spreadsheet features?\n\nIt covers roughly 90% of the surface - inline editing on every cell, copy/paste\nas TSV, range selection, column groups, and a built-in formula engine. Budget\n2-4 hours per grid for typical spreadsheet workflows, plus 1-2 days if you lean\nheavily on the full HyperFormula function set.\n\n### Does SvGrid support formulas like Handsontable?\n\nYes, a built-in formula engine with cell references, ranges, and common\nfunctions (see the spreadsheet-formulas guide and demo 27). It is a practical\nsubset rather than the complete HyperFormula library, so verify your specific\nfunctions before porting.\n\n### Is SvGrid licensed like Handsontable?\n\nNo. The `@svgrid/grid` core is MIT and free for commercial use - no per-seat\nlicence key - whereas Handsontable's free licence excludes commercial use and\nits commercial plans are per developer; the facts box has the statement as\nread from handsontable.com. The optional `@svgrid/enterprise` pack (Excel and\nPDF export, print, pivot, import) is priced per developer.\n\n### Does SvGrid have a blank spreadsheet like Handsontable?\n\nYes, on the plain grid: column-letter headers, a row gutter, a name box\nand formula bar, gridlines, range selection and a fill handle, with\nHyperFormula underneath when you want its function library.\n\n<div data-docs-demo=\"207-blank-sheet\" data-height=\"560\"></div>\n\n## What you end up with\n\nSpreadsheet behaviour: inline edit, drag a cell range, copy it out as TSV.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection statusBar />\n```\n\n## See also\n\n- [SvGrid vs Handsontable](https://svgrid.com/compare/handsontable/) - the side-by-side comparison, with a source and date for every claim\n- [Spreadsheet formulas](./spreadsheet-formulas.md) - sv-grid's\n built-in formula engine, and the rest of the spreadsheet demos\n- [Demo 27 (Spreadsheet + Ribbon)](https://svgrid.com/demos/27-spreadsheet-ribbon/) - live\n- [Migrating from AG Grid](./migrating-from-ag-grid.md)\n"
6112
+ "markdown": "# Migrating from Handsontable\n\nHandsontable's mental model is \"spreadsheet-in-a-page\": every cell\neditable, formulas, copy/paste with the OS clipboard, frozen rows\nand columns. Sv-grid covers ~90% of the same surface; the port is\nmostly a configuration translation, except for HyperFormula (see\nthe Formulas section below).\n\n> Estimated effort: **2-4 hours** per grid for typical spreadsheet\n> workflows. **Add 1-2 days** if you depend heavily on\n> HyperFormula features sv-grid doesn't ship.\n\n<!-- facts:start handsontable -->\n> **Facts, checked 9 Nov 2026.** `handsontable` 18.1.2, Commercial (see licence file), last published 7 Oct 2026, 1,250,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). Handsontable pricing, as its site states it: Handsontable's Hobby licence is free for personal, exploratory projects and cannot be used in commercial settings; handsontable.com lists Standard from $999 per developer and Priority from $1,299 per developer, with Enterprise on custom terms (https://handsontable.com/pricing, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Handsontable](https://svgrid.com/compare/handsontable/).\n<!-- facts:end -->\n\n## Imports\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import Handsontable from 'handsontable'\n- import 'handsontable/dist/handsontable.full.min.css'\n\n+ import {\n+ SvGrid, tableFeatures,\n+ rowSortingFeature, columnFilteringFeature, rowSelectionFeature,\n+ + } from '@svgrid/grid'\n```\n\n## Schema → columns\n\nHandsontable accepts a 2D array OR an array of objects. Sv-grid\nexpects the array-of-objects form:\n\n```diff\n- const hot = new Handsontable(container, {\n- data: [\n- ['Tesla', 2017, 'black', false],\n- ['Nissan', 2018, 'blue', true ],\n- ],\n- colHeaders: ['Make', 'Year', 'Color', 'In stock'],\n- columns: [\n- {},\n- { type: 'numeric' },\n- {},\n- { type: 'checkbox' },\n- ],\n- })\n\n+ type Car = { make: string; year: number; color: string; inStock: boolean }\n+ const rows: Car[] = [\n+ { make: 'Tesla', year: 2017, color: 'black', inStock: false },\n+ { make: 'Nissan', year: 2018, color: 'blue', inStock: true },\n+ ]\n+ const columns: ColumnDef<typeof features, Car>[] = [\n+ { field: 'make', header: 'Make' },\n+ { field: 'year', header: 'Year', editorType: 'number' },\n+ { field: 'color', header: 'Color' },\n+ { field: 'inStock', header: 'In stock', editorType: 'checkbox' },\n+ ]\n+ <SvGrid data={rows} columns={columns} features={features} />\n```\n\nIf your existing data is a 2D array, transform once at boot:\n\n```ts\nconst headerRow = raw[0]\nconst rows = raw.slice(1).map((cells) =>\n Object.fromEntries(cells.map((v, i) => [headerRow[i], v])),\n)\n```\n\n## Editor types\n\n| Handsontable | sv-grid |\n| ---------------------------------------------- | ------------------------------------------------ |\n| `type: 'text'` | `editorType: 'text'` |\n| `type: 'numeric'` | `editorType: 'number'` |\n| `type: 'date'` | `editorType: 'date'` |\n| `type: 'checkbox'` | `editorType: 'checkbox'` |\n| `type: 'dropdown', source: [...]` | `editorType: 'list', editorOptions: [...]` |\n| `type: 'autocomplete'` | `editorType: 'list'` (combobox UI ships) |\n| `type: 'time'` | `editorType: 'time'` |\n| `type: 'password'` | Use `editorType: 'text'` + a custom snippet |\n\n## Selection + clipboard\n\n| Handsontable | sv-grid |\n| --------------------------------------- | ------------------------------------------------------- |\n| `selectCells([[0,0,2,5]])` | `api.selectCells(ranges)`, with `selectionMode='cell'` |\n| `getSelected()` | `api.getSelected()`: the same `[[startRow, startCol, endRow, endCol], ...]`, normalised to top-left / bottom-right and `[]` rather than `undefined` when nothing is selected |\n| `getSelectedLast()` | `api.getSelectedLast()`: `[startRow, startCol, endRow, endCol]` with the orientation kept (a range dragged upwards has `startRow > endRow`), or `undefined` |\n| `getSelectedRange()` / `getSelectedRangeLast()` | `api.getSelectedRange()` / `api.getSelectedRangeLast()`: `{ from, to, highlight }` per rectangle, the same three fields as `CellRange`; plain data rather than a class, so use `Math.min` / `Math.max` for the corners |\n| `copyPaste: true` (default) | Ships built-in; `enableCellSelection={true}`. The spreadsheet shell also exchanges formats, number formats and formulas with Excel and Google Sheets through the HTML clipboard |\n| `fillHandle: true` | Ships built-in (drag-to-fill on selected range) |\n| `moveCells: true` | `moveCells` - drag the range border to move it, Ctrl / Cmd to copy. On by default; see [move or copy a range](./editing/move-cells.md) |\n\n## Frozen rows / columns\n\n| Handsontable | sv-grid |\n| ---------------------------------- | ------------------------------------------------------ |\n| `fixedColumnsStart: 2` | `api.setColumnPinning({ left: ['firstColId', 'secondColId'] })` |\n| `fixedRowsTop: 1` | `pinnedTopRows` (and `pinnedBottomRows`) |\n| `manualColumnFreeze: true` | Ships via the column-header right-click menu |\n| `mergeCells: [{ row, col, rowspan, colspan }]` | `mergedCells={[{ rowIndex, colIndex, rowSpan, colSpan }]}` on `<SvGrid>`; the spreadsheet shell adds Merge & Center, Merge Across, Merge Cells and Unmerge on the ribbon and the cell menu |\n\nFreeze panes on a plain `<SvGrid>`: two pinned columns and the sticky\nletter and number headers, with formulas underneath.\n\n<div data-docs-demo=\"208-freeze-panes\" data-height=\"520\"></div>\n\n## Hooks → callbacks\n\n| Handsontable hook | sv-grid |\n| -------------------------------------- | ------------------------------------------------------- |\n| `afterChange((changes, src) => {})` | `onCellValueChange((e) => {...})` |\n| `afterSelection(...)` | `onActiveCellChange((cell) => {...})` |\n| `afterFilter(...)` | `onFiltersChange((f) => {...})` |\n| `afterColumnSort(...)` | `onSortingChange((s) => {...})` |\n| `beforeChange((changes) => false)` | Throw / return `false` from your inline validator; see [validation](./editing/validation.md) |\n| `afterCreateRow / afterRemoveRow` | Wrap `api.addRow / removeRow` in your own emitter; on the spreadsheet shell, `onChange` reports a `structure` reason with the edit |\n| `afterMergeCells / afterUnmergeCells` | On the spreadsheet shell, `onChange` reports `merges`; `api.getMergedCells()` reads the grid's list |\n| `afterSetCellMeta` (comments, validation) | `onChange` reports `comments`, `validation`, `conditional-formats` and `protection` on the shell |\n| `afterFilter` (the Filters plugin) | `onChange` reports `filter` on the shell; `getState().sheets[name].autoFilter` is the state |\n\nOn the spreadsheet shell (`<SvSheet>` in `@svgrid/enterprise`) one callback\ncovers the lot: `onChange(reasons)` fires once per tick with every kind of\nchange since the last call (`cells`, `formats`, `sizes`, `hidden`,\n`freeze`, `sheets`, `comments`, `protection`, `validation`,\n`conditional-formats`, `merges`, `filter`, `structure` with the insert or\ndelete, `restore`), which is where Handsontable's `afterChange` autosave\ngoes. `getState()` gives the document to save and `setState()` puts it back.\n\n## Cell meta → the sheet document\n\nHandsontable keeps comments, validators, read-only flags and merges as cell\nmeta on the instance. On the spreadsheet shell they are parts of the\ndocument, per sheet, saved with it and moved by inserts and deletes:\n\n| Handsontable | sv-grid spreadsheet shell |\n| --------------------------------------------------- | --------------------------------------------------------------- |\n| `comments: true`, `cell: [{ row, col, comment }]` | Review > New Comment, Shift+F2, the cell menu; `getState().sheets[name].comments` |\n| `validator: 'numeric'`, `allowInvalid: false` | Data > Data Validation (Whole number, Decimal, List, Date, Text length, Custom), Stop or Warning; `validation` in the state |\n| `readOnly: true` on a cell | Format Cells > Protection unlocks the cells that may change, Review > Protect Sheet locks the rest; `locked: false` in the format entry, `protected` per sheet |\n| `mergeCells` | Merge & Center on the ribbon and the cell menu; `merges` in the state |\n| `filters: true` + `dropdownMenu` | Ctrl+Shift+L or Data > Filter: Excel's AutoFilter arrows and menus; `autoFilter` in the state |\n| conditional formatting (a custom renderer) | Home > Styles > Conditional Formatting: Highlight Cells, Top/Bottom, Data Bars, Color Scales, Icon Sets, Manage Rules; `conditionalFormats` in the state |\n| `autofill` rules | The fill handle's own series: numbers, dates by day or by the set gap, `Item 1`, weekdays and months |\n\nOn a plain `<SvGrid>` the free grid has the primitives instead: `notes`\nand `editableComments` for comments, `editable` and `valueParser` on a\ncolumn for read-only cells and validation, `mergedCells` for merges, the\nfilter row and its Excel-style menu for filtering, and value-driven\n`conditionalFormats` by column.\n\n## Formulas (HyperFormula)\n\nSv-grid ships a focused subset under\n[`spreadsheet-formulas`](./spreadsheet-formulas.md) (SUM, IF, COUNTIF\n+ cell refs + ranges + cycle detection). It does NOT ship the full\nHyperFormula library.\n\nIf you need HyperFormula's full surface (~400 functions, complex\ndate math, R1C1 refs), the integration pattern is:\n\n```ts\nimport { HyperFormula } from 'hyperformula'\n\nconst hf = HyperFormula.buildFromArray(rowsAsMatrix, { licenseKey: 'gpl-v3' })\n\n// On every cell edit, push to HF + read back the computed value:\nfunction onCellValueChange(e) {\n hf.setCellContents({ sheet: 0, row: e.rowIndex, col: columnIndex }, e.newValue)\n // ...read computed cells you depend on + push back via api.setCellValue\n}\n```\n\nThis is the only piece where the swap isn't 1:1. Most teams find\nsv-grid's shipped subset covers 80% of real-world formula usage; if\nyours is in the 20%, HyperFormula plugs in with the pattern above.\n\n## Licence\n\nHandsontable's grid needs a commercial licence for commercial use; its free\nHobby licence is for personal, exploratory projects, as its pricing page\nstates (the wording and the date are in the facts box at the top of this\npage). Sv-grid's `@svgrid/grid` is MIT with no licence key, and the\n[Enterprise tier](../enterprise/licensing.md) needs no key to evaluate:\nit works without one and displays a watermark and a console\nmessage until you add one, so there is no cutoff to plan around.\n\n## What you get for free vs Handsontable\n\n- **No commercial licence for the grid.** @svgrid/grid is MIT.\n- **Modern Svelte 5 ergonomics.** `$state` arrays beat\n `loadData(...)`.\n- **Grid features a spreadsheet does not have.** Row grouping with\n aggregation, master/detail, server-side data (grouping, tree, pivot and transactions on the server are Enterprise) and integrated\n charts in the same package.\n- **CSP-clean.** Handsontable's HyperFormula path needs CSP `eval`\n exceptions in some configurations.\n\n## What you give up\n\n- **Full HyperFormula surface as a supported feature.** Sv-grid ships a\n subset and lets you wire HyperFormula yourself; see above.\n- **Nested headers.** Handsontable's `nestedHeaders` have no\n counterpart beyond column groups.\n- **Threaded comments and a validation Input Message.** Comments on the\n shell are Excel's notes, one text per cell; validation shows its message\n on a refusal, not on arrival.\n- **Excel export without a paid pack.** Handsontable's export plugin writes\n XLSX through its asynchronous method; on sv-grid, CSV, TSV and JSON export\n are free and Excel, PDF and print output are in `@svgrid/enterprise`.\n\n## Frequently asked questions\n\n### Can SvGrid replace Handsontable's spreadsheet features?\n\nIt covers roughly 90% of the surface - inline editing on every cell, copy/paste\nas TSV, range selection, column groups, and a built-in formula engine. Budget\n2-4 hours per grid for typical spreadsheet workflows, plus 1-2 days if you lean\nheavily on the full HyperFormula function set.\n\n### Does SvGrid support formulas like Handsontable?\n\nYes, a built-in formula engine with cell references, ranges, and common\nfunctions (see the spreadsheet-formulas guide and demo 27). It is a practical\nsubset rather than the complete HyperFormula library, so verify your specific\nfunctions before porting.\n\n### Is SvGrid licensed like Handsontable?\n\nNo. The `@svgrid/grid` core is MIT and free for commercial use - no per-seat\nlicence key - whereas Handsontable's free licence excludes commercial use and\nits commercial plans are per developer; the facts box has the statement as\nread from handsontable.com. The optional `@svgrid/enterprise` pack (Excel and\nPDF export, print, pivot, import) is priced per developer.\n\n### Does SvGrid have a blank spreadsheet like Handsontable?\n\nYes, on the plain grid: column-letter headers, a row gutter, a name box\nand formula bar, gridlines, range selection and a fill handle, with\nHyperFormula underneath when you want its function library.\n\n<div data-docs-demo=\"207-blank-sheet\" data-height=\"560\"></div>\n\n## What you end up with\n\nSpreadsheet behaviour: inline edit, drag a cell range, copy it out as TSV.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection statusBar />\n```\n\n## See also\n\n- [SvGrid vs Handsontable](https://svgrid.com/compare/handsontable/) - the side-by-side comparison, with a source and date for every claim\n- [Spreadsheet formulas](./spreadsheet-formulas.md) - sv-grid's\n built-in formula engine, and the rest of the spreadsheet demos\n- [Demo 27 (Spreadsheet + Ribbon)](https://svgrid.com/demos/27-spreadsheet-ribbon/) - live\n- [Migrating from AG Grid](./migrating-from-ag-grid.md)\n"
6113
6113
  },
6114
6114
  {
6115
6115
  "slug": "help/migrating-from-jqxgrid",
6116
6116
  "path": "docs/help/migrating-from-jqxgrid.md",
6117
6117
  "title": "Migrating from jqxGrid (jQWidgets)",
6118
6118
  "section": "Help",
6119
- "markdown": "# Migrating from jqxGrid (jQWidgets)\n\njqxGrid and SvGrid come from the **same team**. jqxGrid is the mature,\nmulti-framework grid (jQuery core with Angular, React, and Vue\nintegrations); SvGrid is the Svelte-5-native member of the family. If\nyour team is consolidating on Svelte 5, this guide maps a jqxGrid setup\nonto SvGrid - and the concepts will feel familiar because the\nengineering lineage is shared.\n\n> Estimated effort: **2-4 hours** per grid. Most of the work is moving\n> from the imperative jQuery source / API to reactive `$state`.\n\n<!-- facts:start jqxgrid -->\n> **Facts, checked 9 Nov 2026.** `jqwidgets-scripts` 27.2.0, https://www.jqwidgets.com/license/, last published 31 Oct 2026, 50,500 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). jqxGrid pricing, as its site states it: jqwidgets.com lists the Developer License at $499 (on sale at $399) per developer, perpetual and royalty-free with one year of updates and support, and the Team License at $1,499 for five or more developers; the free Community License excludes the Grid (https://www.jqwidgets.com/license/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs jqxGrid (jQWidgets)](https://svgrid.com/compare/jqxgrid/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| jqxGrid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `$('#grid').jqxGrid({ ... })` | `<SvGrid ... />` |\n| `source` / `jqxDataAdapter` | `data` array (or `externalSort/Filter`) |\n| `{ text: 'Name', datafield: 'name' }` | `{ header: 'Name', field: 'name' }` |\n| `cellsrenderer` | `cell: (c) => renderSnippet(...)` |\n| `columntype` / cell editor | `editorType` |\n| `cellsformat: 'c2'` | `format: { type: 'currency', currency: 'USD' }` |\n| `sortable` / `filterable` | `rowSortingFeature` / `columnFilteringFeature` |\n| `groupable` + `groups` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `pageable` | `showPagination` |\n| `jqxGrid('updatebounddata')` | reassign the `$state` data array |\n| `editable` + `cellendedit` event | `enableInlineEditing` + `onCellValueChange` |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- const source = { localdata: rows, datatype: 'array' }\n- const adapter = new $.jqx.dataAdapter(source)\n- $('#grid').jqxGrid({\n- source: adapter, sortable: true, filterable: true, pageable: true, editable: true,\n- columns: [\n- { text: 'Name', datafield: 'name' },\n- { text: 'Amount', datafield: 'amount', cellsformat: 'c2' },\n- ],\n- })\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination enableInlineEditing />\n```\n\n## What changes\n\n- **jQuery source / dataAdapter → reactive `$state`.** No\n `updatebounddata()`; reassign the array.\n- **`cellsrenderer` → snippet cells** via `renderSnippet`.\n- **jQWidgets theme → `--sg-*` tokens** (or your Tailwind theme).\n\n## A note on family\n\nMoving to SvGrid keeps you with the same team's Svelte-5 grid rather\nthan switching vendors. The Enterprise tier and support come from the same people behind jqxGrid.\nIf your app is still on jQuery / Angular / React / Vue, jqxGrid remains\nthe right tool; reach for SvGrid when the codebase is Svelte 5.\n\n## Frequently asked questions\n\n### Is SvGrid made by the same team as jqxGrid?\n\nYes. SvGrid is built by jQWidgets - the same team behind jqxGrid - specifically\nfor Svelte 5, rebuilt around runes and snippets with an MIT community core.\n\n### How do jqxGrid columns map to SvGrid?\n\n`text` becomes `header`, `datafield` becomes `field`, `cellsrenderer` becomes a\n`renderSnippet` cell, and `cellsformat` maps onto the `format` config. Sorting /\nfiltering / grouping move to registered features.\n\n### Should I migrate my jQuery jqxGrid app to SvGrid?\n\nOnly if you are moving that app to Svelte 5. If it stays on jQuery / Angular /\nReact / Vue, jqxGrid is the better fit. SvGrid is the native choice once the\nstack is Svelte.\n\n## What you end up with\n\nEditing plus range selection and a totals row - the jqxGrid working set.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection summary sortable />\n```\n\n## See also\n\n- [SvGrid vs jqxGrid](https://svgrid.com/compare/jqxgrid/) - the side-by-side comparison\n- [About SvGrid](https://svgrid.com/about/) - built by jQWidgets\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n"
6119
+ "markdown": "# Migrating from jqxGrid (jQWidgets)\n\njqxGrid and SvGrid come from the **same team**. jqxGrid is the mature,\nmulti-framework grid (jQuery core with Angular, React, and Vue\nintegrations); SvGrid is the Svelte-5-native member of the family. If\nyour team is consolidating on Svelte 5, this guide maps a jqxGrid setup\nonto SvGrid - and the concepts will feel familiar because the\nengineering lineage is shared.\n\n> Estimated effort: **2-4 hours** per grid. Most of the work is moving\n> from the imperative jQuery source / API to reactive `$state`.\n\n<!-- facts:start jqxgrid -->\n> **Facts, checked 9 Nov 2026.** `jqwidgets-scripts` 27.2.0, https://www.jqwidgets.com/license/, last published 31 Oct 2026, 50,500 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). jqxGrid pricing, as its site states it: jqwidgets.com lists the Developer License at $499 (on sale at $399) per developer, perpetual and royalty-free with one year of updates and support, and the Team License at $1,499 for five or more developers; the free Community License excludes the Grid (https://www.jqwidgets.com/license/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs jqxGrid (jQWidgets)](https://svgrid.com/compare/jqxgrid/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| jqxGrid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `$('#grid').jqxGrid({ ... })` | `<SvGrid ... />` |\n| `source` / `jqxDataAdapter` | `data` array (or `externalSort/Filter`) |\n| `{ text: 'Name', datafield: 'name' }` | `{ header: 'Name', field: 'name' }` |\n| `cellsrenderer` | `cell: (c) => renderSnippet(...)` |\n| `columntype` / cell editor | `editorType` |\n| `cellsformat: 'c2'` | `format: { type: 'currency', currency: 'USD' }` |\n| `sortable` / `filterable` | `rowSortingFeature` / `columnFilteringFeature` |\n| `groupable` + `groups` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `pageable` | `showPagination` |\n| `jqxGrid('updatebounddata')` | reassign the `$state` data array |\n| `editable` + `cellendedit` event | `enableInlineEditing` + `onCellValueChange` |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- const source = { localdata: rows, datatype: 'array' }\n- const adapter = new $.jqx.dataAdapter(source)\n- $('#grid').jqxGrid({\n- source: adapter, sortable: true, filterable: true, pageable: true, editable: true,\n- columns: [\n- { text: 'Name', datafield: 'name' },\n- { text: 'Amount', datafield: 'amount', cellsformat: 'c2' },\n- ],\n- })\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination enableInlineEditing />\n```\n\n## What changes\n\n- **jQuery source / dataAdapter → reactive `$state`.** No\n `updatebounddata()`; reassign the array.\n- **`cellsrenderer` → snippet cells** via `renderSnippet`.\n- **jQWidgets theme → `--sg-*` tokens** (or your Tailwind theme).\n\n## A note on family\n\nMoving to SvGrid keeps you with the same team's Svelte-5 grid rather\nthan switching vendors. The Enterprise tier and support come from the same people behind jqxGrid.\nIf your app is still on jQuery / Angular / React / Vue, jqxGrid remains\nthe right tool; reach for SvGrid when the codebase is Svelte 5.\n\n## Frequently asked questions\n\n### Is SvGrid made by the same team as jqxGrid?\n\nYes. SvGrid is built by jQWidgets - the same team behind jqxGrid - specifically\nfor Svelte 5, rebuilt around runes and snippets with an MIT community core.\n\n### How do jqxGrid columns map to SvGrid?\n\n`text` becomes `header`, `datafield` becomes `field`, `cellsrenderer` becomes a\n`renderSnippet` cell, and `cellsformat` maps onto the `format` config. Sorting /\nfiltering / grouping move to registered features.\n\n### Should I migrate my jQuery jqxGrid app to SvGrid?\n\nOnly if you are moving that app to Svelte 5. If it stays on jQuery / Angular /\nReact / Vue, jqxGrid is the better fit. SvGrid is the native choice once the\nstack is Svelte.\n\n## What you end up with\n\nEditing plus range selection and a totals row - the jqxGrid working set.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection summary sortable />\n```\n\n## See also\n\n- [SvGrid vs jqxGrid](https://svgrid.com/compare/jqxgrid/) - the side-by-side comparison\n- [About SvGrid](https://svgrid.com/about/) - built by jQWidgets\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n"
6120
6120
  },
6121
6121
  {
6122
6122
  "slug": "help/migrating-from-kendo-grid",
6123
6123
  "path": "docs/help/migrating-from-kendo-grid.md",
6124
6124
  "title": "Migrating from Kendo UI Grid (Telerik)",
6125
6125
  "section": "Help",
6126
- "markdown": "# Migrating from Kendo UI Grid (Telerik)\n\nKendo UI Grid is the polished, commercial enterprise grid from Telerik /\nProgress, available for jQuery, React, Vue, and Angular. Moving to SvGrid\ntrades the multi-framework commercial suite for a Svelte-5-native grid\nwith an MIT core - and a much cheaper paid tier.\n\n> Estimated effort: **3-6 hours** per grid - column / event translation\n> plus re-theming.\n>\n> Kendo's exact API differs across its jQuery / React / Vue / Angular\n> flavours; map your version's column and event names onto the SvGrid\n> equivalents below.\n\n<!-- facts:start kendo-ui-grid -->\n> **Facts, checked 9 Nov 2026.** `@progress/kendo-react-grid` 16.3.0, Commercial (see licence file), last published 2 Nov 2026, 406,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). Kendo UI Grid pricing, as its site states it: telerik.com lists the Kendo UI bundle at $799, $849 or $1,299 per developer per year depending on the support tier, and KendoReact alone from $649 per developer per year; no free tier for the grid (https://www.telerik.com/purchase.aspx, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Kendo UI Grid (Telerik)](https://svgrid.com/compare/kendo-ui-grid/).\n<!-- facts:end -->\n\n## Concept map\n\n| Kendo UI Grid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `columns: [{ field, title }]` | `columns: [{ field, header }]` |\n| `template` / cell render | `cell: (c) => renderSnippet(...)` |\n| `editable` + editor | `enableInlineEditing` + `editorType` |\n| `sortable` / `filterable` | `rowSortingFeature` / `columnFilteringFeature` |\n| `group` / aggregates | `columnGroupingFeature` + `api.setGroupBy()` |\n| `dataSource` (local / remote) | `data` array or `externalSort/Filter` |\n| `pageable` | `showPagination` |\n| Kendo themes (Default / Bootstrap) | `--sg-*` tokens / Tailwind |\n| Grid events (`change`, `dataBound`) | `onCellValueChange`, `onApiReady`, ... |\n\n## Shape of the change\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- $('#grid').kendoGrid({\n- dataSource: { data: rows, pageSize: 25 },\n- sortable: true, filterable: true, pageable: true, editable: true,\n- columns: [\n- { field: 'name', title: 'Name' },\n- { field: 'amount', title: 'Amount', format: '{0:c}' },\n- ],\n- })\n\n+ <SvGrid\n+ data={rows}\n+ columns={[\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]}\n+ features={tableFeatures({ rowSortingFeature, columnFilteringFeature })}\n+ showPagination enableInlineEditing />\n```\n\n## Why teams switch\n\n- **Svelte 5 native** instead of a jQuery / multi-framework suite.\n- **MIT community core** - no per-developer license for the grid itself.\n- **Paid support at a fraction of enterprise-suite pricing.**\n- **No design-system lock-in** - theme via `--sg-*` tokens.\n\n## What to weigh\n\n- Kendo bundles a **huge component suite** (charts, scheduler, editor)\n and decades of vendor support. If you need the whole suite across\n several frameworks, that has real value.\n\n## Frequently asked questions\n\n### Is SvGrid a cheaper alternative to Kendo UI Grid?\n\nFor a Svelte stack, yes: the `@svgrid/grid` core is MIT (free, including\ncommercial use), and the paid `@svgrid/enterprise` tier is a fraction of enterprise\ncomponent-suite pricing.\n\n### How long does a Kendo Grid migration take?\n\nAbout 3-6 hours per grid. It is a column / event rename pass plus re-theming;\nthe columns and editing concepts map closely.\n\n### Does SvGrid match Kendo's feature set?\n\nFor the grid itself it covers the common enterprise surface - sorting,\nExcel-style filters, grouping, virtualization, editing, master/detail, tree.\nKendo's value beyond that is the wider suite and multi-framework parity.\n\n## What you end up with\n\nSorting, filtering, grouping and editing, with the column menu Kendo users expect.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} groupable sortable filterable editable />\n```\n\n## See also\n\n- [SvGrid vs Kendo UI Grid](https://svgrid.com/compare/kendo-ui-grid/) - the side-by-side comparison\n- [Pricing](https://svgrid.com/pricing/) - the SvGrid Enterprise tiers\n- [Architecture](./architecture.md) - engine + render-component split\n"
6126
+ "markdown": "# Migrating from Kendo UI Grid (Telerik)\n\nKendo UI Grid is the polished, commercial enterprise grid from Telerik /\nProgress, available for jQuery, React, Vue, and Angular. Moving to SvGrid\ntrades the multi-framework commercial suite for a Svelte-5-native grid\nwith an MIT core - and a much cheaper paid tier.\n\n> Estimated effort: **3-6 hours** per grid - column / event translation\n> plus re-theming.\n>\n> Kendo's exact API differs across its jQuery / React / Vue / Angular\n> flavours; map your version's column and event names onto the SvGrid\n> equivalents below.\n\n<!-- facts:start kendo-ui-grid -->\n> **Facts, checked 9 Nov 2026.** `@progress/kendo-react-grid` 16.3.0, Commercial (see licence file), last published 2 Nov 2026, 406,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). Kendo UI Grid pricing, as its site states it: telerik.com lists the Kendo UI bundle at $799, $849 or $1,299 per developer per year depending on the support tier, and KendoReact alone from $649 per developer per year; no free tier for the grid (https://www.telerik.com/purchase.aspx, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Kendo UI Grid (Telerik)](https://svgrid.com/compare/kendo-ui-grid/).\n<!-- facts:end -->\n\n## Concept map\n\n| Kendo UI Grid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `columns: [{ field, title }]` | `columns: [{ field, header }]` |\n| `template` / cell render | `cell: (c) => renderSnippet(...)` |\n| `editable` + editor | `enableInlineEditing` + `editorType` |\n| `sortable` / `filterable` | `rowSortingFeature` / `columnFilteringFeature` |\n| `group` / aggregates | `columnGroupingFeature` + `api.setGroupBy()` |\n| `dataSource` (local / remote) | `data` array or `externalSort/Filter` |\n| `pageable` | `showPagination` |\n| Kendo themes (Default / Bootstrap) | `--sg-*` tokens / Tailwind |\n| Grid events (`change`, `dataBound`) | `onCellValueChange`, `onApiReady`, ... |\n\n## Shape of the change\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- $('#grid').kendoGrid({\n- dataSource: { data: rows, pageSize: 25 },\n- sortable: true, filterable: true, pageable: true, editable: true,\n- columns: [\n- { field: 'name', title: 'Name' },\n- { field: 'amount', title: 'Amount', format: '{0:c}' },\n- ],\n- })\n\n+ <SvGrid\n+ data={rows}\n+ columns={[\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]}\n+ features={tableFeatures({ rowSortingFeature, columnFilteringFeature })}\n+ showPagination enableInlineEditing />\n```\n\n## Why teams switch\n\n- **Svelte 5 native** instead of a jQuery / multi-framework suite.\n- **MIT community core** - no per-developer license for the grid itself.\n- **Paid support at a fraction of enterprise-suite pricing.**\n- **No design-system lock-in** - theme via `--sg-*` tokens.\n\n## What to weigh\n\n- Kendo bundles a **huge component suite** (charts, scheduler, editor)\n and decades of vendor support. If you need the whole suite across\n several frameworks, that has real value.\n\n## Frequently asked questions\n\n### Is SvGrid a cheaper alternative to Kendo UI Grid?\n\nFor a Svelte stack, yes: the `@svgrid/grid` core is MIT (free, including\ncommercial use), and the paid `@svgrid/enterprise` tier is a fraction of enterprise\ncomponent-suite pricing.\n\n### How long does a Kendo Grid migration take?\n\nAbout 3-6 hours per grid. It is a column / event rename pass plus re-theming;\nthe columns and editing concepts map closely.\n\n### Does SvGrid match Kendo's feature set?\n\nFor the grid itself it covers the common enterprise surface - sorting,\nExcel-style filters, grouping, virtualization, editing, master/detail, tree.\nKendo's value beyond that is the wider suite and multi-framework parity.\n\n## What you end up with\n\nSorting, filtering, grouping and editing, with the column menu Kendo users expect.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} groupable sortable filterable editable />\n```\n\n## See also\n\n- [SvGrid vs Kendo UI Grid](https://svgrid.com/compare/kendo-ui-grid/) - the side-by-side comparison\n- [Pricing](https://svgrid.com/pricing/) - the SvGrid Enterprise tiers\n- [Architecture](./architecture.md) - engine + render-component split\n"
6127
6127
  },
6128
6128
  {
6129
6129
  "slug": "help/migrating-from-mui-x",
6130
6130
  "path": "docs/help/migrating-from-mui-x.md",
6131
6131
  "title": "Migrating from MUI X DataGrid",
6132
6132
  "section": "Help",
6133
- "markdown": "# Migrating from MUI X DataGrid\n\nMUI X DataGrid is the most common starting point for teams already\non Material UI. Its MIT Community edition plus paid Pro and Premium\ntiers is a split similar to sv-grid's, except that column pinning,\nmaster/detail, tree data, row grouping, aggregation and cell selection\nare paid in MUI X and free here. The port is mostly mechanical.\n\n> Estimated effort: **1-3 hours** per grid, depending on how heavily\n> you've leant on `apiRef.current.*` calls.\n\n<!-- facts:start mui-x-datagrid -->\n> **Facts, checked 9 Nov 2026.** `@mui/x-data-grid` 9.17.0, MIT, last published 30 Oct 2026, 13,100,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `@mui/x-data-grid` 9.17.0 158.4 KB JS, no separate stylesheet, @emotion/react + @emotion/styled + @mui/material + @mui/system + react + react-dom external (measured 9 Nov 2026). MUI X DataGrid pricing, as its site states it: MUI X Community is free under MIT. mui.com lists Pro at $299 per year per developer, Premium at $599 per year per developer and Enterprise at $1,399 per year per developer, with perpetual and annual licence models (https://mui.com/pricing/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/).\n<!-- facts:end -->\n\n## Package map\n\n| MUI X | sv-grid |\n| ------------------------------------ | ---------------------------------------- |\n| `@mui/x-data-grid` | `@svgrid/grid` |\n| `@mui/x-data-grid-pro` | `@svgrid/enterprise` (export, import, pivot, AI) |\n| `@mui/x-data-grid-premium` | All of `@svgrid/enterprise` ships in one tier |\n\n## Imports\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import { DataGrid, GridColDef, GridRowsProp } from '@mui/x-data-grid'\n\n+ import { SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef } from '@svgrid/grid'\n```\n\n## Column defs\n\n```diff\n- const columns: GridColDef[] = [\n- { field: 'id', headerName: 'ID', width: 90 },\n- { field: 'lastName', headerName: 'Last', width: 150, editable: true },\n- { field: 'age', headerName: 'Age', type: 'number', width: 110, editable: true },\n- ]\n\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'id', header: 'ID', width: 90 },\n+ { field: 'lastName', header: 'Last', width: 150 },\n+ { field: 'age', header: 'Age', width: 110, editorType: 'number' },\n+ ]\n```\n\n- `headerName` → `header`\n- `type: 'number'` → `editorType: 'number'`\n- `editable: true` → omit (every column is editable by default; set\n `editable: false` to OPT OUT)\n\n## Mounting\n\n```diff\n- <DataGrid rows={rows} columns={columns} pageSize={25} checkboxSelection />\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ showPagination={true} pageSize={25}\n+ selectionMode=\"row\" />\n```\n\n`tableFeatures({ rowSortingFeature, columnFilteringFeature,\nrowSelectionFeature })` registers what you'd implicitly get from\nMUI X.\n\n## `apiRef` translation\n\n| MUI X (`apiRef.current.*`) | sv-grid (`api.*`) |\n| -------------------------------------------- | ------------------------------------------ |\n| `setRows(rows)` | mutate the `$state` array you bound to `data` |\n| `updateRows([{id, ...patch}])` | `api.setCellValue(rowIndex, field, value)` |\n| `setSortModel([{field, sort}])` | `api.setSort(field, sort)` |\n| `setFilterModel({items: [...]})` | `api.setFilter(field, {operator, value})` |\n| `setPage(0)` / `setPageSize(50)` | Use the built-in pager; for headless control register `pageSize` prop |\n| `selectRow(id)` | Toggle the row checkbox via the wrapper's UI; programmatic select is `api.selectRows([id])` (`api.toggleRowSelected(id)` to flip one) |\n| `getSelectedRows()` | `api.getSelectedRows()` (ids via `api.getSelectedRowIds()`) |\n| `exportDataAsExcel()` | `api.exportData({ format: 'xlsx' })` (Enterprise) |\n| `setColumnVisibilityModel({field: false})` | `api.setColumnVisible('field', false)` |\n\n## Custom cells\n\n```diff\n- {\n- field: 'status',\n- renderCell: ({ row }) => <Chip label={row.status} color={row.status === 'active' ? 'success' : 'default'} />,\n- }\n\n+ {\n+ field: 'status',\n+ cell: (ctx) => renderSnippet(StatusChip, { status: ctx.row.original.status }),\n+ }\n```\n\n## Selection\n\n```diff\n- <DataGrid checkboxSelection\n- onRowSelectionModelChange={(ids) => setSelected(ids)} />\n\n+ <SvGrid\n+ {data} {columns} features={features}\n+ selectionMode=\"row\"\n+ onRowSelectionChange={(selection, rows) => setSelected(rows)} />\n```\n\n`selectionMode` values: `'row'` (the MUI X equivalent), `'cell'`\n(spreadsheet-style range), `'both'`, `'none'`.\n\n## Inline editing\n\nMUI X had `processRowUpdate(newRow, oldRow)` returning the new row.\nSv-grid's equivalent is `onCellValueChange`:\n\n```diff\n- <DataGrid processRowUpdate={async (newRow) => {\n- await api.savePatch(newRow)\n- return newRow\n- }} />\n\n+ <SvGrid\n+ onCellValueChange={async (e) => {\n+ await api.savePatch({ id: e.row.id, [e.columnId]: e.newValue })\n+ }}\n+ />\n```\n\nFor full-row editing (one Save button per row), see\n[Full-row editing](./editing/full-row.md).\n\n## Server-side data\n\nMUI X's `pagination + serverSideMode + filterMode='server'` maps to\nsv-grid's `externalSort + externalFilter`, or to `createServerDataSource` (free:\npaging or infinite scroll, sort, filter) when the server pages, and to the\nEnterprise `createServerRowModel` when it also groups; the demo below runs\nthe free controller against a fake API with sort, filter and infinite scroll:\n\n<div data-docs-demo=\"148-server-row-model\" data-height=\"520\"></div>\n\nMUI X's server-side row grouping and aggregation (Premium) map to the\nEnterprise row model, which loads each group level on expand and pages,\nedits and selects across rows the grid never fetched:\n\n<div data-docs-demo=\"467-server-row-model-1m\" data-height=\"600\"></div>\n\n```svelte\n<SvGrid\n data={rows} columns={columns} features={features}\n externalSort={true}\n externalFilter={true}\n onSortingChange={async (s) => { rows = await fetchPage({ sort: s }) }}\n onFiltersChange={async (f) => { rows = await fetchPage({ filters: f.columns }) }}\n/>\n```\n\n## Slots / customisation\n\nMUI X exposes a `slots` object. Sv-grid doesn't - instead, every\nvisual piece is a CSS custom property (`--sg-*`); see\n[design tokens](./tokens.md). Override at any DOM level:\n\n```css\n.grid-host { --sg-accent: #db2777; --sg-radius: 10px; }\n```\n\nRow height is the exception: it is the `rowHeight` prop rather than a\ntoken, because the virtualizer needs it as a number.\n\nFor full theme presets (Ant, MUI, Fluent, Base Web, shadcn) see\n[demo 74](https://svgrid.com/demos/74-theme-integrations/).\n\n## What you get for free vs MUI X\n\n- **No Emotion / no Material theme dependency.** The facts box at the\n top of the page has both packages measured the same way; the MUI X\n figure excludes React, Material UI, Emotion and the MUI system, which\n it needs and sv-grid does not.\n- **Pro and Premium features in the MIT core.** Column pinning, master/detail,\n tree data, row grouping, aggregation and cell selection carry Pro or\n Premium badges on mui.com; they are free here. Excel export and pivot\n are paid on both sides.\n- **In-grid AI helpers against your own model.** MUI X's AI assistant is a\n Premium feature that routes prompts through MUI's hosted service unless\n you build your own; sv-grid's natural-language filter, smart fill and\n summarise helpers are free and run against a provider you register.\n- **CSP-clean.** No `eval`.\n\n## What you give up\n\n- **The Material Design look out of the box.** Sv-grid ships\n unstyled-by-token; the [MUI preset in demo 74](https://svgrid.com/demos/74-theme-integrations/)\n is one drop-in.\n- **MUI form-field integration.** Bind directly to your own MUI\n inputs in custom cell components if you want them.\n- **The rest of MUI X in one design system.** Date pickers, charts and\n the tree view share MUI X's theme; sv-grid ships its own date and time\n editors and charts, themed through the same `--sg-*` tokens.\n- **A hosted AI assistant with starter credits.** Premium licence holders\n get MUI's service; sv-grid's helpers need a model provider you host or\n pay for yourself.\n\n## Frequently asked questions\n\n### How hard is it to migrate from MUI X DataGrid to SvGrid?\n\nMostly mechanical - typically 1-3 hours per grid, depending on how heavily you\nrelied on `apiRef.current.*`. The Community/Enterprise split mirrors MUI X's, so the\nlicensing mental model carries over directly.\n\n### Can SvGrid keep the Material Design look?\n\nYes. SvGrid ships unstyled-by-token and re-themes through `--sg-*` CSS\nvariables; the MUI preset in demo 74 is a drop-in starting point. You can also\nbind your own MUI inputs inside custom cell components.\n\n### Is SvGrid cheaper than MUI X Pro/Premium?\n\nSvGrid's Community tier is MIT and free for commercial use, and `@svgrid/enterprise`\nis priced per developer ($599 for the grid, $999 with the spreadsheet and\nStudio, both covering unlimited apps). MUI X Pro and\nPremium are also per developer per year; the prices mui.com listed on the\ndate we read it are in the facts box at the top of this page. The difference\nis what the free tier holds: most of what MUI X sells in Pro and Premium is\nin SvGrid's MIT core, so compare feature needs before prices on the\n[pricing page](https://svgrid.com/pricing/).\n\n## What you end up with\n\nSorting, filtering, pagination and inline editing - the MUI X DataGrid feature set.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable editable pageable pageSize={3} />\n```\n\n## See also\n\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/) - the side-by-side comparison, with a source and date for every claim\n- [Server-side data](./server-side-data.md) - the row model the demo above uses\n- [AI assistant](./ai.md) - the free in-grid helpers\n- [Migrating from AG Grid](./migrating-from-ag-grid.md)\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md)\n- [Design tokens](./tokens.md)\n"
6133
+ "markdown": "# Migrating from MUI X DataGrid\n\nMUI X DataGrid is the most common starting point for teams already\non Material UI. Its MIT Community edition plus paid Pro and Premium\ntiers is a split similar to sv-grid's, except that column pinning,\nmaster/detail, tree data, row grouping, aggregation and cell selection\nare paid in MUI X and free here. The port is mostly mechanical.\n\n> Estimated effort: **1-3 hours** per grid, depending on how heavily\n> you've leant on `apiRef.current.*` calls.\n\n<!-- facts:start mui-x-datagrid -->\n> **Facts, checked 9 Nov 2026.** `@mui/x-data-grid` 9.17.0, MIT, last published 30 Oct 2026, 13,100,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `@mui/x-data-grid` 9.17.0 158.4 KB JS, no separate stylesheet, @emotion/react + @emotion/styled + @mui/material + @mui/system + react + react-dom external (measured 9 Nov 2026). MUI X DataGrid pricing, as its site states it: MUI X Community is free under MIT. mui.com lists Pro at $299 per year per developer, Premium at $599 per year per developer and Enterprise at $1,399 per year per developer, with perpetual and annual licence models (https://mui.com/pricing/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/).\n<!-- facts:end -->\n\n## Package map\n\n| MUI X | sv-grid |\n| ------------------------------------ | ---------------------------------------- |\n| `@mui/x-data-grid` | `@svgrid/grid` |\n| `@mui/x-data-grid-pro` | `@svgrid/enterprise` (export, import, pivot, AI) |\n| `@mui/x-data-grid-premium` | All of `@svgrid/enterprise` ships in one tier |\n\n## Imports\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import { DataGrid, GridColDef, GridRowsProp } from '@mui/x-data-grid'\n\n+ import { SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef } from '@svgrid/grid'\n```\n\n## Column defs\n\n```diff\n- const columns: GridColDef[] = [\n- { field: 'id', headerName: 'ID', width: 90 },\n- { field: 'lastName', headerName: 'Last', width: 150, editable: true },\n- { field: 'age', headerName: 'Age', type: 'number', width: 110, editable: true },\n- ]\n\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'id', header: 'ID', width: 90 },\n+ { field: 'lastName', header: 'Last', width: 150 },\n+ { field: 'age', header: 'Age', width: 110, editorType: 'number' },\n+ ]\n```\n\n- `headerName` → `header`\n- `type: 'number'` → `editorType: 'number'`\n- `editable: true` → omit (every column is editable by default; set\n `editable: false` to OPT OUT)\n\n## Mounting\n\n```diff\n- <DataGrid rows={rows} columns={columns} pageSize={25} checkboxSelection />\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ showPagination={true} pageSize={25}\n+ selectionMode=\"row\" />\n```\n\n`tableFeatures({ rowSortingFeature, columnFilteringFeature,\nrowSelectionFeature })` registers what you'd implicitly get from\nMUI X.\n\n## `apiRef` translation\n\n| MUI X (`apiRef.current.*`) | sv-grid (`api.*`) |\n| -------------------------------------------- | ------------------------------------------ |\n| `setRows(rows)` | mutate the `$state` array you bound to `data` |\n| `updateRows([{id, ...patch}])` | `api.setCellValue(rowIndex, field, value)` |\n| `setSortModel([{field, sort}])` | `api.setSort(field, sort)` |\n| `setFilterModel({items: [...]})` | `api.setFilter(field, {operator, value})` |\n| `setPage(0)` / `setPageSize(50)` | Use the built-in pager; for headless control register `pageSize` prop |\n| `selectRow(id)` | Toggle the row checkbox via the wrapper's UI; programmatic select is `api.selectRows([id])` (`api.toggleRowSelected(id)` to flip one) |\n| `getSelectedRows()` | `api.getSelectedRows()` (ids via `api.getSelectedRowIds()`) |\n| `exportDataAsExcel()` | `api.exportData({ format: 'xlsx' })` (Enterprise) |\n| `setColumnVisibilityModel({field: false})` | `api.setColumnVisible('field', false)` |\n\n## Custom cells\n\n```diff\n- {\n- field: 'status',\n- renderCell: ({ row }) => <Chip label={row.status} color={row.status === 'active' ? 'success' : 'default'} />,\n- }\n\n+ {\n+ field: 'status',\n+ cell: (ctx) => renderSnippet(StatusChip, { status: ctx.row.original.status }),\n+ }\n```\n\n## Selection\n\n```diff\n- <DataGrid checkboxSelection\n- onRowSelectionModelChange={(ids) => setSelected(ids)} />\n\n+ <SvGrid\n+ {data} {columns} features={features}\n+ selectionMode=\"row\"\n+ onRowSelectionChange={(selection, rows) => setSelected(rows)} />\n```\n\n`selectionMode` values: `'row'` (the MUI X equivalent), `'cell'`\n(spreadsheet-style range), `'both'`, `'none'`.\n\n## Inline editing\n\nMUI X had `processRowUpdate(newRow, oldRow)` returning the new row.\nSv-grid's equivalent is `onCellValueChange`:\n\n```diff\n- <DataGrid processRowUpdate={async (newRow) => {\n- await api.savePatch(newRow)\n- return newRow\n- }} />\n\n+ <SvGrid\n+ onCellValueChange={async (e) => {\n+ await api.savePatch({ id: e.row.id, [e.columnId]: e.newValue })\n+ }}\n+ />\n```\n\nFor full-row editing (one Save button per row), see\n[Full-row editing](./editing/full-row.md).\n\n## Server-side data\n\nMUI X's `pagination + serverSideMode + filterMode='server'` maps to\nsv-grid's `externalSort + externalFilter`, or to `createServerDataSource` (free:\npaging or infinite scroll, sort, filter) when the server pages, and to the\nEnterprise `createServerRowModel` when it also groups; the demo below runs\nthe free controller against a fake API with sort, filter and infinite scroll:\n\n<div data-docs-demo=\"148-server-row-model\" data-height=\"520\"></div>\n\nMUI X's server-side row grouping and aggregation (Premium) map to the\nEnterprise row model, which loads each group level on expand and pages,\nedits and selects across rows the grid never fetched:\n\n<div data-docs-demo=\"467-server-row-model-1m\" data-height=\"600\"></div>\n\n```svelte\n<SvGrid\n data={rows} columns={columns} features={features}\n externalSort={true}\n externalFilter={true}\n onSortingChange={async (s) => { rows = await fetchPage({ sort: s }) }}\n onFiltersChange={async (f) => { rows = await fetchPage({ filters: f.columns }) }}\n/>\n```\n\n## Slots / customisation\n\nMUI X exposes a `slots` object. Sv-grid doesn't - instead, every\nvisual piece is a CSS custom property (`--sg-*`); see\n[design tokens](./tokens.md). Override at any DOM level:\n\n```css\n.grid-host { --sg-accent: #db2777; --sg-radius: 10px; }\n```\n\nRow height is the exception: it is the `rowHeight` prop rather than a\ntoken, because the virtualizer needs it as a number.\n\nFor full theme presets (Ant, MUI, Fluent, Base Web, shadcn) see\n[demo 74](https://svgrid.com/demos/74-theme-integrations/).\n\n## What you get for free vs MUI X\n\n- **No Emotion / no Material theme dependency.** The facts box at the\n top of the page has both packages measured the same way; the MUI X\n figure excludes React, Material UI, Emotion and the MUI system, which\n it needs and sv-grid does not.\n- **Pro and Premium features in the MIT core.** Column pinning, master/detail,\n tree data, row grouping, aggregation and cell selection carry Pro or\n Premium badges on mui.com; they are free here. Excel export and pivot\n are paid on both sides.\n- **In-grid AI helpers against your own model.** MUI X's AI assistant is a\n Premium feature that routes prompts through MUI's hosted service unless\n you build your own; sv-grid's natural-language filter, smart fill and\n summarise helpers are free and run against a provider you register.\n- **CSP-clean.** No `eval`.\n\n## What you give up\n\n- **The Material Design look out of the box.** Sv-grid ships\n unstyled-by-token; the [MUI preset in demo 74](https://svgrid.com/demos/74-theme-integrations/)\n is one drop-in.\n- **MUI form-field integration.** Bind directly to your own MUI\n inputs in custom cell components if you want them.\n- **The rest of MUI X in one design system.** Date pickers, charts and\n the tree view share MUI X's theme; sv-grid ships its own date and time\n editors and charts, themed through the same `--sg-*` tokens.\n- **A hosted AI assistant with starter credits.** Premium licence holders\n get MUI's service; sv-grid's helpers need a model provider you host or\n pay for yourself.\n\n## Frequently asked questions\n\n### How hard is it to migrate from MUI X DataGrid to SvGrid?\n\nMostly mechanical - typically 1-3 hours per grid, depending on how heavily you\nrelied on `apiRef.current.*`. The Community/Enterprise split mirrors MUI X's, so the\nlicensing mental model carries over directly.\n\n### Can SvGrid keep the Material Design look?\n\nYes. SvGrid ships unstyled-by-token and re-themes through `--sg-*` CSS\nvariables; the MUI preset in demo 74 is a drop-in starting point. You can also\nbind your own MUI inputs inside custom cell components.\n\n### Is SvGrid cheaper than MUI X Pro/Premium?\n\nSvGrid's Community tier is MIT and free for commercial use, and `@svgrid/enterprise`\nis priced per developer ($599 for the grid, $999 with the spreadsheet and\nStudio, both covering unlimited apps). MUI X Pro and\nPremium are also per developer per year; the prices mui.com listed on the\ndate we read it are in the facts box at the top of this page. The difference\nis what the free tier holds: most of what MUI X sells in Pro and Premium is\nin SvGrid's MIT core, so compare feature needs before prices on the\n[pricing page](https://svgrid.com/pricing/).\n\n## What you end up with\n\nSorting, filtering, pagination and inline editing - the MUI X DataGrid feature set.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable editable pageable pageSize={3} />\n```\n\n## See also\n\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/) - the side-by-side comparison, with a source and date for every claim\n- [Server-side data](./server-side-data.md) - the row model the demo above uses\n- [AI assistant](./ai.md) - the free in-grid helpers\n- [Migrating from AG Grid](./migrating-from-ag-grid.md)\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md)\n- [Design tokens](./tokens.md)\n"
6134
6134
  },
6135
6135
  {
6136
6136
  "slug": "help/migrating-from-nocodb",
@@ -6144,77 +6144,77 @@ export const docs = [
6144
6144
  "path": "docs/help/migrating-from-primevue-datatable.md",
6145
6145
  "title": "Migrating from PrimeVue / PrimeNG / PrimeReact DataTable",
6146
6146
  "section": "Help",
6147
- "markdown": "\n# Migrating from PrimeVue / PrimeNG / PrimeReact DataTable\n\nThe Prime DataTable is the grid in the Prime UI suites for Vue, Angular,\nand React. Its columns are declared as `<Column>` children with `field`\n/ `header` props - a shape that maps cleanly onto SvGrid's `ColumnDef`\narray. The port is mostly translating columns and events into a Svelte 5\ncomponent. Since PrimeVue 5 the package ships under the PrimeUI licence\nrather than MIT: free for individuals and small organisations under the\nlimits in its licence file, per developer with a licence key otherwise.\nThe facts box below quotes the terms as read on the date shown.\n\n> Estimated effort: **2-4 hours** per grid.\n\n<!-- facts:start primevue-datatable -->\n> **Facts, checked 9 Nov 2026.** `primevue` 5.0.4, Commercial (see licence file), last published 22 Oct 2026, 3,290,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). PrimeVue / PrimeNG / PrimeReact DataTable pricing, as its site states it: primeui.dev lists the Commercial Suite at a launch price of $599 per developer (standard $799) as a perpetual licence with one year of updates, renewable at $399 per developer per year; the Community License is free for individuals, non-profits and organisations under $1M revenue, fewer than five developers, fewer than ten employees and under $3M in venture funding (https://primeui.dev/pricing, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs PrimeVue / PrimeNG / PrimeReact DataTable](https://svgrid.com/compare/primevue-datatable/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Prime DataTable | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `<DataTable :value=\"rows\">` | `<SvGrid data={rows}>` |\n| `<Column field=\"name\" header=\"Name\" />` | `{ field: 'name', header: 'Name' }` |\n| `sortable` | `rowSortingFeature` |\n| `:filters` / filter display | `columnFilteringFeature` |\n| `<template #body>` cell slot | `cell: (c) => renderSnippet(...)` |\n| `editMode` + cell editor template | `editorType` (+ custom `cellEditor`) |\n| `:expandedRows` / row expansion | `rowExpandingFeature` |\n| `lazy` + `@page` / `@sort` / `@filter` | `externalSort` / `externalFilter` + events |\n| `paginator` + `:rows` | `showPagination` (+ page size) |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <DataTable :value=\"rows\" paginator :rows=\"25\" sortMode=\"multiple\">\n- <Column field=\"name\" header=\"Name\" sortable />\n- <Column field=\"amount\" header=\"Amount\" sortable>\n- <template #body=\"{ data }\">{{ fmt(data.amount) }}</template>\n- </Column>\n- </DataTable>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## What changes\n\n- **`<Column>` children → a `ColumnDef[]` array.**\n- **`#body` slots → `cell` snippets** via `renderSnippet`.\n- **Prime theming → `--sg-*` tokens / Tailwind** - no Prime theme\n dependency.\n- **`lazy` → `externalSort` / `externalFilter`** with refetch on events.\n\n## Frequently asked questions\n\n### How do Prime `<Column>` definitions map to SvGrid?\n\nEach `<Column field header />` becomes a `{ field, header }` object in SvGrid's\n`columns` array, and `#body` templates become `renderSnippet` cells. Sorting and\nfiltering move from per-column props to the `rowSortingFeature` /\n`columnFilteringFeature` registration.\n\n### Is SvGrid open-source like the Prime grid core?\n\n`@svgrid/grid` is MIT with no licence key or eligibility conditions. The\nPrime cores were MIT through PrimeVue 4; from PrimeVue 5 the npm package\ncarries the PrimeUI licence, whose free Community License depends on your\nrevenue, developer count, headcount and funding, and whose Commercial\nLicense is per developer with a key. The other difference is framework:\nSvGrid is Svelte 5, not Vue / Angular / React.\n\n### Do I lose Prime theming?\n\nYou re-create the look with `--sg-*` tokens (optionally driven by your Tailwind\ntheme), so the grid matches your design system without a Prime theme dependency.\n\n## What you end up with\n\nSorting, filtering, pagination and row selection in one pass.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} selectable />\n```\n\n## See also\n\n- [SvGrid vs Prime DataTable](https://svgrid.com/compare/primevue-datatable/) - the side-by-side comparison\n- [Tailwind integration](./tailwind.md) - match your design system\n- [Cell components](./cells/cell-components.md) - snippet cells + editors\n"
6147
+ "markdown": "\n# Migrating from PrimeVue / PrimeNG / PrimeReact DataTable\n\nThe Prime DataTable is the grid in the Prime UI suites for Vue, Angular,\nand React. Its columns are declared as `<Column>` children with `field`\n/ `header` props - a shape that maps cleanly onto SvGrid's `ColumnDef`\narray. The port is mostly translating columns and events into a Svelte 5\ncomponent. Since PrimeVue 5 the package ships under the PrimeUI licence\nrather than MIT: free for individuals and small organisations under the\nlimits in its licence file, per developer with a licence key otherwise.\nThe facts box below quotes the terms as read on the date shown.\n\n> Estimated effort: **2-4 hours** per grid.\n\n<!-- facts:start primevue-datatable -->\n> **Facts, checked 9 Nov 2026.** `primevue` 5.0.4, Commercial (see licence file), last published 22 Oct 2026, 3,290,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). PrimeVue / PrimeNG / PrimeReact DataTable pricing, as its site states it: primeui.dev lists the Commercial Suite at a launch price of $599 per developer (standard $799) as a perpetual licence with one year of updates, renewable at $399 per developer per year; the Community License is free for individuals, non-profits and organisations under $1M revenue, fewer than five developers, fewer than ten employees and under $3M in venture funding (https://primeui.dev/pricing, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs PrimeVue / PrimeNG / PrimeReact DataTable](https://svgrid.com/compare/primevue-datatable/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Prime DataTable | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `<DataTable :value=\"rows\">` | `<SvGrid data={rows}>` |\n| `<Column field=\"name\" header=\"Name\" />` | `{ field: 'name', header: 'Name' }` |\n| `sortable` | `rowSortingFeature` |\n| `:filters` / filter display | `columnFilteringFeature` |\n| `<template #body>` cell slot | `cell: (c) => renderSnippet(...)` |\n| `editMode` + cell editor template | `editorType` (+ custom `cellEditor`) |\n| `:expandedRows` / row expansion | `rowExpandingFeature` |\n| `lazy` + `@page` / `@sort` / `@filter` | `externalSort` / `externalFilter` + events |\n| `paginator` + `:rows` | `showPagination` (+ page size) |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <DataTable :value=\"rows\" paginator :rows=\"25\" sortMode=\"multiple\">\n- <Column field=\"name\" header=\"Name\" sortable />\n- <Column field=\"amount\" header=\"Amount\" sortable>\n- <template #body=\"{ data }\">{{ fmt(data.amount) }}</template>\n- </Column>\n- </DataTable>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## What changes\n\n- **`<Column>` children → a `ColumnDef[]` array.**\n- **`#body` slots → `cell` snippets** via `renderSnippet`.\n- **Prime theming → `--sg-*` tokens / Tailwind** - no Prime theme\n dependency.\n- **`lazy` → `externalSort` / `externalFilter`** with refetch on events.\n\n## Frequently asked questions\n\n### How do Prime `<Column>` definitions map to SvGrid?\n\nEach `<Column field header />` becomes a `{ field, header }` object in SvGrid's\n`columns` array, and `#body` templates become `renderSnippet` cells. Sorting and\nfiltering move from per-column props to the `rowSortingFeature` /\n`columnFilteringFeature` registration.\n\n### Is SvGrid open-source like the Prime grid core?\n\n`@svgrid/grid` is MIT with no licence key or eligibility conditions. The\nPrime cores were MIT through PrimeVue 4; from PrimeVue 5 the npm package\ncarries the PrimeUI licence, whose free Community License depends on your\nrevenue, developer count, headcount and funding, and whose Commercial\nLicense is per developer with a key. The other difference is framework:\nSvGrid is Svelte 5, not Vue / Angular / React.\n\n### Do I lose Prime theming?\n\nYou re-create the look with `--sg-*` tokens (optionally driven by your Tailwind\ntheme), so the grid matches your design system without a Prime theme dependency.\n\n## What you end up with\n\nSorting, filtering, pagination and row selection in one pass.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} selectable />\n```\n\n## See also\n\n- [SvGrid vs Prime DataTable](https://svgrid.com/compare/primevue-datatable/) - the side-by-side comparison\n- [Tailwind integration](./tailwind.md) - match your design system\n- [Cell components](./cells/cell-components.md) - snippet cells + editors\n"
6148
6148
  },
6149
6149
  {
6150
6150
  "slug": "help/migrating-from-react-data-grid",
6151
6151
  "path": "docs/help/migrating-from-react-data-grid.md",
6152
6152
  "title": "Migrating from React Data Grid (adazzle)",
6153
6153
  "section": "Help",
6154
- "markdown": "# Migrating from React Data Grid (adazzle)\n\n`react-data-grid` is a fast, Excel-like React grid with a clean column\nAPI. SvGrid mirrors its shape - virtualized DOM, typed columns, inline\nediting, custom cells - on Svelte 5 runes instead of React, so the port\nis mostly a column rename pass.\n\n> Estimated effort: **2-4 hours** per grid, more if you lean on custom\n> editors.\n\n<!-- facts:start react-data-grid -->\n> **Facts, checked 9 Nov 2026.** `react-data-grid` 7.0.0-beta.64, MIT, last published 5 Nov 2026, 2,650,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `react-data-grid` 7.0.0-beta.64 25.6 KB JS + 2.3 KB CSS, react + react-dom external (measured 9 Nov 2026). React Data Grid pricing, as its site states it: react-data-grid is MIT on npm and free; no licence is sold (https://github.com/adazzle/react-data-grid, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs React Data Grid (adazzle)](https://svgrid.com/compare/react-data-grid/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| react-data-grid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `<DataGrid columns rows />` | `<SvGrid columns data />` |\n| `{ key: 'name', name: 'Name' }` | `{ field: 'name', header: 'Name' }` |\n| `renderCell: (p) => <Badge .../>` | `cell: (c) => renderSnippet(Badge, {...})` |\n| `renderEditCell` / `editor` | `editorType` (+ custom `cellEditor`) |\n| `rowKeyGetter={(r) => r.id}` | `getRowId={(r) => r.id}` |\n| `onRowsChange={setRows}` | `onCellValueChange` (+ your state) |\n| `sortColumns` / `onSortColumnsChange` | `rowSortingFeature` + `onSortingChange` |\n| `<DataGrid className=...>` theming | `--sg-*` tokens / Tailwind |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import DataGrid, { textEditor } from 'react-data-grid'\n- const columns = [\n- { key: 'name', name: 'Name' },\n- { key: 'amount', name: 'Amount', renderEditCell: textEditor },\n- ]\n- <DataGrid columns={columns} rows={rows} rowKeyGetter={(r) => r.id} onRowsChange={setRows} />\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', editorType: 'text' },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features}\n+ getRowId={(r) => r.id} enableInlineEditing\n+ onCellValueChange={(e) => save(e)} />\n```\n\n## What changes\n\n- **React → Svelte 5.** `renderCell` / `renderEditCell` become\n `renderSnippet` snippets and `editorType`.\n- **`onRowsChange` → `onCellValueChange`.** SvGrid writes to its own\n working copy; subscribe to commits and persist where you like.\n- **Filtering UI is built in.** react-data-grid leaves the filter row to\n you; SvGrid ships the Excel-style menu.\n\n## Frequently asked questions\n\n### How do react-data-grid columns map to SvGrid?\n\n`key` becomes `field`, `name` becomes `header`, `renderCell` becomes a\n`renderSnippet` cell, and editors map to `editorType` (or a custom `cellEditor`\nsnippet). `rowKeyGetter` becomes `getRowId`.\n\n### Is SvGrid free like react-data-grid?\n\nYes - `@svgrid/grid` is MIT-licensed, like react-data-grid. SvGrid adds an\noptional paid Enterprise pack for export, import, pivot, and AI.\n\n### Does SvGrid have built-in filtering?\n\nYes. react-data-grid asks you to build the filter UI; SvGrid ships an\nExcel-style column filter menu and a global filter out of the box.\n\n## What you end up with\n\nInline editing with cell-range selection, which is the reason most people picked it.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection sortable />\n```\n\n## See also\n\n- [SvGrid vs React Data Grid](https://svgrid.com/compare/react-data-grid/) - the side-by-side comparison\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - sibling React-to-Svelte guide\n- [Cell components](./cells/cell-components.md) - custom cells + editors\n"
6154
+ "markdown": "# Migrating from React Data Grid (adazzle)\n\n`react-data-grid` is a fast, Excel-like React grid with a clean column\nAPI. SvGrid mirrors its shape - virtualized DOM, typed columns, inline\nediting, custom cells - on Svelte 5 runes instead of React, so the port\nis mostly a column rename pass.\n\n> Estimated effort: **2-4 hours** per grid, more if you lean on custom\n> editors.\n\n<!-- facts:start react-data-grid -->\n> **Facts, checked 9 Nov 2026.** `react-data-grid` 7.0.0-beta.64, MIT, last published 5 Nov 2026, 2,650,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `react-data-grid` 7.0.0-beta.64 25.6 KB JS + 2.3 KB CSS, react + react-dom external (measured 9 Nov 2026). React Data Grid pricing, as its site states it: react-data-grid is MIT on npm and free; no licence is sold (https://github.com/adazzle/react-data-grid, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs React Data Grid (adazzle)](https://svgrid.com/compare/react-data-grid/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| react-data-grid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `<DataGrid columns rows />` | `<SvGrid columns data />` |\n| `{ key: 'name', name: 'Name' }` | `{ field: 'name', header: 'Name' }` |\n| `renderCell: (p) => <Badge .../>` | `cell: (c) => renderSnippet(Badge, {...})` |\n| `renderEditCell` / `editor` | `editorType` (+ custom `cellEditor`) |\n| `rowKeyGetter={(r) => r.id}` | `getRowId={(r) => r.id}` |\n| `onRowsChange={setRows}` | `onCellValueChange` (+ your state) |\n| `sortColumns` / `onSortColumnsChange` | `rowSortingFeature` + `onSortingChange` |\n| `<DataGrid className=...>` theming | `--sg-*` tokens / Tailwind |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import DataGrid, { textEditor } from 'react-data-grid'\n- const columns = [\n- { key: 'name', name: 'Name' },\n- { key: 'amount', name: 'Amount', renderEditCell: textEditor },\n- ]\n- <DataGrid columns={columns} rows={rows} rowKeyGetter={(r) => r.id} onRowsChange={setRows} />\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', editorType: 'text' },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features}\n+ getRowId={(r) => r.id} enableInlineEditing\n+ onCellValueChange={(e) => save(e)} />\n```\n\n## What changes\n\n- **React → Svelte 5.** `renderCell` / `renderEditCell` become\n `renderSnippet` snippets and `editorType`.\n- **`onRowsChange` → `onCellValueChange`.** SvGrid writes to its own\n working copy; subscribe to commits and persist where you like.\n- **Filtering UI is built in.** react-data-grid leaves the filter row to\n you; SvGrid ships the Excel-style menu.\n\n## Frequently asked questions\n\n### How do react-data-grid columns map to SvGrid?\n\n`key` becomes `field`, `name` becomes `header`, `renderCell` becomes a\n`renderSnippet` cell, and editors map to `editorType` (or a custom `cellEditor`\nsnippet). `rowKeyGetter` becomes `getRowId`.\n\n### Is SvGrid free like react-data-grid?\n\nYes - `@svgrid/grid` is MIT-licensed, like react-data-grid. SvGrid adds an\noptional paid Enterprise pack for export, import, pivot, and AI.\n\n### Does SvGrid have built-in filtering?\n\nYes. react-data-grid asks you to build the filter UI; SvGrid ships an\nExcel-style column filter menu and a global filter out of the box.\n\n## What you end up with\n\nInline editing with cell-range selection, which is the reason most people picked it.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} editable enableCellSelection sortable />\n```\n\n## See also\n\n- [SvGrid vs React Data Grid](https://svgrid.com/compare/react-data-grid/) - the side-by-side comparison\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - sibling React-to-Svelte guide\n- [Cell components](./cells/cell-components.md) - custom cells + editors\n"
6155
6155
  },
6156
6156
  {
6157
6157
  "slug": "help/migrating-from-shadcn-data-table",
6158
6158
  "path": "docs/help/migrating-from-shadcn-data-table.md",
6159
6159
  "title": "Migrating from the shadcn-svelte data table",
6160
6160
  "section": "Help",
6161
- "markdown": "# Migrating from the shadcn-svelte data table\n\nThe shadcn-svelte **Data Table** is not the same thing as its `<Table>`\ncomponent. `<Table>` is presentation only. The data table is a recipe\nbuilt on TanStack Table v9 that you copy into your own repo: it already\ndoes sorting, filtering, pagination, row selection, and column\nvisibility. If that is what you have, this is your page. If you have the\nplain styled `<Table>` and hand-wrote the sorting yourself, read\n[Migrating from a UI-kit table](./migrating-from-ui-kit-tables.md)\ninstead.\n\n> Estimated effort: **20-40 min** per table. Your column definitions and\n> your feature registration carry over almost unchanged. Most of the work\n> is deleting files.\n\n<!-- facts:start tanstack-table,svelte-ui-kit-tables -->\n> **Facts, checked 9 Nov 2026.** `@tanstack/svelte-table` 9.3.1, MIT, last published 12 Oct 2026, 539,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `@tanstack/svelte-table` 9.3.1 36.5 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). TanStack Table pricing, as its site states it: TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table (https://tanstack.com/table/latest, read 12 Sep 2026). Flowbite / Skeleton / shadcn-svelte tables pricing, as its site states it: Flowbite Svelte, Skeleton and shadcn-svelte are MIT component libraries; their table components are free and no grid licence is sold (https://github.com/themesberg/flowbite-svelte, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/), [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/).\n<!-- facts:end -->\n\n## Why this port is small\n\nSvGrid speaks the same vocabulary TanStack Table v9 introduced. These are\nreal exports from `@svgrid/grid`, not lookalikes:\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```ts\nimport {\n tableFeatures,\n rowSortingFeature,\n columnFilteringFeature,\n rowPaginationFeature,\n rowSelectionFeature,\n rowExpandingFeature,\n columnGroupingFeature,\n} from '@svgrid/grid'\n```\n\nSo `data-table-features.ts` moves over as-is, and `columns.ts` keeps its\nshape. What does not move is the rendering scaffolding, because SvGrid\nships a renderer and there is nothing to translate it into.\n\n## What happens to each file\n\nThe shadcn recipe generates seven files. Five of them stop existing.\n\n| Their file | Becomes |\n| --- | --- |\n| `data-table-features.ts` | Kept. Same `tableFeatures({...})` call. |\n| `columns.ts` | Kept, minus the `header` render helpers. `{ accessorKey: 'email' }` becomes `{ field: 'email' }`. |\n| `data-table.svelte` | Deleted. It is `<table>` markup over `getHeaderGroups()` / `getRowModel()`, which `<SvGrid>` does. |\n| `data-table-checkbox.svelte` | Deleted. `showRowSelection` renders the column and the select-all header. |\n| `data-table-email-button.svelte` | Deleted. `sortable` makes every header a sort toggle. |\n| `data-table-actions.svelte` | Kept if you want it, as a cell snippet. |\n| `+page.svelte` | Kept. It now renders one component instead of wiring state. |\n\n## Let the codemod do the mechanical part\n\n```bash\nnpx @svgrid/migrate src/routes/payments # preview\nnpx @svgrid/migrate src/routes/payments --write # apply\n```\n\nIt previews by default and only writes with `--write`. Point it at the whole\nroute directory rather than a single file: the shadcn layout keeps `features`\nin its own module, and the codemod follows that import to work out which\nprops the component needs.\n\nIt renames the column keys, re-points the imports, strips the features SvGrid\ndoes not have, and deletes the markup along with the `$state` / `onXChange`\npairs. Custom `cell` and `header` render functions are kept and reported, never\nguessed at. Run your type-checker afterwards - the markup is regenerated rather\nthan edited.\n\n## Before and after\n\n```diff\n- <script lang=\"ts\">\n- import { createSvelteTable, FlexRender } from '$lib/components/ui/data-table'\n- import { columns } from './columns'\n- import { features } from './data-table-features'\n-\n- let sorting = $state([])\n- let filters = $state([])\n- let selection = $state({})\n- let pagination = $state({ pageIndex: 0, pageSize: 10 })\n-\n- const table = createSvelteTable({\n- get data() { return data },\n- columns, features,\n- state: { get sorting() { return sorting }, /* ...and three more... */ },\n- onSortingChange: (u) => (sorting = typeof u === 'function' ? u(sorting) : u),\n- // ...one of these per piece of state...\n- })\n- </script>\n-\n- <Table.Root>\n- <Table.Header>\n- {#each table.getHeaderGroups() as headerGroup}\n- <!-- ...twenty more lines of markup... -->\n- {/each}\n- </Table.Header>\n- <Table.Body>\n- {#each table.getRowModel().rows as row}\n- <!-- ...and twenty more... -->\n- {/each}\n- </Table.Body>\n- </Table.Root>\n\n+ <script lang=\"ts\">\n+ import { SvGrid } from '@svgrid/grid'\n+ import '@svgrid/grid/themes/shadcn.css'\n+ import { columns } from './columns'\n+ import { features } from './data-table-features'\n+ </script>\n+\n+ <SvGrid\n+ {data} {columns} {features}\n+ sortable filterable showGlobalFilter showRowSelection\n+ pageable showPagination pageSize={10}\n+ fitColumns\n+ />\n```\n\nThe four `$state` declarations and their four `onXChange` updaters go\naway with the markup. Sorting, filtering, paging, and selection state\nlive in the grid.\n\nColumn visibility moves too: mark a column `visible: false` to start it\nhidden, and users toggle it from the header column menu. You no longer\nmaintain a dropdown for it.\n\n## One command\n\nUse the CLI you already have:\n\n```sh\nnpx shadcn-svelte@latest add https://svgrid.com/r/data-table.json\n```\n\nThat writes a ready-to-edit `data-table.svelte` into\n`$lib/components/ui/data-table/`, beside your other components, and\ninstalls `@svgrid/grid`. It is a normal registry item, so `components.json`\naliases and your existing setup are respected.\n\nIf you are not in a shadcn project, the same component ships through\nSvGrid's own scaffolder:\n\n```sh\nnpx @svgrid/ui add data-table\n```\n\nBoth routes write the identical file. It is yours to edit, exactly like a\nshadcn recipe. The difference is where the engine lives.\n\n## The upgrade path is the real difference\n\nBoth approaches put a file you own in your repo. Only one puts the\n*engine* in a package.\n\nWhen TanStack Table went from v8 to v9, `useReactTable` became `useTable`,\nthe `get*RowModel()` options became a features object, and `flexRender`\nbecame a component. Every copied data-table had to be hand-fixed, because\nthere is no `npm update` for code you pasted. With SvGrid the file you own\nstays a thin call site, and the behaviour behind it updates with a version\nbump.\n\nThe cost scales with how much you pasted. A seven-file starter is an\nafternoon. The richer community grids paste a whole subsystem - row\nrendering, virtualization, context menus, paste dialogs, hooks, config -\nand every one of those files is now yours to carry through the next\nupstream change.\n\n## Keep your design system\n\nYou do not have to leave shadcn's look, and you do not have to leave its\nmarkup either.\n\n- **Tokens.** `import '@svgrid/grid/themes/shadcn.css'` is a one-line\n match for the default palette in light and dark. To follow a customized\n theme, map your `--background` / `--border` / `--primary` onto the\n `--sg-*` properties - see the [shadcn integration\n guide](./shadcn.md).\n- **Your own markup.** If you want to keep shadcn's `Table.Root` /\n `Table.Row` elements exactly as they are, use the headless core from\n `@svgrid/grid/core` and swap only the engine. See\n [Why headless?](../why-headless.md).\n\n## What you gain\n\nThe official recipe stops where its docs stop: sorting, filtering,\npagination, row selection, column visibility. Community registry items go\nfurther, at the cost of pasting a whole multi-file subsystem into your\nrepo. In SvGrid each of these is a prop:\n\n- **Virtualization** (`virtualization`) - the official recipe has none, so\n a large table is a rewrite rather than a prop.\n- **Inline editing** (`editable`) and cell-range selection\n (`enableCellSelection`).\n- **Grouping with aggregation** (`groupable`) and tree data (`treeData`).\n- **Export** to CSV, and to Excel / PDF on the paid tier.\n- **WAI-ARIA grid semantics and keyboard navigation**, instead of a plain\n `<table>` you keep accessible yourself.\n\n## What you give up\n\n- **A file you can read top to bottom.** The recipe's behaviour is\n visible in your repo; SvGrid's is behind a prop. That is the trade the\n whole page is about.\n- **Pixel-level control of the markup**, unless you go headless.\n- **Weight.** For a short static list that never sorts, the copied\n recipe is smaller.\n\n## Frequently asked questions\n\n### Does SvGrid replace shadcn-svelte?\n\nNo. It replaces one recipe in it. Your buttons, dialogs, and inputs stay\nexactly as they are, and the grid reads the same theme variables they do.\n\n### Can I keep my existing columns.ts?\n\nMostly. Rename `accessorKey` to `field`, drop the `header` render\nhelpers that only existed to draw a sort button, and the rest carries\nover. Custom cells become snippets via `renderSnippet`.\n\n### Do I have to give up owning the code?\n\nYou still own the call site - `npx @svgrid/ui add data-table` writes a\nfile into your project that you edit freely. What you stop owning is the\nrow-model plumbing, which is the part that broke on the v8 to v9 upgrade.\n\n### Is this worth it for a 50-row table?\n\nProbably not, unless you want editing, grouping, or export later. The\ncase gets strong when the table grows past what fits on a page, or when\nthe next feature request is one you would otherwise hand-build.\n\n## What you end up with\n\nSorting, filtering, pagination and selection - the shadcn data-table demo, as props.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} selectable />\n```\n\n## See also\n\n- [shadcn-svelte integration](./shadcn.md) - matching the theme\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - the\n underlying API map\n- [Migrating from a UI-kit table](./migrating-from-ui-kit-tables.md) - for\n the plain styled `<Table>`\n- [Why headless?](../why-headless.md) - keeping your own markup\n"
6161
+ "markdown": "# Migrating from the shadcn-svelte data table\n\nThe shadcn-svelte **Data Table** is not the same thing as its `<Table>`\ncomponent. `<Table>` is presentation only. The data table is a recipe\nbuilt on TanStack Table v9 that you copy into your own repo: it already\ndoes sorting, filtering, pagination, row selection, and column\nvisibility. If that is what you have, this is your page. If you have the\nplain styled `<Table>` and hand-wrote the sorting yourself, read\n[Migrating from a UI-kit table](./migrating-from-ui-kit-tables.md)\ninstead.\n\n> Estimated effort: **20-40 min** per table. Your column definitions and\n> your feature registration carry over almost unchanged. Most of the work\n> is deleting files.\n\n<!-- facts:start tanstack-table,svelte-ui-kit-tables -->\n> **Facts, checked 9 Nov 2026.** `@tanstack/svelte-table` 9.3.1, MIT, last published 12 Oct 2026, 539,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `@tanstack/svelte-table` 9.3.1 36.5 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). TanStack Table pricing, as its site states it: TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table (https://tanstack.com/table/latest, read 12 Sep 2026). Flowbite / Skeleton / shadcn-svelte tables pricing, as its site states it: Flowbite Svelte, Skeleton and shadcn-svelte are MIT component libraries; their table components are free and no grid licence is sold (https://github.com/themesberg/flowbite-svelte, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/), [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/).\n<!-- facts:end -->\n\n## Why this port is small\n\nSvGrid speaks the same vocabulary TanStack Table v9 introduced. These are\nreal exports from `@svgrid/grid`, not lookalikes:\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```ts\nimport {\n tableFeatures,\n rowSortingFeature,\n columnFilteringFeature,\n rowPaginationFeature,\n rowSelectionFeature,\n rowExpandingFeature,\n columnGroupingFeature,\n} from '@svgrid/grid'\n```\n\nSo `data-table-features.ts` moves over as-is, and `columns.ts` keeps its\nshape. What does not move is the rendering scaffolding, because SvGrid\nships a renderer and there is nothing to translate it into.\n\n## What happens to each file\n\nThe shadcn recipe generates seven files. Five of them stop existing.\n\n| Their file | Becomes |\n| --- | --- |\n| `data-table-features.ts` | Kept. Same `tableFeatures({...})` call. |\n| `columns.ts` | Kept, minus the `header` render helpers. `{ accessorKey: 'email' }` becomes `{ field: 'email' }`. |\n| `data-table.svelte` | Deleted. It is `<table>` markup over `getHeaderGroups()` / `getRowModel()`, which `<SvGrid>` does. |\n| `data-table-checkbox.svelte` | Deleted. `showRowSelection` renders the column and the select-all header. |\n| `data-table-email-button.svelte` | Deleted. `sortable` makes every header a sort toggle. |\n| `data-table-actions.svelte` | Kept if you want it, as a cell snippet. |\n| `+page.svelte` | Kept. It now renders one component instead of wiring state. |\n\n## Let the codemod do the mechanical part\n\n```bash\nnpx @svgrid/migrate src/routes/payments # preview\nnpx @svgrid/migrate src/routes/payments --write # apply\n```\n\nIt previews by default and only writes with `--write`. Point it at the whole\nroute directory rather than a single file: the shadcn layout keeps `features`\nin its own module, and the codemod follows that import to work out which\nprops the component needs.\n\nIt renames the column keys, re-points the imports, strips the features SvGrid\ndoes not have, and deletes the markup along with the `$state` / `onXChange`\npairs. Custom `cell` and `header` render functions are kept and reported, never\nguessed at. Run your type-checker afterwards - the markup is regenerated rather\nthan edited.\n\n## Before and after\n\n```diff\n- <script lang=\"ts\">\n- import { createSvelteTable, FlexRender } from '$lib/components/ui/data-table'\n- import { columns } from './columns'\n- import { features } from './data-table-features'\n-\n- let sorting = $state([])\n- let filters = $state([])\n- let selection = $state({})\n- let pagination = $state({ pageIndex: 0, pageSize: 10 })\n-\n- const table = createSvelteTable({\n- get data() { return data },\n- columns, features,\n- state: { get sorting() { return sorting }, /* ...and three more... */ },\n- onSortingChange: (u) => (sorting = typeof u === 'function' ? u(sorting) : u),\n- // ...one of these per piece of state...\n- })\n- </script>\n-\n- <Table.Root>\n- <Table.Header>\n- {#each table.getHeaderGroups() as headerGroup}\n- <!-- ...twenty more lines of markup... -->\n- {/each}\n- </Table.Header>\n- <Table.Body>\n- {#each table.getRowModel().rows as row}\n- <!-- ...and twenty more... -->\n- {/each}\n- </Table.Body>\n- </Table.Root>\n\n+ <script lang=\"ts\">\n+ import { SvGrid } from '@svgrid/grid'\n+ import '@svgrid/grid/themes/shadcn.css'\n+ import { columns } from './columns'\n+ import { features } from './data-table-features'\n+ </script>\n+\n+ <SvGrid\n+ {data} {columns} {features}\n+ sortable filterable showGlobalFilter showRowSelection\n+ pageable showPagination pageSize={10}\n+ fitColumns\n+ />\n```\n\nThe four `$state` declarations and their four `onXChange` updaters go\naway with the markup. Sorting, filtering, paging, and selection state\nlive in the grid.\n\nColumn visibility moves too: mark a column `visible: false` to start it\nhidden, and users toggle it from the header column menu. You no longer\nmaintain a dropdown for it.\n\n## One command\n\nUse the CLI you already have:\n\n```sh\nnpx shadcn-svelte@latest add https://svgrid.com/r/data-table.json\n```\n\nThat writes a ready-to-edit `data-table.svelte` into\n`$lib/components/ui/data-table/`, beside your other components, and\ninstalls `@svgrid/grid`. It is a normal registry item, so `components.json`\naliases and your existing setup are respected.\n\nIf you are not in a shadcn project, the same component ships through\nSvGrid's own scaffolder:\n\n```sh\nnpx @svgrid/ui add data-table\n```\n\nBoth routes write the identical file. It is yours to edit, exactly like a\nshadcn recipe. The difference is where the engine lives.\n\n## The upgrade path is the real difference\n\nBoth approaches put a file you own in your repo. Only one puts the\n*engine* in a package.\n\nWhen TanStack Table went from v8 to v9, `useReactTable` became `useTable`,\nthe `get*RowModel()` options became a features object, and `flexRender`\nbecame a component. Every copied data-table had to be hand-fixed, because\nthere is no `npm update` for code you pasted. With SvGrid the file you own\nstays a thin call site, and the behaviour behind it updates with a version\nbump.\n\nThe cost scales with how much you pasted. A seven-file starter is an\nafternoon. The richer community grids paste a whole subsystem - row\nrendering, virtualization, context menus, paste dialogs, hooks, config -\nand every one of those files is now yours to carry through the next\nupstream change.\n\n## Keep your design system\n\nYou do not have to leave shadcn's look, and you do not have to leave its\nmarkup either.\n\n- **Tokens.** `import '@svgrid/grid/themes/shadcn.css'` is a one-line\n match for the default palette in light and dark. To follow a customized\n theme, map your `--background` / `--border` / `--primary` onto the\n `--sg-*` properties - see the [shadcn integration\n guide](./shadcn.md).\n- **Your own markup.** If you want to keep shadcn's `Table.Root` /\n `Table.Row` elements exactly as they are, use the headless core from\n `@svgrid/grid/core` and swap only the engine. See\n [Why headless?](../why-headless.md).\n\n## What you gain\n\nThe official recipe stops where its docs stop: sorting, filtering,\npagination, row selection, column visibility. Community registry items go\nfurther, at the cost of pasting a whole multi-file subsystem into your\nrepo. In SvGrid each of these is a prop:\n\n- **Virtualization** (`virtualization`) - the official recipe has none, so\n a large table is a rewrite rather than a prop.\n- **Inline editing** (`editable`) and cell-range selection\n (`enableCellSelection`).\n- **Grouping with aggregation** (`groupable`) and tree data (`treeData`).\n- **Export** to CSV, and to Excel / PDF on the paid tier.\n- **WAI-ARIA grid semantics and keyboard navigation**, instead of a plain\n `<table>` you keep accessible yourself.\n\n## What you give up\n\n- **A file you can read top to bottom.** The recipe's behaviour is\n visible in your repo; SvGrid's is behind a prop. That is the trade the\n whole page is about.\n- **Pixel-level control of the markup**, unless you go headless.\n- **Weight.** For a short static list that never sorts, the copied\n recipe is smaller.\n\n## Frequently asked questions\n\n### Does SvGrid replace shadcn-svelte?\n\nNo. It replaces one recipe in it. Your buttons, dialogs, and inputs stay\nexactly as they are, and the grid reads the same theme variables they do.\n\n### Can I keep my existing columns.ts?\n\nMostly. Rename `accessorKey` to `field`, drop the `header` render\nhelpers that only existed to draw a sort button, and the rest carries\nover. Custom cells become snippets via `renderSnippet`.\n\n### Do I have to give up owning the code?\n\nYou still own the call site - `npx @svgrid/ui add data-table` writes a\nfile into your project that you edit freely. What you stop owning is the\nrow-model plumbing, which is the part that broke on the v8 to v9 upgrade.\n\n### Is this worth it for a 50-row table?\n\nProbably not, unless you want editing, grouping, or export later. The\ncase gets strong when the table grows past what fits on a page, or when\nthe next feature request is one you would otherwise hand-build.\n\n## What you end up with\n\nSorting, filtering, pagination and selection - the shadcn data-table demo, as props.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} selectable />\n```\n\n## See also\n\n- [shadcn-svelte integration](./shadcn.md) - matching the theme\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - the\n underlying API map\n- [Migrating from a UI-kit table](./migrating-from-ui-kit-tables.md) - for\n the plain styled `<Table>`\n- [Why headless?](../why-headless.md) - keeping your own markup\n"
6162
6162
  },
6163
6163
  {
6164
6164
  "slug": "help/migrating-from-smart-grid",
6165
6165
  "path": "docs/help/migrating-from-smart-grid.md",
6166
6166
  "title": "Migrating from Smart.Grid (Smart Web Components)",
6167
6167
  "section": "Help",
6168
- "markdown": "\n# Migrating from Smart.Grid (Smart Web Components)\n\nSmart.Grid (htmlelements.com) and SvGrid come from the **same team**.\nSmart.Grid is a framework-agnostic custom element - it even works inside\nSvelte today - while SvGrid is the reactive, Svelte-5-native option. If\nyou are standardising a Svelte app on idiomatic components, this guide\nmaps a Smart.Grid setup onto SvGrid.\n\n> Estimated effort: **2-4 hours** per grid. The main shift is from a\n> custom element bound through properties / attributes to a reactive\n> Svelte component.\n\n<!-- facts:start smart-grid -->\n> **Facts, checked 9 Nov 2026.** `smart-webcomponents` 27.1.0, ISC, last published 2 Oct 2026, 5,690 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). Smart.Grid pricing, as its site states it: htmlelements.com lists the Developer License at $499 (on sale at $399) per developer, perpetual, or $39 per month, and the Team License at $1,499 per five developers; the free Community License covers the core components but not the Grid (https://www.htmlelements.com/license/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Smart.Grid (Smart Web Components)](https://svgrid.com/compare/smart-grid/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Smart.Grid (web component) | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `<smart-grid>` element | `<SvGrid ... />` |\n| `grid.dataSource = new Smart.DataAdapter`| `data` array (or `externalSort/Filter`) |\n| `{ label: 'Name', dataField: 'name' }` | `{ header: 'Name', field: 'name' }` |\n| `template` / cell template | `cell: (c) => renderSnippet(...)` |\n| `editing` + editor | `enableInlineEditing` + `editorType` |\n| `sorting` / `filtering` | `rowSortingFeature` / `columnFilteringFeature` |\n| `grouping` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `paging` + `pager` | `showPagination` |\n| element events (`change`, ...) | `onCellValueChange`, `onApiReady` |\n| `theme` attribute | `--sg-*` tokens / Tailwind |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <smart-grid id=\"grid\" sorting filtering paging></smart-grid>\n- <script>\n- const grid = document.querySelector('#grid')\n- grid.dataSource = new Smart.DataAdapter({ dataSource: rows, dataFields: [...] })\n- grid.columns = [\n- { label: 'Name', dataField: 'name' },\n- { label: 'Amount', dataField: 'amount' },\n- ]\n- </script>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## What changes\n\n- **Custom element + DataAdapter → reactive `$state`.** Data binding is a\n prop, not an assigned `dataSource`.\n- **Element templates → Svelte snippets** via `renderSnippet`.\n- **`theme` attribute → `--sg-*` tokens.**\n\n## A note on family\n\nSmart.Grid is a genuinely good framework-agnostic grid from the same\nteam, and it runs in Svelte as a web component. The reason to pick SvGrid\nis *idiomatic* Svelte: runes, snippets, and an MIT core. If you need one\ngrid across Angular, React, Vue, Svelte, and vanilla, Smart.Grid is the\nright call.\n\n## Frequently asked questions\n\n### Is SvGrid from the same team as Smart.Grid?\n\nYes. Smart Web Components (htmlelements.com) and SvGrid are built by the same\nteam. SvGrid is the Svelte-5-native option with an MIT community core.\n\n### Can I keep using Smart.Grid in Svelte instead?\n\nYes - Smart.Grid is a web component and works inside Svelte. Choose SvGrid when\nyou want an idiomatic Svelte component (runes + snippets) rather than a custom\nelement bound through properties.\n\n### How do Smart.Grid columns map to SvGrid?\n\n`label` becomes `header`, `dataField` becomes `field`, and element templates\nbecome `renderSnippet` cells. Sorting / filtering / grouping move to registered\nfeatures.\n\n## More examples\n\n### Smart.Chart integration\n\nMounts a <smart-chart> web component (htmlelements.com) and pipes the grid's displayed rows into its dataSource. Re-aggregates on every filter / sort.\n\n<div data-docs-demo=\"77-smart-chart\" data-height=\"460\"></div>\n\n## What you end up with\n\nSorting, filtering and editing with a footer summary.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable editable summary />\n```\n\n## The ported grid, running\n\nSorting, filtering, editing and a totals row - the working set a Smart grid\nusually carries, as props.\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n city: string\n age: number\n salary: number\n }\n\n const seed: Person[] = [\n { id: 1, name: 'Ada Lovelace', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', city: 'Portland', age: 54, salary: 155000 },\n ]\n\n let rows = $state<Person[]>(seed.map((p) => ({ ...p })))\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 180, editorType: 'text' },\n { field: 'city', header: 'City', width: 150, editorType: 'text' },\n { field: 'salary', header: 'Salary', width: 150, editorType: 'number', summary: 'sum',\n format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n\n<SvGrid data={rows} {columns} sortable filterable editable summary />\n```\n\n## See also\n\n- [SvGrid vs Smart.Grid](https://svgrid.com/compare/smart-grid/) - the side-by-side comparison\n- [About SvGrid](https://svgrid.com/about/) - built by jQWidgets / htmlelements.com\n- [Cell components](./cells/cell-components.md) - snippet-based custom cells\n"
6168
+ "markdown": "\n# Migrating from Smart.Grid (Smart Web Components)\n\nSmart.Grid (htmlelements.com) and SvGrid come from the **same team**.\nSmart.Grid is a framework-agnostic custom element - it even works inside\nSvelte today - while SvGrid is the reactive, Svelte-5-native option. If\nyou are standardising a Svelte app on idiomatic components, this guide\nmaps a Smart.Grid setup onto SvGrid.\n\n> Estimated effort: **2-4 hours** per grid. The main shift is from a\n> custom element bound through properties / attributes to a reactive\n> Svelte component.\n\n<!-- facts:start smart-grid -->\n> **Facts, checked 9 Nov 2026.** `smart-webcomponents` 27.1.0, ISC, last published 2 Oct 2026, 5,690 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). Smart.Grid pricing, as its site states it: htmlelements.com lists the Developer License at $499 (on sale at $399) per developer, perpetual, or $39 per month, and the Team License at $1,499 per five developers; the free Community License covers the core components but not the Grid (https://www.htmlelements.com/license/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Smart.Grid (Smart Web Components)](https://svgrid.com/compare/smart-grid/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Smart.Grid (web component) | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `<smart-grid>` element | `<SvGrid ... />` |\n| `grid.dataSource = new Smart.DataAdapter`| `data` array (or `externalSort/Filter`) |\n| `{ label: 'Name', dataField: 'name' }` | `{ header: 'Name', field: 'name' }` |\n| `template` / cell template | `cell: (c) => renderSnippet(...)` |\n| `editing` + editor | `enableInlineEditing` + `editorType` |\n| `sorting` / `filtering` | `rowSortingFeature` / `columnFilteringFeature` |\n| `grouping` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `paging` + `pager` | `showPagination` |\n| element events (`change`, ...) | `onCellValueChange`, `onApiReady` |\n| `theme` attribute | `--sg-*` tokens / Tailwind |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <smart-grid id=\"grid\" sorting filtering paging></smart-grid>\n- <script>\n- const grid = document.querySelector('#grid')\n- grid.dataSource = new Smart.DataAdapter({ dataSource: rows, dataFields: [...] })\n- grid.columns = [\n- { label: 'Name', dataField: 'name' },\n- { label: 'Amount', dataField: 'amount' },\n- ]\n- </script>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## What changes\n\n- **Custom element + DataAdapter → reactive `$state`.** Data binding is a\n prop, not an assigned `dataSource`.\n- **Element templates → Svelte snippets** via `renderSnippet`.\n- **`theme` attribute → `--sg-*` tokens.**\n\n## A note on family\n\nSmart.Grid is a genuinely good framework-agnostic grid from the same\nteam, and it runs in Svelte as a web component. The reason to pick SvGrid\nis *idiomatic* Svelte: runes, snippets, and an MIT core. If you need one\ngrid across Angular, React, Vue, Svelte, and vanilla, Smart.Grid is the\nright call.\n\n## Frequently asked questions\n\n### Is SvGrid from the same team as Smart.Grid?\n\nYes. Smart Web Components (htmlelements.com) and SvGrid are built by the same\nteam. SvGrid is the Svelte-5-native option with an MIT community core.\n\n### Can I keep using Smart.Grid in Svelte instead?\n\nYes - Smart.Grid is a web component and works inside Svelte. Choose SvGrid when\nyou want an idiomatic Svelte component (runes + snippets) rather than a custom\nelement bound through properties.\n\n### How do Smart.Grid columns map to SvGrid?\n\n`label` becomes `header`, `dataField` becomes `field`, and element templates\nbecome `renderSnippet` cells. Sorting / filtering / grouping move to registered\nfeatures.\n\n## More examples\n\n### Smart.Chart integration\n\nMounts a <smart-chart> web component (htmlelements.com) and pipes the grid's displayed rows into its dataSource. Re-aggregates on every filter / sort.\n\n<div data-docs-demo=\"77-smart-chart\" data-height=\"460\"></div>\n\n## What you end up with\n\nSorting, filtering and editing with a footer summary.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable editable summary />\n```\n\n## The ported grid, running\n\nSorting, filtering, editing and a totals row - the working set a Smart grid\nusually carries, as props.\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n city: string\n age: number\n salary: number\n }\n\n const seed: Person[] = [\n { id: 1, name: 'Ada Lovelace', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', city: 'Portland', age: 54, salary: 155000 },\n ]\n\n let rows = $state<Person[]>(seed.map((p) => ({ ...p })))\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 180, editorType: 'text' },\n { field: 'city', header: 'City', width: 150, editorType: 'text' },\n { field: 'salary', header: 'Salary', width: 150, editorType: 'number', summary: 'sum',\n format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n\n<SvGrid data={rows} {columns} sortable filterable editable summary />\n```\n\n## See also\n\n- [SvGrid vs Smart.Grid](https://svgrid.com/compare/smart-grid/) - the side-by-side comparison\n- [About SvGrid](https://svgrid.com/about/) - built by jQWidgets / htmlelements.com\n- [Cell components](./cells/cell-components.md) - snippet-based custom cells\n"
6169
6169
  },
6170
6170
  {
6171
6171
  "slug": "help/migrating-from-svar-svelte-datagrid",
6172
6172
  "path": "docs/help/migrating-from-svar-svelte-datagrid.md",
6173
6173
  "title": "Migrating from SVAR Svelte DataGrid",
6174
6174
  "section": "Help",
6175
- "markdown": "# Migrating from SVAR Svelte DataGrid\n\nSVAR Svelte DataGrid (`@svar-ui/svelte-grid`, formerly `wx-svelte-grid`) is, like SvGrid, an actual\nSvelte-native grid rather than a wrapper, and both are MIT-licensed and\nfree for commercial use. Because both are component-first with\narray-of-object column definitions, a port is mostly a configuration\ntranslation. The reasons people switch are **capability** (SvGrid adds\nrow grouping with aggregation, master-detail, Excel-style cell-range\nselection, integrated charts, and pivot) and **architecture** (SvGrid\nalso exposes a headless engine and an MCP server).\n\n> Estimated effort: **2-4 hours** per grid - a prop and event rename\n> pass, plus re-theming.\n>\n> SVAR's exact prop and event names evolve across releases; check the\n> current SVAR docs and map them onto the SvGrid equivalents below.\n\n<!-- facts:start svar-svelte-datagrid -->\n> **Facts, checked 9 Nov 2026.** `@svar-ui/svelte-grid` 2.7.6, MIT, last published 7 Nov 2026, 106,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `@svar-ui/svelte-grid` 2.7.6 53.1 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). SVAR Svelte DataGrid pricing, as its site states it: The DataGrid is MIT and free for commercial use with no paid tier; svar.dev sells PRO editions of its Gantt, Calendar and Kanban components and offers paid consulting and custom development, with no price list for the grid (https://svar.dev/svelte/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/).\n<!-- facts:end -->\n\n## Concept map\n\n| SVAR DataGrid (@svar-ui/svelte-grid) | sv-grid |\n| ----------------------------------------- | ----------------------------------------- |\n| `<Grid {data} {columns} />` | `<SvGrid data={...} columns={...} />` |\n| `columns: [{ id, header, width, ... }]` | `columns: [{ field, header, width, ... }]` |\n| `sizes={{ rowHeight, columnWidth }}` | `rowHeight` prop, `width` per column |\n| `autoRowHeight` | `autoRowHeight` |\n| Per-column `editor` / `template` | `editorType` + `cell` snippet |\n| `sort: true` on a column, `sortMarks` | `rowSortingFeature`; `initialSorting` |\n| Header filters, the Filter component | `columnFilteringFeature`; Excel-style menu and filter row built in |\n| `tree` prop + `data` with children | `treeData={{ parentField }}` or `rowExpandingFeature` with sub rows |\n| `select` / `multiselect` | `selectionMode=\"row\"` (+ checkbox column) |\n| `split={{ left: n }}` (pinned columns) | `initialColumnPinning` or `pinned: 'left' \\| 'right'` per column |\n| `reorder` (row drag) | `rowDragManaged` |\n| `undo` prop, undo/redo hotkeys | `history` prop; `api.undo()` / `api.redo()` |\n| Context menu component | `contextMenu` prop with items |\n| `init={(api) => ...}` | `onApiReady={(api) => ...}` |\n| `api.exec('sort-rows', { key, order })` | `api.setSort(field, 'asc' \\| 'desc')` |\n| `api.exec('filter-rows', { filter })` | `api.setFilter(field, { operator, value })` |\n| `api.on('...')` event bus | `onXxx` callback props |\n| Export to CSV, print API | `api.exportCsv()` (free); print in `@svgrid/enterprise` |\n| `RestDataProvider` | server-side row model / `externalSort` + `externalFilter` |\n| Theme / skin components (Willow, Material)| `--sg-*` CSS custom properties / Tailwind |\n\n## Before / after (shape, not exact prop names)\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <script>\n- import { Grid } from '@svar-ui/svelte-grid'\n- const columns = [\n- { id: 'name', header: 'Name', width: 200 },\n- { id: 'amount', header: 'Amount', width: 120, editor: 'text' },\n- ]\n- </script>\n-\n- <Grid {data} {columns} />\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name', width: 200 },\n+ { field: 'amount', header: 'Amount', width: 120, editorType: 'text' },\n+ ]\n+ </script>\n+\n+ <SvGrid data={data} columns={columns} features={features} enableInlineEditing />\n```\n\n## Licensing note\n\nBoth grids are MIT and free for commercial use, so licensing is not the\ndeciding factor. The difference is monetization: SVAR keeps the whole\ngrid free (including CSV export and print) and sells PRO editions of its\nGantt, Calendar and Kanban instead, whereas SvGrid's `@svgrid/grid` core is\nMIT and the optional `@svgrid/enterprise` pack (XLSX/PDF export, print,\npivot, import, support) is the paid piece. CSV, TSV and JSON export and the\nin-grid AI helpers are free on SvGrid too. If all you need is CSV export and\nprint, SVAR gives both free; if you need pivot, Excel or PDF output, or a\nsupport SLA, that is Enterprise on SvGrid.\n\n## What you get with SvGrid\n\n- **Row grouping with aggregation**, **master-detail**, **pinned rows**,\n and left/right column freezing - beyond SVAR's documented surface.\n- A **headless engine** (`createSvGrid` + row models) in addition to the\n component, if you want to compose your own layer.\n- **Excel-style filter menu**, **cell-range selection + TSV copy**, a\n **fill handle**, **integrated charts**, and a documented **imperative API**.\n- **In-grid AI helpers** (natural-language filter, smart fill, summarise),\n free against a model you register, and **@svgrid/mcp** so AI assistants\n answer accurately about your grid and can check the code they write.\n\nThe two features most SVAR teams port for:\n\n<div data-docs-demo=\"07-grouping-aggregation\" data-height=\"520\"></div>\n\n<div data-docs-demo=\"181-master-detail-grid\" data-height=\"520\"></div>\n\n## What you give up\n\n- **Free printing.** SVAR's print API is in the MIT grid; SvGrid's print\n view is in `@svgrid/enterprise`.\n- **The React and Vue versions** of the same grid. SvGrid is Svelte-only.\n- **A smaller bundle.** The facts box at the top of the page has both\n packages measured the same way; the difference is the grouping,\n master/detail, range editing, charts and server-side data (flat paging and infinite scroll free; grouping, tree, pivot and transactions on the server are Enterprise) SvGrid\n carries, and features you never import load as separate chunks.\n- **The SVAR suite around the grid.** Gantt, Calendar, Kanban and File\n Manager share SVAR's core; SvGrid's Kanban and scheduler are views of the\n grid in `@svgrid/enterprise`.\n\n## What to check on the SVAR side\n\n- The **wider SVAR suite** (Gantt, Calendar, Kanban) - if you use several\n SVAR components together, staying on SVAR may be simpler.\n- Any **SVAR-specific column features** you rely on; map each to a SvGrid\n `cell` snippet, `editorType`, or feature before porting.\n- **Filter functions.** SVAR filters with a predicate you write; SvGrid's\n filter takes an operator and a value per column, and `externalFilter`\n hands you the rows when you want to keep a custom predicate.\n\n## Frequently asked questions\n\n### Why move from SVAR Svelte DataGrid to SvGrid?\n\nCapability and architecture, not licensing (both are MIT). SvGrid adds row\ngrouping with aggregation, master-detail, Excel-style cell-range selection,\nintegrated charts, and pivot, and it ships a headless engine plus an MCP server\nalongside the component. Both are Svelte-native, so the core grid\nbehaviour is comparable; SvGrid covers more ground.\n\n### Is the migration a rewrite?\n\nNo. Both are component-first with array-of-object columns, so it is mostly a\nprop / event rename pass (`id` → `field`, edit events → `onCellValueChange`)\nplus re-theming through `--sg-*` tokens.\n\n### Are SvGrid and SVAR both free and MIT?\n\nYes. `@svgrid/grid` and the SVAR DataGrid are both MIT and free for\ncommercial, closed-source use. SVAR keeps the whole grid free and sells PRO\neditions of its Gantt, Calendar and Kanban; on SvGrid, only the optional\n`@svgrid/enterprise` add-on (Excel and PDF export, print, pivot, import,\nsupport) is paid. The dated statement from svar.dev is in the facts box at\nthe top of this page.\n\n### How does SVAR's tree data map to SvGrid?\n\nSVAR takes hierarchical data with the `tree` prop. SvGrid takes either a\nflat array with a parent field (`treeData={{ parentField: 'parentId' }}`)\nor nested rows through `rowExpandingFeature`, and the same grid also does\nmaster/detail, which SVAR's docs do not list. The example below runs the\nflat shape:\n\n## What you end up with\n\nGrouping with aggregation and a totals row, the main capability step up.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} summary groupable sortable filterable />\n```\n\nAnd the tree shape, from the same rows with a parent link, which is what SVAR's `tree` prop becomes:\n\n```svelte {runnable}\n<SvGrid data={rows.map((r, i) => ({ ...r, parentId: i === 0 ? null : rows[0].id }))} {columns} treeData={{ parentField: 'parentId' }} sortable />\n```\n\n## See also\n\n- [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/) - the side-by-side comparison, with a source and date for every claim and the measured benchmark\n- [Grouping and aggregation](./grouping-aggregation.md) - the first demo above\n- [Master/detail](./rows/master-detail.md) - the second demo above\n- [Architecture](./architecture.md) - the engine + render-component split\n- [Cell components](./cells/cell-components.md) - custom cells and editors\n"
6175
+ "markdown": "# Migrating from SVAR Svelte DataGrid\n\nSVAR Svelte DataGrid (`@svar-ui/svelte-grid`, formerly `wx-svelte-grid`) is, like SvGrid, an actual\nSvelte-native grid rather than a wrapper, and both are MIT-licensed and\nfree for commercial use. Because both are component-first with\narray-of-object column definitions, a port is mostly a configuration\ntranslation. The reasons people switch are **capability** (SvGrid adds\nrow grouping with aggregation, master-detail, Excel-style cell-range\nselection, integrated charts, and pivot) and **architecture** (SvGrid\nalso exposes a headless engine and an MCP server).\n\n> Estimated effort: **2-4 hours** per grid - a prop and event rename\n> pass, plus re-theming.\n>\n> SVAR's exact prop and event names evolve across releases; check the\n> current SVAR docs and map them onto the SvGrid equivalents below.\n\n<!-- facts:start svar-svelte-datagrid -->\n> **Facts, checked 9 Nov 2026.** `@svar-ui/svelte-grid` 2.7.6, MIT, last published 7 Nov 2026, 106,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `@svar-ui/svelte-grid` 2.7.6 53.1 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). SVAR Svelte DataGrid pricing, as its site states it: The DataGrid is MIT and free for commercial use with no paid tier; svar.dev sells PRO editions of its Gantt, Calendar and Kanban components and offers paid consulting and custom development, with no price list for the grid (https://svar.dev/svelte/, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/).\n<!-- facts:end -->\n\n## Concept map\n\n| SVAR DataGrid (@svar-ui/svelte-grid) | sv-grid |\n| ----------------------------------------- | ----------------------------------------- |\n| `<Grid {data} {columns} />` | `<SvGrid data={...} columns={...} />` |\n| `columns: [{ id, header, width, ... }]` | `columns: [{ field, header, width, ... }]` |\n| `sizes={{ rowHeight, columnWidth }}` | `rowHeight` prop, `width` per column |\n| `autoRowHeight` | `autoRowHeight` |\n| Per-column `editor` / `template` | `editorType` + `cell` snippet |\n| `sort: true` on a column, `sortMarks` | `rowSortingFeature`; `initialSorting` |\n| Header filters, the Filter component | `columnFilteringFeature`; Excel-style menu and filter row built in |\n| `tree` prop + `data` with children | `treeData={{ parentField }}` or `rowExpandingFeature` with sub rows |\n| `select` / `multiselect` | `selectionMode=\"row\"` (+ checkbox column) |\n| `split={{ left: n }}` (pinned columns) | `initialColumnPinning` or `pinned: 'left' \\| 'right'` per column |\n| `reorder` (row drag) | `rowDragManaged` |\n| `undo` prop, undo/redo hotkeys | `history` prop; `api.undo()` / `api.redo()` |\n| Context menu component | `contextMenu` prop with items |\n| `init={(api) => ...}` | `onApiReady={(api) => ...}` |\n| `api.exec('sort-rows', { key, order })` | `api.setSort(field, 'asc' \\| 'desc')` |\n| `api.exec('filter-rows', { filter })` | `api.setFilter(field, { operator, value })` |\n| `api.on('...')` event bus | `onXxx` callback props |\n| Export to CSV, print API | `api.exportCsv()` (free); print in `@svgrid/enterprise` |\n| `RestDataProvider` | server-side row model / `externalSort` + `externalFilter` |\n| Theme / skin components (Willow, Material)| `--sg-*` CSS custom properties / Tailwind |\n\n## Before / after (shape, not exact prop names)\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <script>\n- import { Grid } from '@svar-ui/svelte-grid'\n- const columns = [\n- { id: 'name', header: 'Name', width: 200 },\n- { id: 'amount', header: 'Amount', width: 120, editor: 'text' },\n- ]\n- </script>\n-\n- <Grid {data} {columns} />\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name', width: 200 },\n+ { field: 'amount', header: 'Amount', width: 120, editorType: 'text' },\n+ ]\n+ </script>\n+\n+ <SvGrid data={data} columns={columns} features={features} enableInlineEditing />\n```\n\n## Licensing note\n\nBoth grids are MIT and free for commercial use, so licensing is not the\ndeciding factor. The difference is monetization: SVAR keeps the whole\ngrid free (including CSV export and print) and sells PRO editions of its\nGantt, Calendar and Kanban instead, whereas SvGrid's `@svgrid/grid` core is\nMIT and the optional `@svgrid/enterprise` pack (XLSX/PDF export, print,\npivot, import, support) is the paid piece. CSV, TSV and JSON export and the\nin-grid AI helpers are free on SvGrid too. If all you need is CSV export and\nprint, SVAR gives both free; if you need pivot, Excel or PDF output, or a\nsupport SLA, that is Enterprise on SvGrid.\n\n## What you get with SvGrid\n\n- **Row grouping with aggregation**, **master-detail**, **pinned rows**,\n and left/right column freezing - beyond SVAR's documented surface.\n- A **headless engine** (`createSvGrid` + row models) in addition to the\n component, if you want to compose your own layer.\n- **Excel-style filter menu**, **cell-range selection + TSV copy**, a\n **fill handle**, **integrated charts**, and a documented **imperative API**.\n- **In-grid AI helpers** (natural-language filter, smart fill, summarise),\n free against a model you register, and **@svgrid/mcp** so AI assistants\n answer accurately about your grid and can check the code they write.\n\nThe two features most SVAR teams port for:\n\n<div data-docs-demo=\"07-grouping-aggregation\" data-height=\"520\"></div>\n\n<div data-docs-demo=\"181-master-detail-grid\" data-height=\"520\"></div>\n\n## What you give up\n\n- **Free printing.** SVAR's print API is in the MIT grid; SvGrid's print\n view is in `@svgrid/enterprise`.\n- **The React and Vue versions** of the same grid. SvGrid is Svelte-only.\n- **A smaller bundle.** The facts box at the top of the page has both\n packages measured the same way; the difference is the grouping,\n master/detail, range editing, charts and server-side data (flat paging and infinite scroll free; grouping, tree, pivot and transactions on the server are Enterprise) SvGrid\n carries, and features you never import load as separate chunks.\n- **The SVAR suite around the grid.** Gantt, Calendar, Kanban and File\n Manager share SVAR's core; SvGrid's Kanban and scheduler are views of the\n grid in `@svgrid/enterprise`.\n\n## What to check on the SVAR side\n\n- The **wider SVAR suite** (Gantt, Calendar, Kanban) - if you use several\n SVAR components together, staying on SVAR may be simpler.\n- Any **SVAR-specific column features** you rely on; map each to a SvGrid\n `cell` snippet, `editorType`, or feature before porting.\n- **Filter functions.** SVAR filters with a predicate you write; SvGrid's\n filter takes an operator and a value per column, and `externalFilter`\n hands you the rows when you want to keep a custom predicate.\n\n## Frequently asked questions\n\n### Why move from SVAR Svelte DataGrid to SvGrid?\n\nCapability and architecture, not licensing (both are MIT). SvGrid adds row\ngrouping with aggregation, master-detail, Excel-style cell-range selection,\nintegrated charts, and pivot, and it ships a headless engine plus an MCP server\nalongside the component. Both are Svelte-native, so the core grid\nbehaviour is comparable; SvGrid covers more ground.\n\n### Is the migration a rewrite?\n\nNo. Both are component-first with array-of-object columns, so it is mostly a\nprop / event rename pass (`id` → `field`, edit events → `onCellValueChange`)\nplus re-theming through `--sg-*` tokens.\n\n### Are SvGrid and SVAR both free and MIT?\n\nYes. `@svgrid/grid` and the SVAR DataGrid are both MIT and free for\ncommercial, closed-source use. SVAR keeps the whole grid free and sells PRO\neditions of its Gantt, Calendar and Kanban; on SvGrid, only the optional\n`@svgrid/enterprise` add-on (Excel and PDF export, print, pivot, import,\nsupport) is paid. The dated statement from svar.dev is in the facts box at\nthe top of this page.\n\n### How does SVAR's tree data map to SvGrid?\n\nSVAR takes hierarchical data with the `tree` prop. SvGrid takes either a\nflat array with a parent field (`treeData={{ parentField: 'parentId' }}`)\nor nested rows through `rowExpandingFeature`, and the same grid also does\nmaster/detail, which SVAR's docs do not list. The example below runs the\nflat shape:\n\n## What you end up with\n\nGrouping with aggregation and a totals row, the main capability step up.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} summary groupable sortable filterable />\n```\n\nAnd the tree shape, from the same rows with a parent link, which is what SVAR's `tree` prop becomes:\n\n```svelte {runnable}\n<SvGrid data={rows.map((r, i) => ({ ...r, parentId: i === 0 ? null : rows[0].id }))} {columns} treeData={{ parentField: 'parentId' }} sortable />\n```\n\n## See also\n\n- [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/) - the side-by-side comparison, with a source and date for every claim and the measured benchmark\n- [Grouping and aggregation](./grouping-aggregation.md) - the first demo above\n- [Master/detail](./rows/master-detail.md) - the second demo above\n- [Architecture](./architecture.md) - the engine + render-component split\n- [Cell components](./cells/cell-components.md) - custom cells and editors\n"
6176
6176
  },
6177
6177
  {
6178
6178
  "slug": "help/migrating-from-svelte-headless-table",
6179
6179
  "path": "docs/help/migrating-from-svelte-headless-table.md",
6180
6180
  "title": "Migrating from svelte-headless-table",
6181
6181
  "section": "Help",
6182
- "markdown": "\n# Migrating from svelte-headless-table\n\nsvelte-headless-table popularised the headless-table pattern in the\nSvelte ecosystem. It is built on Svelte 4 stores and plugins\n(`addSortBy`, `addColumnFilters`, `addPagination`, ...), and it leaves\nthe markup to you. SvGrid keeps the same headless idea but runs on\nSvelte 5 runes and ships a render component, so the port mostly\n*removes* the table you used to hand-author.\n\n> Estimated effort: **1-3 hours** per grid. Most of the time is deleting\n> the `<table>` markup and `Subscribe` blocks you no longer need.\n\n<!-- facts:start svelte-headless-table pkg:@humanspeak/svelte-headless-table -->\n> **Facts, checked 9 Nov 2026.** `svelte-headless-table` 0.18.3, MIT, last published 28 Oct 2024, 96,400 npm downloads in the 30 days to 8 Nov 2026. `@humanspeak/svelte-headless-table` 7.1.0, MIT, last published 28 Oct 2026, 10,600 npm downloads in the same window. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `svelte-headless-table` 0.18.3 5.7 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). svelte-headless-table pricing, as its site states it: svelte-headless-table is MIT and free; no paid licence or support is offered, and the author states there are no plans for a Svelte 5 port (https://github.com/bryanmylee/svelte-headless-table, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/).\n<!-- facts:end -->\n\n## Know your options first\n\nThe original package's last release and its Svelte 4 peer range are in the\nfacts box above, and its README states there are no current plans for a\nSvelte 5 port, so a Svelte 5 upgrade forces a decision. There are three\nhonest answers and you should know all of them:\n\n1. **`@humanspeak/svelte-headless-table`** - a maintained fork on Svelte 5 with\n the same API; its version, publish date and downloads are in the facts box\n too. Changing one package name is the cheapest path by a wide margin. If\n your table works and you only need Svelte 5, do that.\n2. **TanStack Table v9** - its Svelte adapter declares a Svelte 5 peer range.\n Still headless-only, so you keep writing and maintaining the markup.\n3. **SvGrid** - this page. A different trade: you delete the markup and take a\n renderer instead.\n\nPick SvGrid when the markup is the part you are tired of. Note that the blocker\nis not the syntax and not really the install either: `svelte-headless-table@0.18.3`\ndeclares `svelte@^4`, which npm 11 overrides with a warning (older npm needed\n`--legacy-peer-deps`), and the untouched component then type-checks clean in\nlegacy mode. Slots and `let:` still work in Svelte 5 - they are deprecated in\nfavour of snippets, not removed - so the `Subscribe` blocks below keep\nrendering. The trouble starts when that component moves to runes: the library\ntakes a store, your data is a rune, and the bridge between them is yours to\nwrite and maintain. We ran the whole upgrade and captured every line of it in\n[Upgrading a Svelte 4 table to Svelte 5](./svelte-5-upgrade-data-tables.md).\n\n## Run the codemod\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```bash\nnpx @svgrid/migrate # preview the result\nnpx @svgrid/migrate src --write # apply it\n```\n\nIt translates column definitions and plugin config, deletes the\n`Subscribe`/`Render` scaffolding, and reports anything it cannot map rather than\ndropping it silently. It previews by default. See\n[`@svgrid/migrate`](https://www.npmjs.com/package/@svgrid/migrate) for the full\nmapping and its limits.\n\n## Vocabulary cheat sheet\n\n| svelte-headless-table | sv-grid |\n| ----------------------------------------- | ----------------------------------------- |\n| `createTable(data, plugins)` | `createSvGrid({...})` or `<SvGrid>` |\n| `table.createColumns((t) => [...])` | `columns: ColumnDef[]` |\n| `t.column({ accessor: 'x', header })` | `{ field: 'x', header }` |\n| `t.column({ accessor: (r) => ... })` | `{ id, fieldFn: (r) => ... }` |\n| `t.group({ header, columns })` | `{ header, columns: [...] }` (column group) |\n| `addSortBy()` | `rowSortingFeature` |\n| `addColumnFilters()` / `addTableFilter()` | `columnFilteringFeature` |\n| `addPagination()` | Built in; toggle `showPagination` |\n| `addExpandedRows()` / `addSubRows()` | `rowExpandingFeature` |\n| `addGroupBy()` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `addSelectedRows()` | `rowSelectionFeature` |\n| `addDataExport()` | `@svgrid/enterprise` export pack |\n| `createViewModel(columns)` + `Subscribe` | `<SvGrid>` (no view model to wire) |\n| `pluginStates.sort.sortKeys` | `api.setSort(id, dir)` / `onSortingChange` |\n\n### The shortcut form\n\nThat table maps plugins onto the headless *features*, which is what you want if\nyou keep driving the engine yourself. If you are moving to the `<SvGrid>`\ncomponent, every capability also has a boolean prop, and that is what the\ncodemod emits:\n\n| Plugin | `<SvGrid>` prop |\n| --- | --- |\n| `addSortBy()` | `sortable` |\n| `addTableFilter()` | `filterable showGlobalFilter` |\n| `addColumnFilters()` | `filterable showColumnFilters` |\n| `addPagination({ initialPageSize: 25 })` | `pageable pageSize={25}` |\n| `addSelectedRows()` | `showRowSelection` |\n| `addGroupBy()` | `groupable` |\n| `addSubRows()` | `treeData` |\n| `addColumnOrder()` | `enableColumnReorder` |\n| `addResizedColumns()` | nothing - resizing is built in |\n| `addHiddenColumns()` | `visible: false` on the column |\n| `addExpandedRows()` | `treeData`, or `isDetailRow` + `renderDetailRow` |\n| `addGridLayout()` | nothing - SvGrid owns its layout |\n\nEvery capability is off by default and turned on by its prop; there is no\nplugin registration step.\n\n## Before / after\n\n```diff\n- <script>\n- import { createTable } from 'svelte-headless-table'\n- import { addSortBy, addColumnFilters, addPagination } from 'svelte-headless-table/plugins'\n- import { readable } from 'svelte/store'\n-\n- const table = createTable(readable(data), {\n- sort: addSortBy(), filter: addColumnFilters(), page: addPagination(),\n- })\n- const columns = table.createColumns((t) => [\n- t.column({ accessor: 'name', header: 'Name' }),\n- t.column({ accessor: 'amount', header: 'Amount' }),\n- ])\n- const { headerRows, rows, tableAttrs, tableBodyAttrs } = table.createViewModel(columns)\n- </script>\n-\n- <table {...$tableAttrs}>\n- <thead> ...Subscribe over headerRows... </thead>\n- <tbody {...$tableBodyAttrs}> ...Subscribe over rows... </tbody>\n- </table>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\nType the column array against your row type. `GridColumns<(typeof data)[number]>`\nis the shortest form; a bare `GridColumns` widens the row to\n`Record<string, unknown>` and stops checking `field` against your real keys.\n\n## What you get for free\n\n- **The renderer.** No `createViewModel`, no `Subscribe`, no hand-built\n `<table>`. SvGrid ships virtualization, sticky headers, column\n resize, keyboard nav, and ARIA.\n- **Excel-style filter menu** and **cell-range selection + TSV copy**,\n which are BYO in svelte-headless-table.\n- **Inline editing** with typed editors and validation hooks.\n- **In-grid AI helpers** - natural-language filter, smart fill, summarise -\n free in `@svgrid/grid` against a model you register.\n- **Enterprise features** - Excel and PDF export, print, import and pivot -\n in one paid add-on; CSV, TSV and JSON export are free.\n\nIf you would rather keep driving your own `<table>`, the headless demo\nbelow does exactly that on `createSvGrid`, plugins replaced by features:\n\n<div data-docs-demo=\"186-headless-table\" data-height=\"520\"></div>\n\n## What changes\n\n- **Stores → runes.** Reactive data is `$state` / a plain array, not a\n Svelte store you pass to `createTable`. If your rows came from a `derived`\n store, read it with `$` at the call site or move it to `$derived`.\n- **Plugins → features.** Register `rowSortingFeature` etc. in\n `tableFeatures({...})` instead of `addSortBy()` in the plugin bag.\n- **View model → component.** You stop owning the markup; style through\n `--sg-*` tokens (and Tailwind) instead of `tableAttrs`.\n\n## What does not come across\n\n- **Custom `cell` renderers.** `createRender(MyComponent, props)` has no direct\n equivalent; SvGrid uses a `cell` snippet. The codemod preserves yours as a\n `// TODO port:` comment.\n- **Your own plugins.** There is no plugin system to port them into.\n- **`tableAttrs` / `attrs()` spreading.** The component owns its attributes.\n\n## When not to migrate\n\n- You need control of the exact DOM. Use the fork, or SvGrid's\n [headless core](../why-headless.md) at `@svgrid/grid/core`.\n- Your table is a handful of rows with no interaction. A plain `{#each}` is\n less code than either library.\n- You depend on custom plugins. That is a rewrite, not a migration.\n- You only need Svelte 5 support. Use the fork.\n\n## Frequently asked questions\n\n### How hard is it to move from svelte-headless-table to SvGrid?\n\nUsually 1-3 hours per grid, and `npx @svgrid/migrate` does the mechanical part.\nThe plugin-to-feature mapping is almost one-to-one; most of the work is deleting\nthe `createViewModel` + `Subscribe` + `<table>` markup, because SvGrid renders\nthat for you.\n\n### Does SvGrid use Svelte 5 runes instead of stores?\n\nYes. svelte-headless-table is built on Svelte 4 stores; SvGrid is Svelte-5\nnative (`$state` / `$derived` / `$effect`) with snippets for custom cells.\n\n### Is SvGrid still headless like svelte-headless-table?\n\nYes - `createSvGrid` plus the row-model factories is a headless engine you can\ndrive with your own markup, importable on its own from `@svgrid/grid/core`. The\ndifference is that SvGrid *also* ships a batteries-included `<SvGrid>` component\nso you usually do not have to.\n\n### Is svelte-headless-table still maintained?\n\nThe original's last publish date and Svelte 4 peer range are in the facts box\nat the top of the page, read from npm on the date shown, and its README says\nthere are no current plans for a Svelte 5 port. A community fork,\n`@humanspeak/svelte-headless-table`, is maintained and runs on Svelte 5 with\nthe same API; the facts box has its release line too. The fork is the\nlowest-effort option if you are happy with the library and only need Svelte 5.\n\n## What you end up with\n\nSorting, filtering and selection, on a library that still ships for Svelte 5.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable selectable />\n```\n\n## See also\n\n- [Upgrading a Svelte 4 table to Svelte 5](./svelte-5-upgrade-data-tables.md) - the whole upgrade run and captured, step by step\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/) - the side-by-side comparison, with a source and date for every claim\n- [Headless: build a table](./headless/build-a-table.md) - the demo above, step by step\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - sibling headless guide\n- [Why headless?](../why-headless.md) - the design rationale\n- [Architecture](./architecture.md) - the engine + render-component split\n"
6182
+ "markdown": "\n# Migrating from svelte-headless-table\n\nsvelte-headless-table popularised the headless-table pattern in the\nSvelte ecosystem. It is built on Svelte 4 stores and plugins\n(`addSortBy`, `addColumnFilters`, `addPagination`, ...), and it leaves\nthe markup to you. SvGrid keeps the same headless idea but runs on\nSvelte 5 runes and ships a render component, so the port mostly\n*removes* the table you used to hand-author.\n\n> Estimated effort: **1-3 hours** per grid. Most of the time is deleting\n> the `<table>` markup and `Subscribe` blocks you no longer need.\n\n<!-- facts:start svelte-headless-table pkg:@humanspeak/svelte-headless-table -->\n> **Facts, checked 9 Nov 2026.** `svelte-headless-table` 0.18.3, MIT, last published 28 Oct 2024, 96,400 npm downloads in the 30 days to 8 Nov 2026. `@humanspeak/svelte-headless-table` 7.1.0, MIT, last published 28 Oct 2026, 10,600 npm downloads in the same window. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `svelte-headless-table` 0.18.3 5.7 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). svelte-headless-table pricing, as its site states it: svelte-headless-table is MIT and free; no paid licence or support is offered, and the author states there are no plans for a Svelte 5 port (https://github.com/bryanmylee/svelte-headless-table, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/).\n<!-- facts:end -->\n\n## Know your options first\n\nThe original package's last release and its Svelte 4 peer range are in the\nfacts box above, and its README states there are no current plans for a\nSvelte 5 port, so a Svelte 5 upgrade forces a decision. There are three\nhonest answers and you should know all of them:\n\n1. **`@humanspeak/svelte-headless-table`** - a maintained fork on Svelte 5 with\n the same API; its version, publish date and downloads are in the facts box\n too. Changing one package name is the cheapest path by a wide margin. If\n your table works and you only need Svelte 5, do that.\n2. **TanStack Table v9** - its Svelte adapter declares a Svelte 5 peer range.\n Still headless-only, so you keep writing and maintaining the markup.\n3. **SvGrid** - this page. A different trade: you delete the markup and take a\n renderer instead.\n\nPick SvGrid when the markup is the part you are tired of. Note that the blocker\nis not the syntax and not really the install either: `svelte-headless-table@0.18.3`\ndeclares `svelte@^4`, which npm 11 overrides with a warning (older npm needed\n`--legacy-peer-deps`), and the untouched component then type-checks clean in\nlegacy mode. Slots and `let:` still work in Svelte 5 - they are deprecated in\nfavour of snippets, not removed - so the `Subscribe` blocks below keep\nrendering. The trouble starts when that component moves to runes: the library\ntakes a store, your data is a rune, and the bridge between them is yours to\nwrite and maintain. We ran the whole upgrade and captured every line of it in\n[Upgrading a Svelte 4 table to Svelte 5](./svelte-5-upgrade-data-tables.md).\n\n## Run the codemod\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```bash\nnpx @svgrid/migrate # preview the result\nnpx @svgrid/migrate src --write # apply it\n```\n\nIt translates column definitions and plugin config, deletes the\n`Subscribe`/`Render` scaffolding, and reports anything it cannot map rather than\ndropping it silently. It previews by default. See\n[`@svgrid/migrate`](https://www.npmjs.com/package/@svgrid/migrate) for the full\nmapping and its limits.\n\n## Vocabulary cheat sheet\n\n| svelte-headless-table | sv-grid |\n| ----------------------------------------- | ----------------------------------------- |\n| `createTable(data, plugins)` | `createSvGrid({...})` or `<SvGrid>` |\n| `table.createColumns((t) => [...])` | `columns: ColumnDef[]` |\n| `t.column({ accessor: 'x', header })` | `{ field: 'x', header }` |\n| `t.column({ accessor: (r) => ... })` | `{ id, fieldFn: (r) => ... }` |\n| `t.group({ header, columns })` | `{ header, columns: [...] }` (column group) |\n| `addSortBy()` | `rowSortingFeature` |\n| `addColumnFilters()` / `addTableFilter()` | `columnFilteringFeature` |\n| `addPagination()` | Built in; toggle `showPagination` |\n| `addExpandedRows()` / `addSubRows()` | `rowExpandingFeature` |\n| `addGroupBy()` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `addSelectedRows()` | `rowSelectionFeature` |\n| `addDataExport()` | `@svgrid/enterprise` export pack |\n| `createViewModel(columns)` + `Subscribe` | `<SvGrid>` (no view model to wire) |\n| `pluginStates.sort.sortKeys` | `api.setSort(id, dir)` / `onSortingChange` |\n\n### The shortcut form\n\nThat table maps plugins onto the headless *features*, which is what you want if\nyou keep driving the engine yourself. If you are moving to the `<SvGrid>`\ncomponent, every capability also has a boolean prop, and that is what the\ncodemod emits:\n\n| Plugin | `<SvGrid>` prop |\n| --- | --- |\n| `addSortBy()` | `sortable` |\n| `addTableFilter()` | `filterable showGlobalFilter` |\n| `addColumnFilters()` | `filterable showColumnFilters` |\n| `addPagination({ initialPageSize: 25 })` | `pageable pageSize={25}` |\n| `addSelectedRows()` | `showRowSelection` |\n| `addGroupBy()` | `groupable` |\n| `addSubRows()` | `treeData` |\n| `addColumnOrder()` | `enableColumnReorder` |\n| `addResizedColumns()` | nothing - resizing is built in |\n| `addHiddenColumns()` | `visible: false` on the column |\n| `addExpandedRows()` | `treeData`, or `isDetailRow` + `renderDetailRow` |\n| `addGridLayout()` | nothing - SvGrid owns its layout |\n\nEvery capability is off by default and turned on by its prop; there is no\nplugin registration step.\n\n## Before / after\n\n```diff\n- <script>\n- import { createTable } from 'svelte-headless-table'\n- import { addSortBy, addColumnFilters, addPagination } from 'svelte-headless-table/plugins'\n- import { readable } from 'svelte/store'\n-\n- const table = createTable(readable(data), {\n- sort: addSortBy(), filter: addColumnFilters(), page: addPagination(),\n- })\n- const columns = table.createColumns((t) => [\n- t.column({ accessor: 'name', header: 'Name' }),\n- t.column({ accessor: 'amount', header: 'Amount' }),\n- ])\n- const { headerRows, rows, tableAttrs, tableBodyAttrs } = table.createViewModel(columns)\n- </script>\n-\n- <table {...$tableAttrs}>\n- <thead> ...Subscribe over headerRows... </thead>\n- <tbody {...$tableBodyAttrs}> ...Subscribe over rows... </tbody>\n- </table>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\nType the column array against your row type. `GridColumns<(typeof data)[number]>`\nis the shortest form; a bare `GridColumns` widens the row to\n`Record<string, unknown>` and stops checking `field` against your real keys.\n\n## What you get for free\n\n- **The renderer.** No `createViewModel`, no `Subscribe`, no hand-built\n `<table>`. SvGrid ships virtualization, sticky headers, column\n resize, keyboard nav, and ARIA.\n- **Excel-style filter menu** and **cell-range selection + TSV copy**,\n which are BYO in svelte-headless-table.\n- **Inline editing** with typed editors and validation hooks.\n- **In-grid AI helpers** - natural-language filter, smart fill, summarise -\n free in `@svgrid/grid` against a model you register.\n- **Enterprise features** - Excel and PDF export, print, import and pivot -\n in one paid add-on; CSV, TSV and JSON export are free.\n\nIf you would rather keep driving your own `<table>`, the headless demo\nbelow does exactly that on `createSvGrid`, plugins replaced by features:\n\n<div data-docs-demo=\"186-headless-table\" data-height=\"520\"></div>\n\n## What changes\n\n- **Stores → runes.** Reactive data is `$state` / a plain array, not a\n Svelte store you pass to `createTable`. If your rows came from a `derived`\n store, read it with `$` at the call site or move it to `$derived`.\n- **Plugins → features.** Register `rowSortingFeature` etc. in\n `tableFeatures({...})` instead of `addSortBy()` in the plugin bag.\n- **View model → component.** You stop owning the markup; style through\n `--sg-*` tokens (and Tailwind) instead of `tableAttrs`.\n\n## What does not come across\n\n- **Custom `cell` renderers.** `createRender(MyComponent, props)` has no direct\n equivalent; SvGrid uses a `cell` snippet. The codemod preserves yours as a\n `// TODO port:` comment.\n- **Your own plugins.** There is no plugin system to port them into.\n- **`tableAttrs` / `attrs()` spreading.** The component owns its attributes.\n\n## When not to migrate\n\n- You need control of the exact DOM. Use the fork, or SvGrid's\n [headless core](../why-headless.md) at `@svgrid/grid/core`.\n- Your table is a handful of rows with no interaction. A plain `{#each}` is\n less code than either library.\n- You depend on custom plugins. That is a rewrite, not a migration.\n- You only need Svelte 5 support. Use the fork.\n\n## Frequently asked questions\n\n### How hard is it to move from svelte-headless-table to SvGrid?\n\nUsually 1-3 hours per grid, and `npx @svgrid/migrate` does the mechanical part.\nThe plugin-to-feature mapping is almost one-to-one; most of the work is deleting\nthe `createViewModel` + `Subscribe` + `<table>` markup, because SvGrid renders\nthat for you.\n\n### Does SvGrid use Svelte 5 runes instead of stores?\n\nYes. svelte-headless-table is built on Svelte 4 stores; SvGrid is Svelte-5\nnative (`$state` / `$derived` / `$effect`) with snippets for custom cells.\n\n### Is SvGrid still headless like svelte-headless-table?\n\nYes - `createSvGrid` plus the row-model factories is a headless engine you can\ndrive with your own markup, importable on its own from `@svgrid/grid/core`. The\ndifference is that SvGrid *also* ships a batteries-included `<SvGrid>` component\nso you usually do not have to.\n\n### Is svelte-headless-table still maintained?\n\nThe original's last publish date and Svelte 4 peer range are in the facts box\nat the top of the page, read from npm on the date shown, and its README says\nthere are no current plans for a Svelte 5 port. A community fork,\n`@humanspeak/svelte-headless-table`, is maintained and runs on Svelte 5 with\nthe same API; the facts box has its release line too. The fork is the\nlowest-effort option if you are happy with the library and only need Svelte 5.\n\n## What you end up with\n\nSorting, filtering and selection, on a library that still ships for Svelte 5.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable selectable />\n```\n\n## See also\n\n- [Upgrading a Svelte 4 table to Svelte 5](./svelte-5-upgrade-data-tables.md) - the whole upgrade run and captured, step by step\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/) - the side-by-side comparison, with a source and date for every claim\n- [Headless: build a table](./headless/build-a-table.md) - the demo above, step by step\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - sibling headless guide\n- [Why headless?](../why-headless.md) - the design rationale\n- [Architecture](./architecture.md) - the engine + render-component split\n"
6183
6183
  },
6184
6184
  {
6185
6185
  "slug": "help/migrating-from-syncfusion",
6186
6186
  "path": "docs/help/migrating-from-syncfusion.md",
6187
6187
  "title": "Migrating from Syncfusion Grid",
6188
6188
  "section": "Help",
6189
- "markdown": "# Migrating from Syncfusion Grid\n\nSyncfusion Grid is a comprehensive commercial grid across JavaScript,\nReact, Vue, Angular, and Blazor, with a free community license for\nqualifying small teams. Moving to SvGrid trades the multi-framework suite\nfor a Svelte-5-native grid with an unconditional MIT core.\n\n> Estimated effort: **3-6 hours** per grid.\n>\n> Syncfusion's option names vary across its framework flavours; map your\n> version's columns and events onto the SvGrid equivalents below.\n\n<!-- facts:start syncfusion-grid -->\n> **Facts, checked 9 Nov 2026.** `@syncfusion/ej2-grids` 35.1.42, Commercial (see licence file), last published 27 Oct 2026, 1,060,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). Syncfusion Grid pricing, as its site states it: syncfusion.com lists the Grid SDK at $959 per developer per year and the UI Component Suite at $1,199 per developer per year, minimum five developers; a free Community License applies to companies and individuals with under $1M annual revenue, up to five developers and no more than ten employees (https://www.syncfusion.com/sales/products, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Syncfusion Grid](https://svgrid.com/compare/syncfusion-grid/).\n<!-- facts:end -->\n\n## Concept map\n\n| Syncfusion Grid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `columns: [{ field, headerText }]` | `columns: [{ field, header }]` |\n| `template` / cell template | `cell: (c) => renderSnippet(...)` |\n| `editSettings` + `editType` | `enableInlineEditing` + `editorType` |\n| `allowSorting` / `allowFiltering` | `rowSortingFeature` / `columnFilteringFeature` |\n| `allowGrouping` + `aggregates` | `columnGroupingFeature` + aggregates |\n| `detailTemplate` | `rowExpandingFeature` / master-detail rows |\n| `dataSource` (local / `DataManager`) | `data` array or `externalSort/Filter` |\n| `allowPaging` + `pageSettings` | `showPagination` |\n| `toolbar` export | `@svgrid/enterprise` export pack |\n\n## Shape of the change\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <ejs-grid [dataSource]=\"rows\" [allowSorting]=\"true\" [allowFiltering]=\"true\"\n- [allowPaging]=\"true\" [editSettings]=\"{ allowEditing: true }\">\n- <e-columns>\n- <e-column field=\"name\" headerText=\"Name\"></e-column>\n- <e-column field=\"amount\" headerText=\"Amount\" format=\"C2\"></e-column>\n- </e-columns>\n- </ejs-grid>\n\n+ <SvGrid\n+ data={rows}\n+ columns={[\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]}\n+ features={tableFeatures({ rowSortingFeature, columnFilteringFeature })}\n+ showPagination enableInlineEditing />\n```\n\n## Why teams switch\n\n- **Svelte 5 native** instead of a multi-framework commercial suite.\n- **Unconditional MIT core** - no revenue / headcount threshold like the\n Syncfusion community license.\n- **Optional Enterprise pack** for export / import / pivot / AI.\n\n## What to weigh\n\n- Syncfusion ships **one of the largest feature sets and suites**\n available, plus Blazor support. If you need that breadth or Blazor,\n weigh it against the Svelte-native fit.\n\n## Frequently asked questions\n\n### Is SvGrid free, unlike the Syncfusion commercial license?\n\n`@svgrid/grid` is MIT - free for everyone, with no revenue or team-size\nconditions. Syncfusion's community license is free only for qualifying small\nteams; otherwise it is commercial.\n\n### How do Syncfusion columns map to SvGrid?\n\n`field` stays `field`, `headerText` becomes `header`, and `template` becomes a\n`renderSnippet` cell. Sorting / filtering / grouping move to registered\nfeatures.\n\n### Does SvGrid support Blazor?\n\nNo - SvGrid is Svelte 5 only. If you need Blazor, that is a genuine reason to\nstay on Syncfusion. For a Svelte stack, SvGrid is the native fit.\n\n## What you end up with\n\nSort, filter, group and inline edit - the four things a Syncfusion grid is usually doing.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} groupable sortable filterable editable />\n```\n\n## See also\n\n- [SvGrid vs Syncfusion Grid](https://svgrid.com/compare/syncfusion-grid/) - the side-by-side comparison\n- [Data export and printing - Enterprise](./export.md) - the Enterprise export pack\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n"
6189
+ "markdown": "# Migrating from Syncfusion Grid\n\nSyncfusion Grid is a comprehensive commercial grid across JavaScript,\nReact, Vue, Angular, and Blazor, with a free community license for\nqualifying small teams. Moving to SvGrid trades the multi-framework suite\nfor a Svelte-5-native grid with an unconditional MIT core.\n\n> Estimated effort: **3-6 hours** per grid.\n>\n> Syncfusion's option names vary across its framework flavours; map your\n> version's columns and events onto the SvGrid equivalents below.\n\n<!-- facts:start syncfusion-grid -->\n> **Facts, checked 9 Nov 2026.** `@syncfusion/ej2-grids` 35.1.42, Commercial (see licence file), last published 27 Oct 2026, 1,060,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). Syncfusion Grid pricing, as its site states it: syncfusion.com lists the Grid SDK at $959 per developer per year and the UI Component Suite at $1,199 per developer per year, minimum five developers; a free Community License applies to companies and individuals with under $1M annual revenue, up to five developers and no more than ten employees (https://www.syncfusion.com/sales/products, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Syncfusion Grid](https://svgrid.com/compare/syncfusion-grid/).\n<!-- facts:end -->\n\n## Concept map\n\n| Syncfusion Grid | sv-grid |\n| ---------------------------------------- | ----------------------------------------- |\n| `columns: [{ field, headerText }]` | `columns: [{ field, header }]` |\n| `template` / cell template | `cell: (c) => renderSnippet(...)` |\n| `editSettings` + `editType` | `enableInlineEditing` + `editorType` |\n| `allowSorting` / `allowFiltering` | `rowSortingFeature` / `columnFilteringFeature` |\n| `allowGrouping` + `aggregates` | `columnGroupingFeature` + aggregates |\n| `detailTemplate` | `rowExpandingFeature` / master-detail rows |\n| `dataSource` (local / `DataManager`) | `data` array or `externalSort/Filter` |\n| `allowPaging` + `pageSettings` | `showPagination` |\n| `toolbar` export | `@svgrid/enterprise` export pack |\n\n## Shape of the change\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <ejs-grid [dataSource]=\"rows\" [allowSorting]=\"true\" [allowFiltering]=\"true\"\n- [allowPaging]=\"true\" [editSettings]=\"{ allowEditing: true }\">\n- <e-columns>\n- <e-column field=\"name\" headerText=\"Name\"></e-column>\n- <e-column field=\"amount\" headerText=\"Amount\" format=\"C2\"></e-column>\n- </e-columns>\n- </ejs-grid>\n\n+ <SvGrid\n+ data={rows}\n+ columns={[\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]}\n+ features={tableFeatures({ rowSortingFeature, columnFilteringFeature })}\n+ showPagination enableInlineEditing />\n```\n\n## Why teams switch\n\n- **Svelte 5 native** instead of a multi-framework commercial suite.\n- **Unconditional MIT core** - no revenue / headcount threshold like the\n Syncfusion community license.\n- **Optional Enterprise pack** for export / import / pivot / AI.\n\n## What to weigh\n\n- Syncfusion ships **one of the largest feature sets and suites**\n available, plus Blazor support. If you need that breadth or Blazor,\n weigh it against the Svelte-native fit.\n\n## Frequently asked questions\n\n### Is SvGrid free, unlike the Syncfusion commercial license?\n\n`@svgrid/grid` is MIT - free for everyone, with no revenue or team-size\nconditions. Syncfusion's community license is free only for qualifying small\nteams; otherwise it is commercial.\n\n### How do Syncfusion columns map to SvGrid?\n\n`field` stays `field`, `headerText` becomes `header`, and `template` becomes a\n`renderSnippet` cell. Sorting / filtering / grouping move to registered\nfeatures.\n\n### Does SvGrid support Blazor?\n\nNo - SvGrid is Svelte 5 only. If you need Blazor, that is a genuine reason to\nstay on Syncfusion. For a Svelte stack, SvGrid is the native fit.\n\n## What you end up with\n\nSort, filter, group and inline edit - the four things a Syncfusion grid is usually doing.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} groupBy={['department']} groupable sortable filterable editable />\n```\n\n## See also\n\n- [SvGrid vs Syncfusion Grid](https://svgrid.com/compare/syncfusion-grid/) - the side-by-side comparison\n- [Data export and printing - Enterprise](./export.md) - the Enterprise export pack\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n"
6190
6190
  },
6191
6191
  {
6192
6192
  "slug": "help/migrating-from-tabulator",
6193
6193
  "path": "docs/help/migrating-from-tabulator.md",
6194
6194
  "title": "Migrating from Tabulator",
6195
6195
  "section": "Help",
6196
- "markdown": "# Migrating from Tabulator\n\nTabulator is a feature-rich vanilla-JS table you instantiate against a\nDOM element and drive imperatively (`table.setData(...)`,\n`table.on('cellEdited', ...)`). SvGrid covers a similar feature surface\nbut as a Svelte 5 component: data is a reactive `$state` array, custom\ncells are snippets, and there is no manual mount / teardown.\n\n> Estimated effort: **2-4 hours** per grid - column translation plus\n> swapping imperative `setData` for reactive state.\n\n<!-- facts:start tabulator -->\n> **Facts, checked 9 Nov 2026.** `tabulator-tables` 6.7.1, MIT, last published 22 Oct 2026, 826,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `tabulator-tables` 6.7.1 109.5 KB JS + 3.8 KB CSS (measured 9 Nov 2026). Tabulator pricing, as its site states it: Tabulator is MIT and free; no licence or support plan is sold (https://github.com/olifolkerd/tabulator, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Tabulator](https://svgrid.com/compare/tabulator/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Tabulator | sv-grid |\n| -------------------------------------- | ----------------------------------------- |\n| `new Tabulator(el, { data, columns })` | `<SvGrid data={...} columns={...} />` |\n| `{ title: 'Name', field: 'name' }` | `{ header: 'Name', field: 'name' }` |\n| `formatter: fn` / `formatter: 'money'` | `cell: (c) => renderSnippet(...)` or `format` |\n| `editor: 'input' / 'number' / ...` | `editorType: 'text' / 'number' / ...` |\n| `headerFilter: true` | `columnFilteringFeature` |\n| `dataTree: true` | `rowExpandingFeature` + tree row shape |\n| `table.setData(rows)` | reassign the `$state` data array |\n| `table.on('cellEdited', ...)` | `onCellValueChange` |\n| `table.setSort(field, dir)` | `api.setSort(field, dir)` |\n| `layout: 'fitColumns'` | `fitColumns` prop |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import { TabulatorFull as Tabulator } from 'tabulator-tables'\n-\n- const table = new Tabulator('#grid', {\n- data: rows,\n- layout: 'fitColumns',\n- columns: [\n- { title: 'Name', field: 'name' },\n- { title: 'Amount', field: 'amount', formatter: 'money', editor: 'number' },\n- ],\n- })\n- table.on('cellEdited', (cell) => save(cell.getData()))\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' }, editorType: 'number' },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} fitColumns enableInlineEditing\n+ onCellValueChange={(e) => save(e)} />\n```\n\n## What changes\n\n- **Imperative → reactive.** No `setData()` - reassign the `$state`\n array and the grid updates.\n- **String formatters → snippets.** Custom cells are Svelte snippets /\n components, not HTML-string formatters.\n- **Mount / teardown disappears.** No `new Tabulator(el)` and no manual\n destroy on unmount.\n\n## Frequently asked questions\n\n### Is migrating from Tabulator to SvGrid hard?\n\nIt is a configuration translation plus one mindset shift: you stop calling\n`setData()` and instead let a reactive `$state` array drive the grid. Budget\n2-4 hours per grid.\n\n### Is SvGrid free like Tabulator?\n\nYes. `@svgrid/grid` is MIT-licensed like Tabulator. The optional\n`@svgrid/enterprise` pack (export, import, pivot, AI) is the only paid piece.\n\n### Can I keep Tabulator's custom formatters?\n\nYou re-create them as Svelte snippets via `renderSnippet`. Any Svelte\ncomponent or markup works, not just an HTML string.\n\n## What you end up with\n\nSort, filter and edit, with pagination underneath.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable editable pageable pageSize={3} />\n```\n\n## See also\n\n- [SvGrid vs Tabulator](https://svgrid.com/compare/tabulator/) - the side-by-side comparison\n- [Cell components](./cells/cell-components.md) - snippet-based custom cells\n- [Architecture](./architecture.md) - engine + render-component split\n"
6196
+ "markdown": "# Migrating from Tabulator\n\nTabulator is a feature-rich vanilla-JS table you instantiate against a\nDOM element and drive imperatively (`table.setData(...)`,\n`table.on('cellEdited', ...)`). SvGrid covers a similar feature surface\nbut as a Svelte 5 component: data is a reactive `$state` array, custom\ncells are snippets, and there is no manual mount / teardown.\n\n> Estimated effort: **2-4 hours** per grid - column translation plus\n> swapping imperative `setData` for reactive state.\n\n<!-- facts:start tabulator -->\n> **Facts, checked 9 Nov 2026.** `tabulator-tables` 6.7.1, MIT, last published 22 Oct 2026, 826,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `tabulator-tables` 6.7.1 109.5 KB JS + 3.8 KB CSS (measured 9 Nov 2026). Tabulator pricing, as its site states it: Tabulator is MIT and free; no licence or support plan is sold (https://github.com/olifolkerd/tabulator, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Tabulator](https://svgrid.com/compare/tabulator/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| Tabulator | sv-grid |\n| -------------------------------------- | ----------------------------------------- |\n| `new Tabulator(el, { data, columns })` | `<SvGrid data={...} columns={...} />` |\n| `{ title: 'Name', field: 'name' }` | `{ header: 'Name', field: 'name' }` |\n| `formatter: fn` / `formatter: 'money'` | `cell: (c) => renderSnippet(...)` or `format` |\n| `editor: 'input' / 'number' / ...` | `editorType: 'text' / 'number' / ...` |\n| `headerFilter: true` | `columnFilteringFeature` |\n| `dataTree: true` | `rowExpandingFeature` + tree row shape |\n| `table.setData(rows)` | reassign the `$state` data array |\n| `table.on('cellEdited', ...)` | `onCellValueChange` |\n| `table.setSort(field, dir)` | `api.setSort(field, dir)` |\n| `layout: 'fitColumns'` | `fitColumns` prop |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import { TabulatorFull as Tabulator } from 'tabulator-tables'\n-\n- const table = new Tabulator('#grid', {\n- data: rows,\n- layout: 'fitColumns',\n- columns: [\n- { title: 'Name', field: 'name' },\n- { title: 'Amount', field: 'amount', formatter: 'money', editor: 'number' },\n- ],\n- })\n- table.on('cellEdited', (cell) => save(cell.getData()))\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' }, editorType: 'number' },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} fitColumns enableInlineEditing\n+ onCellValueChange={(e) => save(e)} />\n```\n\n## What changes\n\n- **Imperative → reactive.** No `setData()` - reassign the `$state`\n array and the grid updates.\n- **String formatters → snippets.** Custom cells are Svelte snippets /\n components, not HTML-string formatters.\n- **Mount / teardown disappears.** No `new Tabulator(el)` and no manual\n destroy on unmount.\n\n## Frequently asked questions\n\n### Is migrating from Tabulator to SvGrid hard?\n\nIt is a configuration translation plus one mindset shift: you stop calling\n`setData()` and instead let a reactive `$state` array drive the grid. Budget\n2-4 hours per grid.\n\n### Is SvGrid free like Tabulator?\n\nYes. `@svgrid/grid` is MIT-licensed like Tabulator. The optional\n`@svgrid/enterprise` pack (export, import, pivot, AI) is the only paid piece.\n\n### Can I keep Tabulator's custom formatters?\n\nYou re-create them as Svelte snippets via `renderSnippet`. Any Svelte\ncomponent or markup works, not just an HTML string.\n\n## What you end up with\n\nSort, filter and edit, with pagination underneath.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable editable pageable pageSize={3} />\n```\n\n## See also\n\n- [SvGrid vs Tabulator](https://svgrid.com/compare/tabulator/) - the side-by-side comparison\n- [Cell components](./cells/cell-components.md) - snippet-based custom cells\n- [Architecture](./architecture.md) - engine + render-component split\n"
6197
6197
  },
6198
6198
  {
6199
6199
  "slug": "help/migrating-from-tanstack-table",
6200
6200
  "path": "docs/help/migrating-from-tanstack-table.md",
6201
6201
  "title": "Migrating from TanStack Table",
6202
6202
  "section": "Help",
6203
- "markdown": "# Migrating from TanStack Table\n\nTanStack Table (formerly React Table) is the closest conceptual\nsibling to sv-grid - both are \"headless data grid\" libraries with\nexplicit row-model pipelines, and the v9 Svelte adapter,\n`@tanstack/svelte-table`, targets Svelte 5 with the same runes-first\nshape. If you already think in `createTable()` + a features object,\nthis is a half-day port; the vocabulary below covers the v8 names too.\n\n> Estimated effort: **30 min** per grid for read-only views, **2-4\n> hours** for editing-heavy grids.\n\n<!-- facts:start tanstack-table -->\n> **Facts, checked 9 Nov 2026.** `@tanstack/svelte-table` 9.3.1, MIT, last published 12 Oct 2026, 539,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `@tanstack/svelte-table` 9.3.1 36.5 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). TanStack Table pricing, as its site states it: TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table (https://tanstack.com/table/latest, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/).\n<!-- facts:end -->\n\n> **On v9?** SvGrid exports the same feature vocabulary, so most of this\n> page is automated. Run `npx @svgrid/migrate src` to preview the port, and\n> see [Migrating from the shadcn-svelte data table](./migrating-from-shadcn-data-table.md)\n> if that is where your table came from.\n\n## Vocabulary cheat sheet\n\n| TanStack Table | sv-grid |\n| ------------------------------- | ---------------------------------------- |\n| `useReactTable({...})` | `createSvGrid({...})` or `<SvGrid>` |\n| `getCoreRowModel()` | `createCoreRowModel()` |\n| `getSortedRowModel()` | Auto-registered by `rowSortingFeature` |\n| `getFilteredRowModel()` | Auto-registered by `columnFilteringFeature` |\n| `getPaginationRowModel()` | Built in; toggle with `showPagination` |\n| `getGroupedRowModel()` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `getExpandedRowModel()` | `rowExpandingFeature` |\n| `flexRender(col.cell, ctx)` | `renderSnippet(MyCell, props)` OR string field |\n| `columnHelper.accessor('field')` | `{ field: 'field' }` directly |\n| `table.getRowModel().rows` | `api.getDisplayedRows()` |\n| `table.setColumnFilters(...)` | `api.setFilter(id, {...})` per column |\n| `meta` (column / table) | n/a - use column `cellClass` / `cell` instead |\n\n## Imports\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import {\n- useReactTable, getCoreRowModel, getSortedRowModel, getFilteredRowModel,\n- flexRender, createColumnHelper,\n- } from '@tanstack/react-table'\n\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ renderSnippet,\n+ } from '@svgrid/grid'\n```\n\n## Column definitions\n\n```diff\n- const columnHelper = createColumnHelper<Order>()\n- const columns = [\n- columnHelper.accessor('orderId', { header: 'Order ID' }),\n- columnHelper.accessor('amount', { header: 'Amount', cell: ({ getValue }) => fmt(getValue<number>()) }),\n- ]\n\n+ const columns: ColumnDef<typeof features, Order>[] = [\n+ { field: 'orderId', header: 'Order ID' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n```\n\nThe `fieldFn` form is the same:\n\n```diff\n- columnHelper.accessor((row) => row.profile.name, { id: 'name', header: 'Name' })\n\n+ { id: 'name', fieldFn: (row) => row.profile.name, header: 'Name' }\n```\n\n## Mounting\n\n```diff\n- const table = useReactTable({\n- data, columns,\n- getCoreRowModel: getCoreRowModel(),\n- getSortedRowModel: getSortedRowModel(),\n- getFilteredRowModel: getFilteredRowModel(),\n- })\n- return (\n- <table>\n- <thead>{table.getHeaderGroups().map(/* ... */)}</thead>\n- <tbody>{table.getRowModel().rows.map(/* ... */)}</tbody>\n- </table>\n- )\n\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+\n+ <SvGrid data={rows} columns={columns} features={features} />\n```\n\nSv-grid bundles the render + virtualization. No `<table>` you have\nto author yourself.\n\n## Imperative state\n\n| TanStack Table | sv-grid |\n| ------------------------------------------- | ---------------------------------------- |\n| `table.setColumnFilters([{id, value}])` | `api.setFilter(id, { operator: 'contains', value })` |\n| `table.getColumn('id')?.setFilterValue(v)` | `api.setFilter(id, { operator: 'contains', value: v })` |\n| `table.setSorting([{id, desc}])` | `api.setSort(id, desc ? 'desc' : 'asc')` |\n| `table.setRowSelection(state)` | n/a - use `api.clearRowSelection()` + UI events |\n| `table.toggleAllRowsExpanded()` | `api.expandAllGroups()` |\n| `table.getColumn('x').toggleVisibility()` | `api.setColumnVisible('x', false)` |\n| `table.options.meta?.foo` | Pass through component props / `$state` |\n\n## Custom cells\n\n```diff\n- {\n- header: 'Status',\n- cell: ({ row }) => <Badge tone={row.original.status} />,\n- }\n\n+ {\n+ field: 'status', header: 'Status',\n+ cell: (ctx) => renderSnippet(StatusCell, { tone: ctx.row.original.status }),\n+ }\n```\n\n`renderSnippet` accepts any Svelte 5 snippet; `renderComponent`\naccepts a regular component if you prefer SFCs.\n\n## Server-side data\n\n```diff\n- const [pagination, setPagination] = useState({pageIndex: 0, pageSize: 25})\n- const [sorting, setSorting] = useState([])\n- const table = useReactTable({\n- data, columns, manualPagination: true, manualSorting: true,\n- state: { pagination, sorting },\n- onPaginationChange: setPagination,\n- onSortingChange: setSorting,\n- })\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ externalSort={true}\n+ externalFilter={true}\n+ onSortingChange={(clauses) => refetch({ sort: clauses })}\n+ onFiltersChange={(f) => refetch({ filters: f.columns })}\n+ />\n```\n\nOr hand the fetching to a controller and pass it as one prop:\n`createServerDataSource(source, { mode: 'infinite' })` owns the blocks,\nthe sort and filter round trips and the placeholder rows, and the\nmanual-flag wiring above disappears. TanStack Table has no server row\nmodel; per-group fetching is yours to write, where the Enterprise\n`createServerRowModel` does it over the same contract.\n\n<div data-docs-demo=\"148-server-row-model\" data-height=\"480\"></div>\n\n## What you get for free vs TanStack Table\n\n- **Renderer**. TanStack is headless-only; you draw the `<table>`.\n Sv-grid ships an opinionated, fast renderer with virtualization,\n scroll, sticky headers, column resize, keyboard nav, ARIA.\n- **Inline editing**. `editorType: 'text' | 'number' | 'date' |\n 'list' | ...` - built-in editors with parsing + validation hooks.\n- **CSP-clean**. No `eval`. The shipped tests confirm.\n- **In-grid AI helpers**. Natural-language filter, smart fill, summarise\n and classify, free in `@svgrid/grid` against a model you register.\n- **Enterprise features**. Excel and PDF export, print, import and pivot\n in one paid add-on; CSV, TSV and JSON export are free.\n\nThe headless demo below is the shape a TanStack app has today - a plain\n`<table>` driven by the engine - running on `createSvGrid`:\n\n<div data-docs-demo=\"186-headless-table\" data-height=\"520\"></div>\n\n## What you give up\n\n- **One engine across frameworks.** TanStack has adapters for React,\n Vue, Solid, Qwik, Lit and Svelte; sv-grid is Svelte 5-native and has\n no React wrapper.\n- **The smallest engine-only bundle.** The facts box at the top of the\n page has both packages measured the same way; the difference is the\n renderer, virtualizer, filter UI and editors you would otherwise write.\n- **TanStack Virtual and the shadcn-svelte data table recipe.** Sv-grid\n virtualizes on its own and has its own\n [shadcn-svelte guide](./migrating-from-shadcn-data-table.md).\n- **`meta` for arbitrary side-channel state.** Pass props directly.\n- **`@tanstack/match-sorter-utils` integration**. The grid's filter\n has a `contains` operator; if you need fuzzy match, plug your\n matcher in via the column's `cell` snippet or use\n `externalFilter={true}` and run the matcher yourself.\n\n## Frequently asked questions\n\n### How long does it take to migrate from TanStack Table to SvGrid?\n\nAbout 30 minutes per read-only grid and 2-4 hours for editing-heavy grids.\nBoth libraries use explicit row-model pipelines, so `createTable()` +\n`getCoreRowModel()` maps almost directly onto `createSvGrid()` +\n`createCoreRowModel()`.\n\n### Why switch from TanStack Table to SvGrid on Svelte 5?\n\nTanStack Table is headless-only and multi-framework, so you build the entire\nDOM, virtualization, filter UI, and editing layer yourself. SvGrid keeps the\nheadless core but ships a batteries-included `<SvGrid>` render component -\nvirtualization, Excel-style filters, cell selection, and inline editing work\nin one prop pass.\n\n### Is SvGrid free like TanStack Table?\n\nYes - `@svgrid/grid` is MIT-licensed, like TanStack Table. SvGrid\nadditionally offers an optional paid `@svgrid/enterprise` pack for Excel and\nPDF export, print, pivot and import, which TanStack Table does not provide at\nall; TanStack sells Enterprise Support as consulting instead, as the facts box\nabove notes.\n\n### Does the TanStack Svelte adapter support Svelte 5?\n\nYes. The v9 line of `@tanstack/svelte-table` declares a Svelte 5 peer range\nand reads rune data through getters; the version in the facts box above is\nthe one read from npm. If you are on the v8 adapter with Svelte 4 stores,\nthe port to sv-grid is the same work as the port to v9, minus the markup.\n\n## What you end up with\n\nThe same model you had, with the rendering already written.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} selectable />\n```\n\n## See also\n\n- [SvGrid vs TanStack Table](https://svgrid.com/compare/tanstack-table/) -\n the side-by-side comparison, with a source and date for every claim\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](./comparison.md) -\n the measured benchmark, TanStack column included\n- [Architecture](./architecture.md) - the headless engine + render\n component split\n- [Migrating from AG Grid](./migrating-from-ag-grid.md) - sibling guide\n- [Why headless?](../why-headless.md) - the design rationale\n"
6203
+ "markdown": "# Migrating from TanStack Table\n\nTanStack Table (formerly React Table) is the closest conceptual\nsibling to sv-grid - both are \"headless data grid\" libraries with\nexplicit row-model pipelines, and the v9 Svelte adapter,\n`@tanstack/svelte-table`, targets Svelte 5 with the same runes-first\nshape. If you already think in `createTable()` + a features object,\nthis is a half-day port; the vocabulary below covers the v8 names too.\n\n> Estimated effort: **30 min** per grid for read-only views, **2-4\n> hours** for editing-heavy grids.\n\n<!-- facts:start tanstack-table -->\n> **Facts, checked 9 Nov 2026.** `@tanstack/svelte-table` 9.3.1, MIT, last published 12 Oct 2026, 539,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `@tanstack/svelte-table` 9.3.1 36.5 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). TanStack Table pricing, as its site states it: TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table (https://tanstack.com/table/latest, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/).\n<!-- facts:end -->\n\n> **On v9?** SvGrid exports the same feature vocabulary, so most of this\n> page is automated. Run `npx @svgrid/migrate src` to preview the port, and\n> see [Migrating from the shadcn-svelte data table](./migrating-from-shadcn-data-table.md)\n> if that is where your table came from.\n\n## Vocabulary cheat sheet\n\n| TanStack Table | sv-grid |\n| ------------------------------- | ---------------------------------------- |\n| `useReactTable({...})` | `createSvGrid({...})` or `<SvGrid>` |\n| `getCoreRowModel()` | `createCoreRowModel()` |\n| `getSortedRowModel()` | Auto-registered by `rowSortingFeature` |\n| `getFilteredRowModel()` | Auto-registered by `columnFilteringFeature` |\n| `getPaginationRowModel()` | Built in; toggle with `showPagination` |\n| `getGroupedRowModel()` | `columnGroupingFeature` + `api.setGroupBy()` |\n| `getExpandedRowModel()` | `rowExpandingFeature` |\n| `flexRender(col.cell, ctx)` | `renderSnippet(MyCell, props)` OR string field |\n| `columnHelper.accessor('field')` | `{ field: 'field' }` directly |\n| `table.getRowModel().rows` | `api.getDisplayedRows()` |\n| `table.setColumnFilters(...)` | `api.setFilter(id, {...})` per column |\n| `meta` (column / table) | n/a - use column `cellClass` / `cell` instead |\n\n## Imports\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- import {\n- useReactTable, getCoreRowModel, getSortedRowModel, getFilteredRowModel,\n- flexRender, createColumnHelper,\n- } from '@tanstack/react-table'\n\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ renderSnippet,\n+ } from '@svgrid/grid'\n```\n\n## Column definitions\n\n```diff\n- const columnHelper = createColumnHelper<Order>()\n- const columns = [\n- columnHelper.accessor('orderId', { header: 'Order ID' }),\n- columnHelper.accessor('amount', { header: 'Amount', cell: ({ getValue }) => fmt(getValue<number>()) }),\n- ]\n\n+ const columns: ColumnDef<typeof features, Order>[] = [\n+ { field: 'orderId', header: 'Order ID' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n```\n\nThe `fieldFn` form is the same:\n\n```diff\n- columnHelper.accessor((row) => row.profile.name, { id: 'name', header: 'Name' })\n\n+ { id: 'name', fieldFn: (row) => row.profile.name, header: 'Name' }\n```\n\n## Mounting\n\n```diff\n- const table = useReactTable({\n- data, columns,\n- getCoreRowModel: getCoreRowModel(),\n- getSortedRowModel: getSortedRowModel(),\n- getFilteredRowModel: getFilteredRowModel(),\n- })\n- return (\n- <table>\n- <thead>{table.getHeaderGroups().map(/* ... */)}</thead>\n- <tbody>{table.getRowModel().rows.map(/* ... */)}</tbody>\n- </table>\n- )\n\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+\n+ <SvGrid data={rows} columns={columns} features={features} />\n```\n\nSv-grid bundles the render + virtualization. No `<table>` you have\nto author yourself.\n\n## Imperative state\n\n| TanStack Table | sv-grid |\n| ------------------------------------------- | ---------------------------------------- |\n| `table.setColumnFilters([{id, value}])` | `api.setFilter(id, { operator: 'contains', value })` |\n| `table.getColumn('id')?.setFilterValue(v)` | `api.setFilter(id, { operator: 'contains', value: v })` |\n| `table.setSorting([{id, desc}])` | `api.setSort(id, desc ? 'desc' : 'asc')` |\n| `table.setRowSelection(state)` | n/a - use `api.clearRowSelection()` + UI events |\n| `table.toggleAllRowsExpanded()` | `api.expandAllGroups()` |\n| `table.getColumn('x').toggleVisibility()` | `api.setColumnVisible('x', false)` |\n| `table.options.meta?.foo` | Pass through component props / `$state` |\n\n## Custom cells\n\n```diff\n- {\n- header: 'Status',\n- cell: ({ row }) => <Badge tone={row.original.status} />,\n- }\n\n+ {\n+ field: 'status', header: 'Status',\n+ cell: (ctx) => renderSnippet(StatusCell, { tone: ctx.row.original.status }),\n+ }\n```\n\n`renderSnippet` accepts any Svelte 5 snippet; `renderComponent`\naccepts a regular component if you prefer SFCs.\n\n## Server-side data\n\n```diff\n- const [pagination, setPagination] = useState({pageIndex: 0, pageSize: 25})\n- const [sorting, setSorting] = useState([])\n- const table = useReactTable({\n- data, columns, manualPagination: true, manualSorting: true,\n- state: { pagination, sorting },\n- onPaginationChange: setPagination,\n- onSortingChange: setSorting,\n- })\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ externalSort={true}\n+ externalFilter={true}\n+ onSortingChange={(clauses) => refetch({ sort: clauses })}\n+ onFiltersChange={(f) => refetch({ filters: f.columns })}\n+ />\n```\n\nOr hand the fetching to a controller and pass it as one prop:\n`createServerDataSource(source, { mode: 'infinite' })` owns the blocks,\nthe sort and filter round trips and the placeholder rows, and the\nmanual-flag wiring above disappears. TanStack Table has no server row\nmodel; per-group fetching is yours to write, where the Enterprise\n`createServerRowModel` does it over the same contract.\n\n<div data-docs-demo=\"148-server-row-model\" data-height=\"480\"></div>\n\n## What you get for free vs TanStack Table\n\n- **Renderer**. TanStack is headless-only; you draw the `<table>`.\n Sv-grid ships an opinionated, fast renderer with virtualization,\n scroll, sticky headers, column resize, keyboard nav, ARIA.\n- **Inline editing**. `editorType: 'text' | 'number' | 'date' |\n 'list' | ...` - built-in editors with parsing + validation hooks.\n- **CSP-clean**. No `eval`. The shipped tests confirm.\n- **In-grid AI helpers**. Natural-language filter, smart fill, summarise\n and classify, free in `@svgrid/grid` against a model you register.\n- **Enterprise features**. Excel and PDF export, print, import and pivot\n in one paid add-on; CSV, TSV and JSON export are free.\n\nThe headless demo below is the shape a TanStack app has today - a plain\n`<table>` driven by the engine - running on `createSvGrid`:\n\n<div data-docs-demo=\"186-headless-table\" data-height=\"520\"></div>\n\n## What you give up\n\n- **One engine across frameworks.** TanStack has adapters for React,\n Vue, Solid, Qwik, Lit and Svelte; sv-grid is Svelte 5-native and has\n no React wrapper.\n- **The smallest engine-only bundle.** The facts box at the top of the\n page has both packages measured the same way; the difference is the\n renderer, virtualizer, filter UI and editors you would otherwise write.\n- **TanStack Virtual and the shadcn-svelte data table recipe.** Sv-grid\n virtualizes on its own and has its own\n [shadcn-svelte guide](./migrating-from-shadcn-data-table.md).\n- **`meta` for arbitrary side-channel state.** Pass props directly.\n- **`@tanstack/match-sorter-utils` integration**. The grid's filter\n has a `contains` operator; if you need fuzzy match, plug your\n matcher in via the column's `cell` snippet or use\n `externalFilter={true}` and run the matcher yourself.\n\n## Frequently asked questions\n\n### How long does it take to migrate from TanStack Table to SvGrid?\n\nAbout 30 minutes per read-only grid and 2-4 hours for editing-heavy grids.\nBoth libraries use explicit row-model pipelines, so `createTable()` +\n`getCoreRowModel()` maps almost directly onto `createSvGrid()` +\n`createCoreRowModel()`.\n\n### Why switch from TanStack Table to SvGrid on Svelte 5?\n\nTanStack Table is headless-only and multi-framework, so you build the entire\nDOM, virtualization, filter UI, and editing layer yourself. SvGrid keeps the\nheadless core but ships a batteries-included `<SvGrid>` render component -\nvirtualization, Excel-style filters, cell selection, and inline editing work\nin one prop pass.\n\n### Is SvGrid free like TanStack Table?\n\nYes - `@svgrid/grid` is MIT-licensed, like TanStack Table. SvGrid\nadditionally offers an optional paid `@svgrid/enterprise` pack for Excel and\nPDF export, print, pivot and import, which TanStack Table does not provide at\nall; TanStack sells Enterprise Support as consulting instead, as the facts box\nabove notes.\n\n### Does the TanStack Svelte adapter support Svelte 5?\n\nYes. The v9 line of `@tanstack/svelte-table` declares a Svelte 5 peer range\nand reads rune data through getters; the version in the facts box above is\nthe one read from npm. If you are on the v8 adapter with Svelte 4 stores,\nthe port to sv-grid is the same work as the port to v9, minus the markup.\n\n## What you end up with\n\nThe same model you had, with the rendering already written.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} selectable />\n```\n\n## See also\n\n- [SvGrid vs TanStack Table](https://svgrid.com/compare/tanstack-table/) -\n the side-by-side comparison, with a source and date for every claim\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](./comparison.md) -\n the measured benchmark, TanStack column included\n- [Architecture](./architecture.md) - the headless engine + render\n component split\n- [Migrating from AG Grid](./migrating-from-ag-grid.md) - sibling guide\n- [Why headless?](../why-headless.md) - the design rationale\n"
6204
6204
  },
6205
6205
  {
6206
6206
  "slug": "help/migrating-from-ui-kit-tables",
6207
6207
  "path": "docs/help/migrating-from-ui-kit-tables.md",
6208
6208
  "title": "Migrating from a UI-kit table (Flowbite / Skeleton / shadcn-svelte)",
6209
6209
  "section": "Help",
6210
- "markdown": "\n# Migrating from a UI-kit table (Flowbite / Skeleton / shadcn-svelte)\n\nFlowbite-Svelte, Skeleton, and shadcn-svelte all ship a styled `<Table>`\ncomponent. They are presentation components: they render the rows you\npass and match your design tokens, but sorting, filtering,\nvirtualization, and editing are yours to write by hand. Moving to SvGrid\nreplaces that hand-rolled logic with built-in features - while still\nletting you match the design system.\n\n> **Not the shadcn-svelte data table.** That one is a separate recipe\n> built on TanStack Table v9, and it already does sorting, filtering,\n> pagination, and row selection. If that is what you have, read\n> [Migrating from the shadcn-svelte data table](./migrating-from-shadcn-data-table.md)\n> instead - the port is different, and smaller.\n\n<!-- facts:start svelte-ui-kit-tables -->\n> **Facts, checked 9 Nov 2026.** `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the 30 days to 8 Nov 2026. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026). Flowbite / Skeleton / shadcn-svelte tables pricing, as its site states it: Flowbite Svelte, Skeleton and shadcn-svelte are MIT component libraries; their table components are free and no grid licence is sold (https://github.com/themesberg/flowbite-svelte, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/).\n<!-- facts:end -->\n\n> Estimated effort: **30 min - 1 hour** per table. You are mostly\n> deleting `{#each}` markup and the sort / filter state you wrote\n> yourself.\n\n## When this guide applies\n\nYou have a table that started simple and has grown:\n\n- You added a click-to-sort handler and a sort-state variable.\n- You added a search box and an array `.filter(...)`.\n- The list got long and now you want virtualization or pagination.\n- You started editing cells inline and it got messy.\n\nThat is the moment a styled `<Table>` stops paying for itself.\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <script>\n- import { Table, TableHead, TableHeadCell, TableBody, TableBodyRow, TableBodyCell } from 'flowbite-svelte'\n- let sortKey = 'name', asc = true\n- $: sorted = [...rows].sort((a, b) => (a[sortKey] > b[sortKey] ? 1 : -1) * (asc ? 1 : -1))\n- </script>\n-\n- <Table>\n- <TableHead>\n- <TableHeadCell on:click={() => (sortKey = 'name')}>Name</TableHeadCell>\n- <TableHeadCell on:click={() => (sortKey = 'amount')}>Amount</TableHeadCell>\n- </TableHead>\n- <TableBody>\n- {#each sorted as row}\n- <TableBodyRow>\n- <TableBodyCell>{row.name}</TableBodyCell>\n- <TableBodyCell>{row.amount}</TableBodyCell>\n- </TableBodyRow>\n- {/each}\n- </TableBody>\n- </Table>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\nThe sort handler, the `sorted` derivation, and the `{#each}` all go\naway. Sorting is a click on the header; filtering is the column menu.\n\n## Keep your design system\n\nSvGrid is unstyled-by-token, so it can match Flowbite / Skeleton /\nshadcn:\n\n- Re-theme through the `--sg-*` CSS custom properties (borders, header\n background, zebra rows, hover, selection, density).\n- Or drive those tokens from your Tailwind theme - see the\n [Tailwind integration](./tailwind.md) guide.\n\n## What you get for free\n\n- **Sorting, filtering, and pagination** without writing handlers.\n- **Row + column virtualization** so thousands of rows stay smooth.\n- **Inline editing**, **cell-range selection + copy**, grouping, tree,\n and master/detail when you need them.\n- **Accessibility** - WAI-ARIA grid roles and keyboard navigation built\n in, instead of a plain `<table>`.\n\n## What you give up\n\n- The **tiny footprint** of a styled `<table>` for a truly static list.\n- A **pixel-perfect**, zero-config match with the rest of the kit - you\n re-create the look with tokens instead (a one-time setup).\n\n## Frequently asked questions\n\n### When should I replace a Flowbite / Skeleton / shadcn table with a data grid?\n\nWhen the table needs sorting, filtering, virtualization, inline editing, or\nlarge datasets. These styled `<Table>` components are presentation only;\nonce you are hand-writing table behaviour, a data grid like SvGrid removes\nthat code. (shadcn-svelte also ships a separate TanStack-based data table -\nsee [that migration guide](./migrating-from-shadcn-data-table.md).)\n\n### Can SvGrid match my existing UI-kit styling?\n\nYes. SvGrid themes through `--sg-*` CSS variables, which you can wire to your\nTailwind / design-system tokens so it matches Flowbite, Skeleton, or\nshadcn-svelte.\n\n### Is it overkill for a small static table?\n\nFor a small, static table that never sorts or filters, a styled `<table>` is\nlighter and fine. Switch when the behaviour - not just the styling - becomes the\nwork.\n\n## What you end up with\n\nEverything the presentational table could not do: sort, filter, page, edit.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} editable />\n```\n\n## See also\n\n- [SvGrid vs UI-kit tables](https://svgrid.com/compare/svelte-ui-kit-tables/) - the side-by-side comparison\n- [Tailwind integration](./tailwind.md) - match your design system\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n"
6210
+ "markdown": "\n# Migrating from a UI-kit table (Flowbite / Skeleton / shadcn-svelte)\n\nFlowbite-Svelte, Skeleton, and shadcn-svelte all ship a styled `<Table>`\ncomponent. They are presentation components: they render the rows you\npass and match your design tokens, but sorting, filtering,\nvirtualization, and editing are yours to write by hand. Moving to SvGrid\nreplaces that hand-rolled logic with built-in features - while still\nletting you match the design system.\n\n> **Not the shadcn-svelte data table.** That one is a separate recipe\n> built on TanStack Table v9, and it already does sorting, filtering,\n> pagination, and row selection. If that is what you have, read\n> [Migrating from the shadcn-svelte data table](./migrating-from-shadcn-data-table.md)\n> instead - the port is different, and smaller.\n\n<!-- facts:start svelte-ui-kit-tables -->\n> **Facts, checked 9 Nov 2026.** `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the 30 days to 8 Nov 2026. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026). Flowbite / Skeleton / shadcn-svelte tables pricing, as its site states it: Flowbite Svelte, Skeleton and shadcn-svelte are MIT component libraries; their table components are free and no grid licence is sold (https://github.com/themesberg/flowbite-svelte, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/).\n<!-- facts:end -->\n\n> Estimated effort: **30 min - 1 hour** per table. You are mostly\n> deleting `{#each}` markup and the sort / filter state you wrote\n> yourself.\n\n## When this guide applies\n\nYou have a table that started simple and has grown:\n\n- You added a click-to-sort handler and a sort-state variable.\n- You added a search box and an array `.filter(...)`.\n- The list got long and now you want virtualization or pagination.\n- You started editing cells inline and it got messy.\n\nThat is the moment a styled `<Table>` stops paying for itself.\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <script>\n- import { Table, TableHead, TableHeadCell, TableBody, TableBodyRow, TableBodyCell } from 'flowbite-svelte'\n- let sortKey = 'name', asc = true\n- $: sorted = [...rows].sort((a, b) => (a[sortKey] > b[sortKey] ? 1 : -1) * (asc ? 1 : -1))\n- </script>\n-\n- <Table>\n- <TableHead>\n- <TableHeadCell on:click={() => (sortKey = 'name')}>Name</TableHeadCell>\n- <TableHeadCell on:click={() => (sortKey = 'amount')}>Amount</TableHeadCell>\n- </TableHead>\n- <TableBody>\n- {#each sorted as row}\n- <TableBodyRow>\n- <TableBodyCell>{row.name}</TableBodyCell>\n- <TableBodyCell>{row.amount}</TableBodyCell>\n- </TableBodyRow>\n- {/each}\n- </TableBody>\n- </Table>\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\nThe sort handler, the `sorted` derivation, and the `{#each}` all go\naway. Sorting is a click on the header; filtering is the column menu.\n\n## Keep your design system\n\nSvGrid is unstyled-by-token, so it can match Flowbite / Skeleton /\nshadcn:\n\n- Re-theme through the `--sg-*` CSS custom properties (borders, header\n background, zebra rows, hover, selection, density).\n- Or drive those tokens from your Tailwind theme - see the\n [Tailwind integration](./tailwind.md) guide.\n\n## What you get for free\n\n- **Sorting, filtering, and pagination** without writing handlers.\n- **Row + column virtualization** so thousands of rows stay smooth.\n- **Inline editing**, **cell-range selection + copy**, grouping, tree,\n and master/detail when you need them.\n- **Accessibility** - WAI-ARIA grid roles and keyboard navigation built\n in, instead of a plain `<table>`.\n\n## What you give up\n\n- The **tiny footprint** of a styled `<table>` for a truly static list.\n- A **pixel-perfect**, zero-config match with the rest of the kit - you\n re-create the look with tokens instead (a one-time setup).\n\n## Frequently asked questions\n\n### When should I replace a Flowbite / Skeleton / shadcn table with a data grid?\n\nWhen the table needs sorting, filtering, virtualization, inline editing, or\nlarge datasets. These styled `<Table>` components are presentation only;\nonce you are hand-writing table behaviour, a data grid like SvGrid removes\nthat code. (shadcn-svelte also ships a separate TanStack-based data table -\nsee [that migration guide](./migrating-from-shadcn-data-table.md).)\n\n### Can SvGrid match my existing UI-kit styling?\n\nYes. SvGrid themes through `--sg-*` CSS variables, which you can wire to your\nTailwind / design-system tokens so it matches Flowbite, Skeleton, or\nshadcn-svelte.\n\n### Is it overkill for a small static table?\n\nFor a small, static table that never sorts or filters, a styled `<table>` is\nlighter and fine. Switch when the behaviour - not just the styling - becomes the\nwork.\n\n## What you end up with\n\nEverything the presentational table could not do: sort, filter, page, edit.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} editable />\n```\n\n## See also\n\n- [SvGrid vs UI-kit tables](https://svgrid.com/compare/svelte-ui-kit-tables/) - the side-by-side comparison\n- [Tailwind integration](./tailwind.md) - match your design system\n- [Getting started](../getting-started.md) - a working grid in ~15 lines\n"
6211
6211
  },
6212
6212
  {
6213
6213
  "slug": "help/migrating-from-vincjo-datatables",
6214
6214
  "path": "docs/help/migrating-from-vincjo-datatables.md",
6215
6215
  "title": "Migrating from @vincjo/datatables",
6216
6216
  "section": "Help",
6217
- "markdown": "# Migrating from @vincjo/datatables\n\n`@vincjo/datatables` is a small, ergonomic Svelte datatable helper: you\nwrap your data in a handler and it gives you reactive sorting,\nfiltering, and pagination while you keep your own table markup. SvGrid\ncovers the same job and then keeps going - virtualization, an\nExcel-style filter menu, inline editing, and a render component - so the\nport is mostly about deciding how much markup you want to keep.\n\n> Estimated effort: **30 min - 2 hours** per table. A simple list is a\n> 30-minute swap; a feature-heavy one trends toward two hours.\n\n<!-- facts:start vincjo-datatables -->\n> **Facts, checked 9 Nov 2026.** `@vincjo/datatables` 2.8.1, MIT, last published 4 Jun 2026, 107,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `@vincjo/datatables` 2.8.1 10.2 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). @vincjo/datatables pricing, as its site states it: @vincjo/datatables is MIT on npm and free; no licence is sold (https://github.com/vincjo/datatables, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| @vincjo/datatables | sv-grid |\n| ------------------------------------------ | ----------------------------------------- |\n| `new TableHandler(data, { rowsPerPage })` | `createSvGrid({...})` or `<SvGrid>` |\n| `<ThSort {table} field=\"name\">` | `rowSortingFeature` (header sort built in) |\n| `<ThFilter {table} field=\"name\">` | `columnFilteringFeature` |\n| `table.global.set(query)` (search) | `api.setState({ globalFilter: query })` |\n| `<Pagination {table} />` / `<RowCount>` | Built in; toggle `showPagination` |\n| `table.rows` (current page rows) | `api.getDisplayedRows()` |\n| Server mode (`table.load(...)`) | `externalSort` / `externalFilter` + refetch |\n| Your own `<table>` + `{#each table.rows}` | `<SvGrid>` render component |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <script>\n- import { TableHandler, ThSort, ThFilter, Pagination } from '@vincjo/datatables'\n- const table = new TableHandler(rows, { rowsPerPage: 25 })\n- </script>\n-\n- <table>\n- <thead><tr>\n- <ThSort {table} field=\"name\">Name</ThSort>\n- <ThSort {table} field=\"amount\">Amount</ThSort>\n- </tr></thead>\n- <tbody>\n- {#each table.rows as row}\n- <tr><td>{row.name}</td><td>{row.amount}</td></tr>\n- {/each}\n- </tbody>\n- </table>\n- <Pagination {table} />\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## Server-side data\n\n```diff\n- // @vincjo/datatables: drive a server with the handler's load hook\n- const table = new TableHandler(rows, { rowsPerPage: 25 })\n- table.load((state) => fetchRows(state)) // sort/filter/page sent to your API\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ externalSort={true} externalFilter={true}\n+ onSortingChange={(clauses) => refetch({ sort: clauses })}\n+ onFiltersChange={(f) => refetch({ filters: f.columns })}\n+ />\n```\n\n## What you get for free\n\n- **A render component** with virtualization - you stop authoring the\n `<table>` and the `{#each}`.\n- **Excel-style filter menu**, **cell-range selection + copy**, and\n **inline editing** with typed editors.\n- **Grouping, tree data, and master/detail** when the table grows up.\n- **Enterprise features** - export, import, pivot, AI.\n\n## What you give up\n\n- **The tiny footprint.** If all you ever need is sort / filter /\n paginate over a small list, `@vincjo/datatables` stays smaller.\n- **Total markup control.** SvGrid renders the table; you theme it with\n `--sg-*` tokens and Tailwind rather than writing every `<td>`.\n\n## Frequently asked questions\n\n### Should I move from @vincjo/datatables to SvGrid?\n\nMove when a hand-rolled table stops being enough - when you need virtualization\nfor large datasets, an Excel-style filter menu, inline editing, grouping, or\ntree data. For a small sort/filter/paginate list, `@vincjo/datatables` is a fine,\nlighter choice.\n\n### Is SvGrid also MIT-licensed like @vincjo/datatables?\n\nYes. `@svgrid/grid` is MIT, like `@vincjo/datatables`. SvGrid adds an\noptional paid `@svgrid/enterprise` pack (export, import, pivot, AI).\n\n### Can I keep my own table markup after switching?\n\nYes. SvGrid has a headless core (`createSvGrid` + row-model factories) you can\ndrive with your own markup, the same way `@vincjo/datatables` lets you. Most\nteams use the `<SvGrid>` component instead because it removes the boilerplate.\n\n## What you end up with\n\nSorting, filtering and pagination without the hand-written table around it.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} />\n```\n\n## See also\n\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/) - the side-by-side comparison\n- [Migrating from svelte-headless-table](./migrating-from-svelte-headless-table.md) - sibling Svelte guide\n- [Why headless?](../why-headless.md) - keep your own markup if you want to\n"
6217
+ "markdown": "# Migrating from @vincjo/datatables\n\n`@vincjo/datatables` is a small, ergonomic Svelte datatable helper: you\nwrap your data in a handler and it gives you reactive sorting,\nfiltering, and pagination while you keep your own table markup. SvGrid\ncovers the same job and then keeps going - virtualization, an\nExcel-style filter menu, inline editing, and a render component - so the\nport is mostly about deciding how much markup you want to keep.\n\n> Estimated effort: **30 min - 2 hours** per table. A simple list is a\n> 30-minute swap; a feature-heavy one trends toward two hours.\n\n<!-- facts:start vincjo-datatables -->\n> **Facts, checked 9 Nov 2026.** `@vincjo/datatables` 2.8.1, MIT, last published 4 Jun 2026, 107,000 npm downloads in the 30 days to 8 Nov 2026. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `@vincjo/datatables` 2.8.1 10.2 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). @vincjo/datatables pricing, as its site states it: @vincjo/datatables is MIT on npm and free; no licence is sold (https://github.com/vincjo/datatables, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/).\n<!-- facts:end -->\n\n## Vocabulary cheat sheet\n\n| @vincjo/datatables | sv-grid |\n| ------------------------------------------ | ----------------------------------------- |\n| `new TableHandler(data, { rowsPerPage })` | `createSvGrid({...})` or `<SvGrid>` |\n| `<ThSort {table} field=\"name\">` | `rowSortingFeature` (header sort built in) |\n| `<ThFilter {table} field=\"name\">` | `columnFilteringFeature` |\n| `table.global.set(query)` (search) | `api.setState({ globalFilter: query })` |\n| `<Pagination {table} />` / `<RowCount>` | Built in; toggle `showPagination` |\n| `table.rows` (current page rows) | `api.getDisplayedRows()` |\n| Server mode (`table.load(...)`) | `externalSort` / `externalFilter` + refetch |\n| Your own `<table>` + `{#each table.rows}` | `<SvGrid>` render component |\n\n## Before / after\n\nThe example at the end of this page runs against these rows:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n\n type Person = {\n id: number\n name: string\n department: string\n city: string\n age: number\n salary: number\n }\n\n const people: Person[] = [\n { id: 1, name: 'Ada Lovelace', department: 'Engineering', city: 'London', age: 36, salary: 142000 },\n { id: 2, name: 'Grace Hopper', department: 'Engineering', city: 'New York', age: 45, salary: 168000 },\n { id: 3, name: 'Linus Torvalds', department: 'Platform', city: 'Portland', age: 54, salary: 155000 },\n { id: 4, name: 'Radia Perlman', department: 'Networking', city: 'Seattle', age: 49, salary: 161000 },\n { id: 5, name: 'Barbara Liskov', department: 'Platform', city: 'Boston', age: 52, salary: 172000 },\n ]\n\n let rows = $state<Person[]>(people)\n\n const columns: GridColumns<Person> = [\n { field: 'name', header: 'Name', width: 200, editorType: 'text' },\n { field: 'department', header: 'Department', width: 150, editorType: 'text' },\n { field: 'city', header: 'City', width: 140, editorType: 'text' },\n { field: 'age', header: 'Age', width: 90, editorType: 'number' },\n { field: 'salary', header: 'Salary', width: 130, editorType: 'number', format: { type: 'currency', currency: 'USD' } },\n ]\n</script>\n```\n\n```diff\n- <script>\n- import { TableHandler, ThSort, ThFilter, Pagination } from '@vincjo/datatables'\n- const table = new TableHandler(rows, { rowsPerPage: 25 })\n- </script>\n-\n- <table>\n- <thead><tr>\n- <ThSort {table} field=\"name\">Name</ThSort>\n- <ThSort {table} field=\"amount\">Amount</ThSort>\n- </tr></thead>\n- <tbody>\n- {#each table.rows as row}\n- <tr><td>{row.name}</td><td>{row.amount}</td></tr>\n- {/each}\n- </tbody>\n- </table>\n- <Pagination {table} />\n\n+ <script lang=\"ts\">\n+ import {\n+ SvGrid, tableFeatures, rowSortingFeature, columnFilteringFeature,\n+ type ColumnDef,\n+ } from '@svgrid/grid'\n+\n+ const features = tableFeatures({ rowSortingFeature, columnFilteringFeature })\n+ const columns: ColumnDef<typeof features, Row>[] = [\n+ { field: 'name', header: 'Name' },\n+ { field: 'amount', header: 'Amount', format: { type: 'currency', currency: 'USD' } },\n+ ]\n+ </script>\n+\n+ <SvGrid data={rows} columns={columns} features={features} showPagination />\n```\n\n## Server-side data\n\n```diff\n- // @vincjo/datatables: drive a server with the handler's load hook\n- const table = new TableHandler(rows, { rowsPerPage: 25 })\n- table.load((state) => fetchRows(state)) // sort/filter/page sent to your API\n\n+ <SvGrid\n+ data={rows} columns={columns} features={features}\n+ externalSort={true} externalFilter={true}\n+ onSortingChange={(clauses) => refetch({ sort: clauses })}\n+ onFiltersChange={(f) => refetch({ filters: f.columns })}\n+ />\n```\n\n## What you get for free\n\n- **A render component** with virtualization - you stop authoring the\n `<table>` and the `{#each}`.\n- **Excel-style filter menu**, **cell-range selection + copy**, and\n **inline editing** with typed editors.\n- **Grouping, tree data, and master/detail** when the table grows up.\n- **Enterprise features** - export, import, pivot, AI.\n\n## What you give up\n\n- **The tiny footprint.** If all you ever need is sort / filter /\n paginate over a small list, `@vincjo/datatables` stays smaller.\n- **Total markup control.** SvGrid renders the table; you theme it with\n `--sg-*` tokens and Tailwind rather than writing every `<td>`.\n\n## Frequently asked questions\n\n### Should I move from @vincjo/datatables to SvGrid?\n\nMove when a hand-rolled table stops being enough - when you need virtualization\nfor large datasets, an Excel-style filter menu, inline editing, grouping, or\ntree data. For a small sort/filter/paginate list, `@vincjo/datatables` is a fine,\nlighter choice.\n\n### Is SvGrid also MIT-licensed like @vincjo/datatables?\n\nYes. `@svgrid/grid` is MIT, like `@vincjo/datatables`. SvGrid adds an\noptional paid `@svgrid/enterprise` pack (export, import, pivot, AI).\n\n### Can I keep my own table markup after switching?\n\nYes. SvGrid has a headless core (`createSvGrid` + row-model factories) you can\ndrive with your own markup, the same way `@vincjo/datatables` lets you. Most\nteams use the `<SvGrid>` component instead because it removes the boilerplate.\n\n## What you end up with\n\nSorting, filtering and pagination without the hand-written table around it.\n\n```svelte {runnable}\n<SvGrid data={rows} {columns} sortable filterable pageable pageSize={3} />\n```\n\n## See also\n\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/) - the side-by-side comparison\n- [Migrating from svelte-headless-table](./migrating-from-svelte-headless-table.md) - sibling Svelte guide\n- [Why headless?](../why-headless.md) - keep your own markup if you want to\n"
6218
6218
  },
6219
6219
  {
6220
6220
  "slug": "help/missing-features",
@@ -6634,7 +6634,7 @@ export const docs = [
6634
6634
  "path": "docs/help/svelte-5-upgrade-data-tables.md",
6635
6635
  "title": "Upgrading a Svelte 4 table to Svelte 5: what breaks, what does not, and the three ways out",
6636
6636
  "section": "Help",
6637
- "markdown": "\n# Upgrading a Svelte 4 table to Svelte 5: what breaks, what does not, and the three ways out\n\nIf your Svelte 4 app has a table built on `svelte-headless-table`, the Svelte 5\nupgrade forces a decision about it, and most of what is written about that\ndecision is guesswork. This page is not. We put the table from the library's\nown documentation into a Svelte 4 project, upgraded it to Svelte 5 one step at a\ntime, and captured every line the tools printed. The script and the captures\nare in the repo at\n[tools/migration-lab](https://github.com/sv-grid/sv-grid/tree/main/tools/migration-lab);\nevery code block below is a quote from those files, checked by a test, and the\nfirst line of each capture records the versions it was made with.\n\n<!-- facts:start svelte-headless-table pkg:@humanspeak/svelte-headless-table -->\n> **Facts, checked 9 Nov 2026.** `svelte-headless-table` 0.18.3, MIT, last published 28 Oct 2024, 96,400 npm downloads in the 30 days to 8 Nov 2026. `@humanspeak/svelte-headless-table` 7.1.0, MIT, last published 28 Oct 2026, 10,600 npm downloads in the same window. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.33 99.9 KB JS + 11.3 KB CSS (measured 10 Nov 2026); `svelte-headless-table` 0.18.3 5.7 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). svelte-headless-table pricing, as its site states it: svelte-headless-table is MIT and free; no paid licence or support is offered, and the author states there are no plans for a Svelte 5 port (https://github.com/bryanmylee/svelte-headless-table, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/).\n<!-- facts:end -->\n\nThe short version: the install warns and goes through, the untouched component\nkeeps compiling, and the trouble starts the day you write your first rune in\nthat component. The three exits are a maintained fork, TanStack Table v9, and\nSvGrid, and they trade different things.\n\n## The component we upgraded\n\nA `PeopleTable` written the way svelte-headless-table's docs teach it: a\n`writable` over a prop, a `$:` to keep it in sync, three plugins, the view model\nread through `<Subscribe>` and `<Render>`, a pager on `pluginStates.page`, and a\n`select` event dispatched from the row.\n\n<!-- captured: fixture/svelte4/PeopleTable.svelte -->\n```svelte\n<script lang=\"ts\">\n // The table as svelte-headless-table's own docs show it: a store for the\n // data, plugins for behaviour, and a view model read through <Subscribe>\n // and <Render>. Svelte 4 throughout: `export let`, `$:`, a dispatcher,\n // `on:click`, and `let:` slot props.\n import { createEventDispatcher } from 'svelte'\n import { writable } from 'svelte/store'\n import { createTable, Subscribe, Render } from 'svelte-headless-table'\n import { addSortBy, addColumnFilters, addPagination } from 'svelte-headless-table/plugins'\n import type { Person } from './people'\n\n export let people: Person[] = []\n export let pageSize = 5\n\n const dispatch = createEventDispatcher<{ select: Person }>()\n\n const data = writable<Person[]>(people)\n $: data.set(people)\n\n const table = createTable(data, {\n sort: addSortBy(),\n filter: addColumnFilters(),\n page: addPagination({ initialPageSize: pageSize }),\n })\n```\n\nThe full file is 83 lines; the 40 that follow are the `<table>`, three levels\nof `<Subscribe let:...>` and a `<p>` with the pager. On the pinned Svelte 4\ntoolchain it checks clean:\n\n<!-- captured: 01-svelte4-baseline.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\nsvelte-check found 0 errors and 0 warnings\n```\n\n## Step 1: the install warns and goes through\n\n`svelte-headless-table` declares `peerDependencies: { svelte: \"^4.0.0\" }`, and\nthat range is not going to change: the facts box above has the last release\ndate, and the project's README says there are no plans for a Svelte 5 port.\nSo what does `npm install svelte@5` do against it? On npm 11, this:\n\n<!-- captured: 02-npm-install-svelte5.txt -->\n```text\n$ npm install svelte@5.57.1\n\nadded 10 packages, removed 10 packages, and changed 2 packages in <time>\nnpm warn ERESOLVE overriding peer dependency\nnpm warn While resolving: svgrid-migration-lab@undefined\nnpm warn Found: svelte@4.2.20\nnpm warn node_modules/svelte\nnpm warn peer svelte@\"^4.0.0 || ^5.0.0-next.0\" from svelte-check@4.7.6\nnpm warn node_modules/svelte-check\nnpm warn dev svelte-check@\"4.7.6\" from the root project\nnpm warn 5 more (svelte-headless-table, svelte-keyed, svelte-render, ...)\nnpm warn\nnpm warn Could not resolve dependency:\nnpm warn peer svelte@\"^4.0.0\" from svelte-headless-table@0.18.3\nnpm warn node_modules/svelte-headless-table\nnpm warn dev svelte-headless-table@\"0.18.3\" from the root project\nnpm warn ERESOLVE overriding peer dependency\nnpm warn ERESOLVE overriding peer dependency\nnpm warn ERESOLVE overriding peer dependency\n\n(exit 0)\n```\n\nExit 0. npm 11 overrides the conflict and tells you so; older npm releases\nstopped with an `ERESOLVE` error until you added `--legacy-peer-deps`, and\npnpm prints an unmet-peer warning of its own. Either way, the peer range is a\nstatement, not a lock: the library's author is telling you it was never run\nagainst the Svelte you now have. Nothing enforces it.\n\n## Step 2: the untouched component still compiles\n\nSvelte 5 compiles Svelte 4 components in legacy mode, so `export let`, `$:`,\n`createEventDispatcher`, `on:click` and `let:` slot props all keep working\nuntil you opt a file into runes. With Svelte 5 installed and the component\nexactly as it was:\n\n<!-- captured: 03-svelte5-legacy-component.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\nsvelte-check found 0 errors and 0 warnings\n```\n\nThis is the part the scare stories leave out. If your only goal is to be on\nSvelte 5, you can install it, keep the table exactly as it is, and ship. The\nlibrary's own components (`<Subscribe>`, `<Render>`) compile in legacy mode\ntoo.\n\n## Step 3: the first rune\n\nWhat you cannot do is make that component a runes component and keep the\ntable as it is. Port `PeopleTable` by the rules in the\n[official migration guide](https://svelte.dev/docs/svelte/v5-migration-guide):\n`export let` becomes `$props()`, `$:` becomes `$derived` and `$effect`, the\ndispatcher becomes a callback prop, `on:click` becomes `onclick`. Leave the\nsvelte-headless-table wiring alone, because the guide has nothing to say about\nit. The checker now says:\n\n<!-- captured: 04-svelte5-runes-component.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\n<lab>\\src\\PeopleTable.svelte:19:35\nWarn: This reference only captures the initial value of `people`. Did you mean to reference it inside a closure instead?\nhttps://svelte.dev/e/state_referenced_locally (svelte)\n\n<lab>\\src\\PeopleTable.svelte:27:44\nWarn: This reference only captures the initial value of `pageSize`. Did you mean to reference it inside a closure instead?\nhttps://svelte.dev/e/state_referenced_locally (svelte)\n\n====================================\nsvelte-check found 0 errors and 2 warnings in 1 file\n```\n\nTwo warnings, no errors, and both are the same fact: the library takes a store,\nyour data is now a rune, and the `writable(people)` you built to bridge them\ncaptured the first value of `people` and nothing after. The fix is an\n`$effect` that writes the prop into the store on every change, which is what\nthe ported fixture does, and it works. It is also a bridge you now own, in\nevery component that has a table, between a reactivity model the library was\nbuilt for and the one your app is written in.\n\nThat is the actual shape of the problem. Not a red build: a component that\ncompiles, with a store-and-slot-props library bolted to a runes app by code you\nmaintain, on a package that will not move. The pieces it leans on are the ones\nSvelte 5 has deprecated: `let:` slot props and `<slot>` in favour of snippets,\nand `<svelte:component>`, which is what `<Render>` uses underneath\n(`svelte-render` 2.0.1, `PropsRenderer.svelte`, peer `svelte ^4.0.0`). They\nkeep working in legacy mode today; deprecated syntax is what the next major\nremoves, and this library will not be updated when it does.\n\n## The three ways out\n\nYou have three honest options, and you should know all of them before picking.\n\n1. **`@humanspeak/svelte-headless-table`**, a maintained fork on Svelte 5 with\n the same API. Its version, publish date and downloads are in the facts box.\n Switching is one package rename; you keep every line of markup and every\n `<Subscribe>`, and you keep the store bridge from step 3. If your table\n works and Svelte 5 is all you need, this is the cheapest exit by a wide\n margin.\n2. **TanStack Table v9.** Its Svelte adapter declares a Svelte 5 peer range and\n its state shape is runes-first, so the bridge goes away. It is still\n headless: you write and maintain the `<table>`, the virtualizer, the filter\n UI and the editors yourself. The\n [TanStack migration guide](./migrating-from-tanstack-table.md) covers the\n vocabulary.\n3. **SvGrid.** A different trade: you delete the markup and take a renderer.\n The column definitions and the plugin list carry over; the `<table>`, the\n `<Subscribe>` blocks, the pager and the store bridge do not exist any more,\n because `<SvGrid>` renders all of that. The rest of this page is what that\n looks like on the fixture above.\n\nPick 1 if the markup is fine and the library being dead does not bother you.\nPick 2 if you want the markup and a maintained engine. Pick 3 if the markup is\nthe part you are tired of.\n\n## Step 4: the codemod\n\n`npx @svgrid/migrate` ports a svelte-headless-table component to SvGrid. It\npreviews by default and only writes with `--write`, and it says what it could\nnot map instead of dropping it. On the Svelte 4 original:\n\n<!-- captured: 05-codemod.txt -->\n```text\n warning A custom `cell` renderer was found. SvGrid renders cells with a snippet, not `createRender`, so this one needs porting by hand - it is preserved as a comment.\n warning The template still reads `pageIndex`, `pageCount`, `shown` from the view model, which is gone. <SvGrid> renders its own pager and row count (`pageable`, `showPagination`), so remove that markup, or read the grid through `onApiReady` if you need the numbers.\n warning The table markup dispatched `select` from a row; that handler went with the `<tr>`. Wire it to `onRowClick={({ row }) => ...}` (or `onCellClick`) on <SvGrid>, and turn the `createEventDispatcher` into a callback prop if the component is moving to runes.\n note Column resizing, keyboard navigation and ARIA grid semantics are built into <SvGrid>; the markup that provided them is intentionally gone.\n\n1 file(s) would change, with 3 warning(s).\nRe-run with --write to apply.\n```\n\nThree warnings, each naming a thing it left for you. Write it, swap the\npackages, and check:\n\n<!-- captured: 06-after-codemod.txt -->\n```text\n$ npm install @svgrid/grid@3.0.4 --legacy-peer-deps\n\nadded 1 package in <time>\n\n(exit 0)\n\n$ npm uninstall svelte-headless-table --legacy-peer-deps\n\nremoved 4 packages in <time>\n\n(exit 0)\n\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\n<lab>\\src\\PeopleTable.svelte:31:23\nError: `$pageIndex` is an illegal variable name. To reference a global variable called `$pageIndex`, use `globalThis.$pageIndex`\nhttps://svelte.dev/e/global_reference_invalid (svelte)\n```\n\n<!-- captured: 06-after-codemod.txt -->\n```text\nsvelte-check found 9 errors and 0 warnings in 1 file\n```\n\nAll nine are the pager `<p>` the second warning named, reading `shown`,\n`$pageIndex` and `$pageCount` from a view model that no longer exists. The\ncodemod does not delete markup outside the `<table>`, because it cannot know\nwhat else is in there; it tells you what will not compile. Delete the `<p>`\n(the grid draws its own pager), turn the `select` event into a callback prop\non `onRowClick` as the third warning says, and:\n\n<!-- captured: 07-hand-edits.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\nsvelte-check found 0 errors and 0 warnings\n```\n\n## What you end up with\n\nThe component after the codemod and the two hand edits:\n\n<!-- captured: PeopleTable.after.svelte -->\n```svelte\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n import type { Person } from './people'\n export let people: Person[] = []\n export let pageSize = 5\n export let onselect: (person: Person) => void = () => {}\n\n const columns: GridColumns<(typeof people)[number]> = [\n {\n header: 'Name',\n field: 'name',\n },\n {\n header: 'Department',\n field: 'department',\n },\n {\n header: 'City',\n field: 'city',\n },\n {\n header: 'Salary',\n field: 'salary',\n // TODO port: cell: ({ value }) => value.toLocaleString('en-US', { style: 'currency', currency: 'USD\n },\n ]\n</script>\n\n<SvGrid data={people} {columns} sortable filterable showColumnFilters pageable pageSize={pageSize}\n onRowClick={({ row }) => onselect(row)} />\n```\n\n83 lines became 31. The `TODO port:` comment is the one thing left to do by\nhand: the currency cell becomes `format: { type: 'currency', currency: 'USD' }`\non the column, and the comment goes. There is no store, no `$:`, no\n`<Subscribe>`, no `let:`, and nothing in the file that Svelte 5 has deprecated,\nso the same component moves to runes whenever you like with the official\nmigration script and nothing else to port.\n\n## What to read next\n\n- [Migrating from svelte-headless-table](./migrating-from-svelte-headless-table.md) -\n the plugin-by-plugin mapping and what the codemod will not do\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/) -\n the side-by-side, with a source and date for every claim\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - exit 2 in detail\n- [Migrating from the shadcn-svelte data table](./migrating-from-shadcn-data-table.md) -\n the same codemod on the TanStack-based recipe\n- [Why headless?](../why-headless.md) - if you want to keep your own markup\n on a maintained Svelte 5 engine, `@svgrid/grid/core` is that too\n\n## Frequently asked questions\n\n### Does svelte-headless-table work with Svelte 5?\n\nIn legacy mode, yes: the install prints a peer-dependency warning and the\nuntouched component type-checks clean, as the captures above show. What does\nnot work is moving a component that uses it to runes without maintaining a\nstore bridge yourself, and the library is not going to be updated for the\ndeprecations that bridge leans on.\n\n### Is the peer dependency conflict an error?\n\nOn npm 11 it is a warning and the install exits 0; on older npm it was an\n`ERESOLVE` error until you passed `--legacy-peer-deps`; pnpm warns. It never\nstops the code from running. It tells you the library was not built against\nthe Svelte you have.\n\n### Should I just use the maintained fork?\n\nIf your table works and all you need is Svelte 5, yes, and this page says so.\nThe fork keeps the same API, so you keep the same markup and the same store\nbridge. Move to SvGrid when the markup and the bridge are what you want gone.\n\n### How much of the port does the codemod do?\n\nThe column definitions, the plugin list and the `<table>` replacement, and it\nnames what it leaves: custom cell renderers, markup outside the table that read\nthe view model, and events the rows dispatched. On the fixture that was two\nhand edits and one `format` option.\n\n### Can I rerun this myself?\n\n`node tools/migration-lab/run.mjs` in the repo. It installs the pinned Svelte 4\ntoolchain into its own folder, upgrades it, runs the codemod from the checkout,\nand rewrites the captured files; the header of each one records the versions.\nBump the pins to try a newer Svelte.\n"
6637
+ "markdown": "\n# Upgrading a Svelte 4 table to Svelte 5: what breaks, what does not, and the three ways out\n\nIf your Svelte 4 app has a table built on `svelte-headless-table`, the Svelte 5\nupgrade forces a decision about it, and most of what is written about that\ndecision is guesswork. This page is not. We put the table from the library's\nown documentation into a Svelte 4 project, upgraded it to Svelte 5 one step at a\ntime, and captured every line the tools printed. The script and the captures\nare in the repo at\n[tools/migration-lab](https://github.com/sv-grid/sv-grid/tree/main/tools/migration-lab);\nevery code block below is a quote from those files, checked by a test, and the\nfirst line of each capture records the versions it was made with.\n\n<!-- facts:start svelte-headless-table pkg:@humanspeak/svelte-headless-table -->\n> **Facts, checked 9 Nov 2026.** `svelte-headless-table` 0.18.3, MIT, last published 28 Oct 2024, 96,400 npm downloads in the 30 days to 8 Nov 2026. `@humanspeak/svelte-headless-table` 7.1.0, MIT, last published 28 Oct 2026, 10,600 npm downloads in the same window. `@svgrid/grid` 3.0.32, MIT, last published 3 Nov 2026, 44,600 npm downloads in the same window. Bundle, minified and gzipped, each package built alone with Svelte external: SvGrid 3.0.34 99.6 KB JS + 11.4 KB CSS (measured 10 Nov 2026); `svelte-headless-table` 0.18.3 5.7 KB JS, no separate stylesheet, svelte external (measured 9 Nov 2026). svelte-headless-table pricing, as its site states it: svelte-headless-table is MIT and free; no paid licence or support is offered, and the author states there are no plans for a Svelte 5 port (https://github.com/bryanmylee/svelte-headless-table, read 12 Sep 2026). SvGrid: MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise. Side by side, with sources: [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/).\n<!-- facts:end -->\n\nThe short version: the install warns and goes through, the untouched component\nkeeps compiling, and the trouble starts the day you write your first rune in\nthat component. The three exits are a maintained fork, TanStack Table v9, and\nSvGrid, and they trade different things.\n\n## The component we upgraded\n\nA `PeopleTable` written the way svelte-headless-table's docs teach it: a\n`writable` over a prop, a `$:` to keep it in sync, three plugins, the view model\nread through `<Subscribe>` and `<Render>`, a pager on `pluginStates.page`, and a\n`select` event dispatched from the row.\n\n<!-- captured: fixture/svelte4/PeopleTable.svelte -->\n```svelte\n<script lang=\"ts\">\n // The table as svelte-headless-table's own docs show it: a store for the\n // data, plugins for behaviour, and a view model read through <Subscribe>\n // and <Render>. Svelte 4 throughout: `export let`, `$:`, a dispatcher,\n // `on:click`, and `let:` slot props.\n import { createEventDispatcher } from 'svelte'\n import { writable } from 'svelte/store'\n import { createTable, Subscribe, Render } from 'svelte-headless-table'\n import { addSortBy, addColumnFilters, addPagination } from 'svelte-headless-table/plugins'\n import type { Person } from './people'\n\n export let people: Person[] = []\n export let pageSize = 5\n\n const dispatch = createEventDispatcher<{ select: Person }>()\n\n const data = writable<Person[]>(people)\n $: data.set(people)\n\n const table = createTable(data, {\n sort: addSortBy(),\n filter: addColumnFilters(),\n page: addPagination({ initialPageSize: pageSize }),\n })\n```\n\nThe full file is 83 lines; the 40 that follow are the `<table>`, three levels\nof `<Subscribe let:...>` and a `<p>` with the pager. On the pinned Svelte 4\ntoolchain it checks clean:\n\n<!-- captured: 01-svelte4-baseline.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\nsvelte-check found 0 errors and 0 warnings\n```\n\n## Step 1: the install warns and goes through\n\n`svelte-headless-table` declares `peerDependencies: { svelte: \"^4.0.0\" }`, and\nthat range is not going to change: the facts box above has the last release\ndate, and the project's README says there are no plans for a Svelte 5 port.\nSo what does `npm install svelte@5` do against it? On npm 11, this:\n\n<!-- captured: 02-npm-install-svelte5.txt -->\n```text\n$ npm install svelte@5.57.1\n\nadded 10 packages, removed 10 packages, and changed 2 packages in <time>\nnpm warn ERESOLVE overriding peer dependency\nnpm warn While resolving: svgrid-migration-lab@undefined\nnpm warn Found: svelte@4.2.20\nnpm warn node_modules/svelte\nnpm warn peer svelte@\"^4.0.0 || ^5.0.0-next.0\" from svelte-check@4.7.6\nnpm warn node_modules/svelte-check\nnpm warn dev svelte-check@\"4.7.6\" from the root project\nnpm warn 5 more (svelte-headless-table, svelte-keyed, svelte-render, ...)\nnpm warn\nnpm warn Could not resolve dependency:\nnpm warn peer svelte@\"^4.0.0\" from svelte-headless-table@0.18.3\nnpm warn node_modules/svelte-headless-table\nnpm warn dev svelte-headless-table@\"0.18.3\" from the root project\nnpm warn ERESOLVE overriding peer dependency\nnpm warn ERESOLVE overriding peer dependency\nnpm warn ERESOLVE overriding peer dependency\n\n(exit 0)\n```\n\nExit 0. npm 11 overrides the conflict and tells you so; older npm releases\nstopped with an `ERESOLVE` error until you added `--legacy-peer-deps`, and\npnpm prints an unmet-peer warning of its own. Either way, the peer range is a\nstatement, not a lock: the library's author is telling you it was never run\nagainst the Svelte you now have. Nothing enforces it.\n\n## Step 2: the untouched component still compiles\n\nSvelte 5 compiles Svelte 4 components in legacy mode, so `export let`, `$:`,\n`createEventDispatcher`, `on:click` and `let:` slot props all keep working\nuntil you opt a file into runes. With Svelte 5 installed and the component\nexactly as it was:\n\n<!-- captured: 03-svelte5-legacy-component.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\nsvelte-check found 0 errors and 0 warnings\n```\n\nThis is the part the scare stories leave out. If your only goal is to be on\nSvelte 5, you can install it, keep the table exactly as it is, and ship. The\nlibrary's own components (`<Subscribe>`, `<Render>`) compile in legacy mode\ntoo.\n\n## Step 3: the first rune\n\nWhat you cannot do is make that component a runes component and keep the\ntable as it is. Port `PeopleTable` by the rules in the\n[official migration guide](https://svelte.dev/docs/svelte/v5-migration-guide):\n`export let` becomes `$props()`, `$:` becomes `$derived` and `$effect`, the\ndispatcher becomes a callback prop, `on:click` becomes `onclick`. Leave the\nsvelte-headless-table wiring alone, because the guide has nothing to say about\nit. The checker now says:\n\n<!-- captured: 04-svelte5-runes-component.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\n<lab>\\src\\PeopleTable.svelte:19:35\nWarn: This reference only captures the initial value of `people`. Did you mean to reference it inside a closure instead?\nhttps://svelte.dev/e/state_referenced_locally (svelte)\n\n<lab>\\src\\PeopleTable.svelte:27:44\nWarn: This reference only captures the initial value of `pageSize`. Did you mean to reference it inside a closure instead?\nhttps://svelte.dev/e/state_referenced_locally (svelte)\n\n====================================\nsvelte-check found 0 errors and 2 warnings in 1 file\n```\n\nTwo warnings, no errors, and both are the same fact: the library takes a store,\nyour data is now a rune, and the `writable(people)` you built to bridge them\ncaptured the first value of `people` and nothing after. The fix is an\n`$effect` that writes the prop into the store on every change, which is what\nthe ported fixture does, and it works. It is also a bridge you now own, in\nevery component that has a table, between a reactivity model the library was\nbuilt for and the one your app is written in.\n\nThat is the actual shape of the problem. Not a red build: a component that\ncompiles, with a store-and-slot-props library bolted to a runes app by code you\nmaintain, on a package that will not move. The pieces it leans on are the ones\nSvelte 5 has deprecated: `let:` slot props and `<slot>` in favour of snippets,\nand `<svelte:component>`, which is what `<Render>` uses underneath\n(`svelte-render` 2.0.1, `PropsRenderer.svelte`, peer `svelte ^4.0.0`). They\nkeep working in legacy mode today; deprecated syntax is what the next major\nremoves, and this library will not be updated when it does.\n\n## The three ways out\n\nYou have three honest options, and you should know all of them before picking.\n\n1. **`@humanspeak/svelte-headless-table`**, a maintained fork on Svelte 5 with\n the same API. Its version, publish date and downloads are in the facts box.\n Switching is one package rename; you keep every line of markup and every\n `<Subscribe>`, and you keep the store bridge from step 3. If your table\n works and Svelte 5 is all you need, this is the cheapest exit by a wide\n margin.\n2. **TanStack Table v9.** Its Svelte adapter declares a Svelte 5 peer range and\n its state shape is runes-first, so the bridge goes away. It is still\n headless: you write and maintain the `<table>`, the virtualizer, the filter\n UI and the editors yourself. The\n [TanStack migration guide](./migrating-from-tanstack-table.md) covers the\n vocabulary.\n3. **SvGrid.** A different trade: you delete the markup and take a renderer.\n The column definitions and the plugin list carry over; the `<table>`, the\n `<Subscribe>` blocks, the pager and the store bridge do not exist any more,\n because `<SvGrid>` renders all of that. The rest of this page is what that\n looks like on the fixture above.\n\nPick 1 if the markup is fine and the library being dead does not bother you.\nPick 2 if you want the markup and a maintained engine. Pick 3 if the markup is\nthe part you are tired of.\n\n## Step 4: the codemod\n\n`npx @svgrid/migrate` ports a svelte-headless-table component to SvGrid. It\npreviews by default and only writes with `--write`, and it says what it could\nnot map instead of dropping it. On the Svelte 4 original:\n\n<!-- captured: 05-codemod.txt -->\n```text\n warning A custom `cell` renderer was found. SvGrid renders cells with a snippet, not `createRender`, so this one needs porting by hand - it is preserved as a comment.\n warning The template still reads `pageIndex`, `pageCount`, `shown` from the view model, which is gone. <SvGrid> renders its own pager and row count (`pageable`, `showPagination`), so remove that markup, or read the grid through `onApiReady` if you need the numbers.\n warning The table markup dispatched `select` from a row; that handler went with the `<tr>`. Wire it to `onRowClick={({ row }) => ...}` (or `onCellClick`) on <SvGrid>, and turn the `createEventDispatcher` into a callback prop if the component is moving to runes.\n note Column resizing, keyboard navigation and ARIA grid semantics are built into <SvGrid>; the markup that provided them is intentionally gone.\n\n1 file(s) would change, with 3 warning(s).\nRe-run with --write to apply.\n```\n\nThree warnings, each naming a thing it left for you. Write it, swap the\npackages, and check:\n\n<!-- captured: 06-after-codemod.txt -->\n```text\n$ npm install @svgrid/grid@3.0.4 --legacy-peer-deps\n\nadded 1 package in <time>\n\n(exit 0)\n\n$ npm uninstall svelte-headless-table --legacy-peer-deps\n\nremoved 4 packages in <time>\n\n(exit 0)\n\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\n<lab>\\src\\PeopleTable.svelte:31:23\nError: `$pageIndex` is an illegal variable name. To reference a global variable called `$pageIndex`, use `globalThis.$pageIndex`\nhttps://svelte.dev/e/global_reference_invalid (svelte)\n```\n\n<!-- captured: 06-after-codemod.txt -->\n```text\nsvelte-check found 9 errors and 0 warnings in 1 file\n```\n\nAll nine are the pager `<p>` the second warning named, reading `shown`,\n`$pageIndex` and `$pageCount` from a view model that no longer exists. The\ncodemod does not delete markup outside the `<table>`, because it cannot know\nwhat else is in there; it tells you what will not compile. Delete the `<p>`\n(the grid draws its own pager), turn the `select` event into a callback prop\non `onRowClick` as the third warning says, and:\n\n<!-- captured: 07-hand-edits.txt -->\n```text\n$ npx svelte-check --tsconfig ./tsconfig.json --output human --threshold warning\nsvelte-check found 0 errors and 0 warnings\n```\n\n## What you end up with\n\nThe component after the codemod and the two hand edits:\n\n<!-- captured: PeopleTable.after.svelte -->\n```svelte\n<script lang=\"ts\">\n import { SvGrid, type GridColumns } from '@svgrid/grid'\n import type { Person } from './people'\n export let people: Person[] = []\n export let pageSize = 5\n export let onselect: (person: Person) => void = () => {}\n\n const columns: GridColumns<(typeof people)[number]> = [\n {\n header: 'Name',\n field: 'name',\n },\n {\n header: 'Department',\n field: 'department',\n },\n {\n header: 'City',\n field: 'city',\n },\n {\n header: 'Salary',\n field: 'salary',\n // TODO port: cell: ({ value }) => value.toLocaleString('en-US', { style: 'currency', currency: 'USD\n },\n ]\n</script>\n\n<SvGrid data={people} {columns} sortable filterable showColumnFilters pageable pageSize={pageSize}\n onRowClick={({ row }) => onselect(row)} />\n```\n\n83 lines became 31. The `TODO port:` comment is the one thing left to do by\nhand: the currency cell becomes `format: { type: 'currency', currency: 'USD' }`\non the column, and the comment goes. There is no store, no `$:`, no\n`<Subscribe>`, no `let:`, and nothing in the file that Svelte 5 has deprecated,\nso the same component moves to runes whenever you like with the official\nmigration script and nothing else to port.\n\n## What to read next\n\n- [Migrating from svelte-headless-table](./migrating-from-svelte-headless-table.md) -\n the plugin-by-plugin mapping and what the codemod will not do\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/) -\n the side-by-side, with a source and date for every claim\n- [Migrating from TanStack Table](./migrating-from-tanstack-table.md) - exit 2 in detail\n- [Migrating from the shadcn-svelte data table](./migrating-from-shadcn-data-table.md) -\n the same codemod on the TanStack-based recipe\n- [Why headless?](../why-headless.md) - if you want to keep your own markup\n on a maintained Svelte 5 engine, `@svgrid/grid/core` is that too\n\n## Frequently asked questions\n\n### Does svelte-headless-table work with Svelte 5?\n\nIn legacy mode, yes: the install prints a peer-dependency warning and the\nuntouched component type-checks clean, as the captures above show. What does\nnot work is moving a component that uses it to runes without maintaining a\nstore bridge yourself, and the library is not going to be updated for the\ndeprecations that bridge leans on.\n\n### Is the peer dependency conflict an error?\n\nOn npm 11 it is a warning and the install exits 0; on older npm it was an\n`ERESOLVE` error until you passed `--legacy-peer-deps`; pnpm warns. It never\nstops the code from running. It tells you the library was not built against\nthe Svelte you have.\n\n### Should I just use the maintained fork?\n\nIf your table works and all you need is Svelte 5, yes, and this page says so.\nThe fork keeps the same API, so you keep the same markup and the same store\nbridge. Move to SvGrid when the markup and the bridge are what you want gone.\n\n### How much of the port does the codemod do?\n\nThe column definitions, the plugin list and the `<table>` replacement, and it\nnames what it leaves: custom cell renderers, markup outside the table that read\nthe view model, and events the rows dispatched. On the fixture that was two\nhand edits and one `format` option.\n\n### Can I rerun this myself?\n\n`node tools/migration-lab/run.mjs` in the repo. It installs the pinned Svelte 4\ntoolchain into its own folder, upgrades it, runs the codemod from the checkout,\nand rewrites the captured files; the header of each one records the versions.\nBump the pins to try a newer Svelte.\n"
6638
6638
  },
6639
6639
  {
6640
6640
  "slug": "help/tailwind",
@@ -7439,7 +7439,7 @@ export const docs = [
7439
7439
  "path": "docs/help/ui-components/sv-rich-text.md",
7440
7440
  "title": "SvRichText",
7441
7441
  "section": "UI components",
7442
- "markdown": "# SvRichText\n\nA lightweight WYSIWYG editor: a formatting toolbar over a `contentEditable`\nregion that emits HTML through a bindable `value`.\n\n`SvRichText` covers the common editing commands - bold, italic, underline,\nstrikethrough, headings, lists, quote, code block, alignment, links, and\nundo/redo - in a compact toolbar you can reorder or trim. It emits an HTML string\nvia `bind:value` and reflects the active formatting state (bold/italic and list\nbuttons light up with the caret). It uses `document.execCommand`, the pragmatic\napproach for a component-kit editor. All color comes from the `--sg-*` tokens.\n\n`value` goes through `sanitizeHtml` on its way in and out, so a stored string\ncannot run a script: event attributes, `javascript:` links and any style but a\ntext alignment are removed.\n\nRelated: [Inputs overview](inputs.md) · [SvForm](sv-form.md) · [Navigation & rich overview](navigation.md)\n\n## Installation\n\nAdd it with the CLI - this drops a ready-to-edit `SvRichText` starter into your app:\n\n<div data-docs-add=\"add rich-text\"></div>\n\nPrefer to see it first? `npx @svgrid/ui try rich-text` opens it in a throwaway sandbox - no project needed.\n\nOr install the package and import it directly. `SvRichText` ships free in\n`@svgrid/grid` (dependency-free):\n\n<div data-docs-install=\"@svgrid/grid\"></div>\n\nThe examples on this page import from `@svgrid/grid`:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvButton, SvRichText } from '@svgrid/grid'\n</script>\n```\n\n```ts\nimport { SvRichText } from '@svgrid/grid'\n```\n\n## Example\n\n<div data-docs-demo=\"341-rich-text-editor\" data-height=\"440\" data-code></div>\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvRichText } from '@svgrid/grid'\n let html = $state('<p>Hello</p>')\n</script>\n\n<SvRichText bind:value={html} placeholder=\"Write something…\" />\n```\n\n## Props\n\n| Prop | Type | Default | Description |\n| ------------- | --------------------------- | --------------------- | -------------------------------------------------------------- |\n| `value` | `string` | `''` | HTML string (bindable). |\n| `onChange` | `(html: string) => void` | - | Fires with the HTML on every edit. |\n| `placeholder` | `string` | `''` | Shown when the editor is empty. |\n| `disabled` | `boolean` | `false` | Dims the editor and blocks all input. |\n| `readonly` | `boolean` | `false` | Hides the toolbar and makes the body non-editable. |\n| `minHeight` | `string` | `160px` | Minimum height of the editable body (any CSS length). |\n| `tools` | `RichTextTool[]` | full default set | Toolbar tools and their order; use `'\\|'` for a separator. |\n| `ariaLabel` | `string` | `Rich text editor` | Accessible name for the editable region. |\n\n`RichTextTool` (exported) is one of:\n`'bold' | 'italic' | 'underline' | 'strike' | 'h1' | 'h2' | 'h3' | 'p' | 'ul' | 'ol' | 'quote' | 'code' | 'alignLeft' | 'alignCenter' | 'alignRight' | 'link' | 'clear' | 'undo' | 'redo' | '|'`.\nThe `'|'` entry draws a visual separator between tool groups.\n\n## Examples\n\n### A trimmed toolbar\n\nPass just the tools you need; group them with `'|'` separators:\n\n```svelte\n<SvRichText\n bind:value={note}\n tools={['bold', 'italic', '|', 'ul', 'ol', '|', 'link']}\n/>\n```\n\n### Read-only rendering\n\nSet `readonly` to show formatted content without a toolbar - useful for previews\nof the same HTML the editor produced:\n\n```svelte\n<SvRichText value={savedHtml} readonly />\n```\n\n### Reacting to edits\n\nPersist on change instead of only on blur:\n\n```svelte\n<SvRichText bind:value={html} onChange={(h) => queueSave(h)} />\n```\n\n### A field bound to a record, with preview\n\nBind `value` straight to a field on your state object so edits flow back into the\nmodel. Track a `dirty` flag from `onChange` to gate the Save button, then flip to\n`readonly` to preview the exact HTML the editor produced:\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvRichText, SvButton } from '@svgrid/grid'\n\n let ticket = $state({ id: 42, notes: '<p>Initial triage notes.</p>' })\n let editing = $state(true)\n let dirty = $state(false)\n\n function save() {\n fetch(`/api/tickets/${ticket.id}`, {\n method: 'PATCH',\n body: JSON.stringify({ notes: ticket.notes }),\n })\n dirty = false\n editing = false\n }\n</script>\n\n{#if editing}\n <SvRichText\n bind:value={ticket.notes}\n tools={['bold', 'italic', '|', 'ul', 'ol', '|', 'link', 'clear']}\n minHeight=\"200px\"\n placeholder=\"Add triage notes...\"\n onChange={() => (dirty = true)}\n />\n <SvButton variant=\"primary\" disabled={!dirty} onclick={save}>Save</SvButton>\n{:else}\n <SvRichText value={ticket.notes} readonly />\n <SvButton onclick={() => (editing = true)}>Edit</SvButton>\n{/if}\n```\n\nTip: an external change to `value` is written into the editor only while it is\nnot focused - writing `innerHTML` mid-caret would collapse the selection - so a\nvalue set from elsewhere applies once the field blurs. `readonly` mode drops the\ntoolbar and renders the same HTML, making it a zero-cost preview of what you saved.\n\n## Accessibility\n\n- The editable region is a `role=\"textbox\"` with `aria-multiline` and an\n `aria-label`.\n- The toolbar is a `role=\"toolbar\"` wired to the body via `aria-controls`.\n- Toggle tools (bold, italic, lists) expose `aria-pressed` reflecting the current\n caret state; every button has a descriptive label.\n\n## Disabled and read-only\n\nDisabled takes the control out of the tab order; read-only keeps it focusable and copyable. Reach for read-only when the value still matters to the reader.\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvRichText } from '@svgrid/grid'\n\n let html = $state('')\n</script>\n\n<SvRichText bind:value={html} disabled />\n\n<SvRichText bind:value={html} readonly />\n```\n\n## See also\n\n- [Navigation overview](navigation.md) - the wayfinding family at a glance.\n- [Inputs overview](inputs.md) - the text and value editors.\n- [SvForm](sv-form.md) - collect rich text alongside other fields.\n"
7442
+ "markdown": "# SvRichText\n\nA lightweight WYSIWYG editor: a formatting toolbar over a `contentEditable`\nregion that emits HTML through a bindable `value`.\n\n`SvRichText` covers the common editing commands - bold, italic, underline,\nstrikethrough, headings, lists, quote, code block, alignment, links, and\nundo/redo - in a compact toolbar you can reorder or trim. It emits an HTML string\nvia `bind:value` and reflects the active formatting state (bold/italic and list\nbuttons light up with the caret). It uses `document.execCommand`, the pragmatic\napproach for a component-kit editor. All color comes from the `--sg-*` tokens.\n\n`value` goes through `sanitizeHtml` on its way in and out, so a stored string\ncannot run a script: event attributes, `javascript:` links and any style but a\ntext alignment are removed.\n\nFour more tools can be listed in `tools` for page and article editing: `image`\n(uploaded through `onImageUpload`), `table`, `hr` and `source` (the HTML itself).\nWith `name`, the editor posts its value with a plain form.\n\nRelated: [Inputs overview](inputs.md) · [SvForm](sv-form.md) · [Navigation & rich overview](navigation.md)\n\n## Installation\n\nAdd it with the CLI - this drops a ready-to-edit `SvRichText` starter into your app:\n\n<div data-docs-add=\"add rich-text\"></div>\n\nPrefer to see it first? `npx @svgrid/ui try rich-text` opens it in a throwaway sandbox - no project needed.\n\nOr install the package and import it directly. `SvRichText` ships free in\n`@svgrid/grid` (dependency-free):\n\n<div data-docs-install=\"@svgrid/grid\"></div>\n\nThe examples on this page import from `@svgrid/grid`:\n\n```svelte {preamble}\n<script lang=\"ts\">\n import { SvButton, SvRichText } from '@svgrid/grid'\n</script>\n```\n\n```ts\nimport { SvRichText } from '@svgrid/grid'\n```\n\n## Example\n\n<div data-docs-demo=\"341-rich-text-editor\" data-height=\"440\" data-code></div>\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvRichText } from '@svgrid/grid'\n let html = $state('<p>Hello</p>')\n</script>\n\n<SvRichText bind:value={html} placeholder=\"Write something…\" />\n```\n\n## Props\n\n| Prop | Type | Default | Description |\n| ------------- | --------------------------- | --------------------- | -------------------------------------------------------------- |\n| `value` | `string` | `''` | HTML string (bindable). |\n| `onChange` | `(html: string) => void` | - | Fires with the HTML on every edit. |\n| `placeholder` | `string` | `''` | Shown when the editor is empty. |\n| `disabled` | `boolean` | `false` | Dims the editor and blocks all input. |\n| `readonly` | `boolean` | `false` | Hides the toolbar and makes the body non-editable. |\n| `minHeight` | `string` | `160px` | Minimum height of the editable body (any CSS length). |\n| `tools` | `RichTextTool[]` | full default set | Toolbar tools and their order; use `'\\|'` for a separator. |\n| `ariaLabel` | `string` | `Rich text editor` | Accessible name for the editable region. |\n| `messages` | `Partial<RichTextMessages>` | - | Any toolbar label or prompt, for localization. |\n| `onImageUpload` | `(file: File) => Promise<string \\| null>` | - | The `image` tool's upload: resolve to the URL to insert. Without it, the tool asks for a URL. |\n| `sanitize` | `(html: string) => string` | `sanitizeHtml` | Replaces the sanitizer both ways, for content the app trusts. |\n| `name` | `string` | - | Posts the value with a native form, as a hidden input. |\n\n`RichTextTool` (exported) is one of:\n`'bold' | 'italic' | 'underline' | 'strike' | 'h1' | 'h2' | 'h3' | 'p' | 'ul' | 'ol' | 'quote' | 'code' | 'alignLeft' | 'alignCenter' | 'alignRight' | 'link' | 'clear' | 'undo' | 'redo' | 'image' | 'table' | 'hr' | 'source' | '|'`.\nThe `'|'` entry draws a visual separator between tool groups. The default\ntoolbar is every tool up to `redo`; `image`, `table`, `hr` and `source` appear\nonly when listed.\n\n## Examples\n\n### A trimmed toolbar\n\nPass just the tools you need; group them with `'|'` separators:\n\n```svelte\n<SvRichText\n bind:value={note}\n tools={['bold', 'italic', '|', 'ul', 'ol', '|', 'link']}\n/>\n```\n\n### Pages and articles: images, tables, the HTML\n\nFor a CMS page or a blog post, list the extra tools, upload images to your\nserver and post the editor with the rest of the form. `sanitize` is replaced\nonly because the administrators who write these pages are trusted to use\ninline styles; keep the default for anything a visitor can write.\n\n```svelte\n<script lang=\"ts\">\n import { SvRichText } from '@svgrid/grid'\n\n async function upload(file: File) {\n const body = new FormData()\n body.append('image', file)\n const answer = await (await fetch('/api/upload', { method: 'POST', body })).json()\n return answer.url ?? null\n }\n</script>\n\n<form method=\"post\">\n <SvRichText\n name=\"content\"\n value={page.content}\n tools={['bold', 'italic', 'h2', 'h3', '|', 'ul', 'ol', '|', 'link', 'image', 'table', 'hr', '|', 'source', 'undo', 'redo']}\n onImageUpload={upload}\n sanitize={(html) => html.replace(/<script[\\s\\S]*?<\\/script>/gi, '')}\n />\n <button>Save</button>\n</form>\n```\n\n`source` swaps the body for a text area with the HTML; the other tools wait\nuntil it is switched off, and what was typed there goes through `sanitize`.\n\n### Read-only rendering\n\nSet `readonly` to show formatted content without a toolbar - useful for previews\nof the same HTML the editor produced:\n\n```svelte\n<SvRichText value={savedHtml} readonly />\n```\n\n### Reacting to edits\n\nPersist on change instead of only on blur:\n\n```svelte\n<SvRichText bind:value={html} onChange={(h) => queueSave(h)} />\n```\n\n### A field bound to a record, with preview\n\nBind `value` straight to a field on your state object so edits flow back into the\nmodel. Track a `dirty` flag from `onChange` to gate the Save button, then flip to\n`readonly` to preview the exact HTML the editor produced:\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvRichText, SvButton } from '@svgrid/grid'\n\n let ticket = $state({ id: 42, notes: '<p>Initial triage notes.</p>' })\n let editing = $state(true)\n let dirty = $state(false)\n\n function save() {\n fetch(`/api/tickets/${ticket.id}`, {\n method: 'PATCH',\n body: JSON.stringify({ notes: ticket.notes }),\n })\n dirty = false\n editing = false\n }\n</script>\n\n{#if editing}\n <SvRichText\n bind:value={ticket.notes}\n tools={['bold', 'italic', '|', 'ul', 'ol', '|', 'link', 'clear']}\n minHeight=\"200px\"\n placeholder=\"Add triage notes...\"\n onChange={() => (dirty = true)}\n />\n <SvButton variant=\"primary\" disabled={!dirty} onclick={save}>Save</SvButton>\n{:else}\n <SvRichText value={ticket.notes} readonly />\n <SvButton onclick={() => (editing = true)}>Edit</SvButton>\n{/if}\n```\n\nTip: an external change to `value` is written into the editor only while it is\nnot focused - writing `innerHTML` mid-caret would collapse the selection - so a\nvalue set from elsewhere applies once the field blurs. `readonly` mode drops the\ntoolbar and renders the same HTML, making it a zero-cost preview of what you saved.\n\n## Accessibility\n\n- The editable region is a `role=\"textbox\"` with `aria-multiline` and an\n `aria-label`.\n- The toolbar is a `role=\"toolbar\"` wired to the body via `aria-controls`.\n- Toggle tools (bold, italic, lists) expose `aria-pressed` reflecting the current\n caret state; every button has a descriptive label.\n\n## Disabled and read-only\n\nDisabled takes the control out of the tab order; read-only keeps it focusable and copyable. Reach for read-only when the value still matters to the reader.\n\n```svelte {runnable}\n<script lang=\"ts\">\n import { SvRichText } from '@svgrid/grid'\n\n let html = $state('')\n</script>\n\n<SvRichText bind:value={html} disabled />\n\n<SvRichText bind:value={html} readonly />\n```\n\n## See also\n\n- [Navigation overview](navigation.md) - the wayfinding family at a glance.\n- [Inputs overview](inputs.md) - the text and value editors.\n- [SvForm](sv-form.md) - collect rich text alongside other fields.\n"
7443
7443
  },
7444
7444
  {
7445
7445
  "slug": "help/ui-components/sv-scroll-area",
@@ -8608,7 +8608,7 @@ export const docs = [
8608
8608
  "path": "docs/_data/comparisons/ag-grid.json",
8609
8609
  "title": "SvGrid vs AG Grid (community + enterprise)",
8610
8610
  "section": "Comparisons",
8611
- "markdown": "# SvGrid vs AG Grid (community + enterprise)\n\nCanonical URL: https://svgrid.com/compare/ag-grid/\nFacts verified: 2026-09-12\nCompetitor package: ag-grid-community\nCompetitor site: https://www.ag-grid.com/\n\nAG Grid is the most widely used JavaScript data grid, sold for React, Angular, Vue and plain JavaScript, with Svelte covered only by community-built integrations.\n\n## Verdict\n\nAG Grid has the longer track record and the wider Enterprise catalogue; SvGrid is Svelte-5-native, smaller on a Svelte page, and puts most of what AG Grid sells as Enterprise into its MIT core. Pick AG Grid for parity with React and Angular teams; pick SvGrid for a Svelte-first stack.\n\nMigration guide: moving from AG Grid (community + enterprise) to SvGrid: https://svgrid.com/docs/help/migrating-from-ag-grid/\n\n## At a glance\n\n| Fact | SvGrid | AG Grid (community + enterprise) | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 36.3.0 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 4 Nov 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (JavaScript, React, Angular, Vue) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 16,400,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 346.4 KB JS, no separate stylesheet | 9 Nov 2026 |\n| Frameworks | Svelte 5 | JavaScript, React, Angular, Vue | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | AG Grid Community is free under MIT. AG Grid Enterprise is listed at $999 USD per developer with one year of updates and Zendesk support; the Enterprise Bundle with AG Charts Enterprise is $1,498 USD per developer | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; ag-grid-community 36.3.0 entry `createGrid + AllCommunityModule + themeQuartz` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nAG Grid is the heavyweight of the category: a Community edition under MIT and an Enterprise edition sold per developer that adds the set filter, range selection, master/detail, tree data, row grouping, pivoting, integrated charts, Excel export and the server-side row model. Its own pages mark each of those as Enterprise, and the sources at the bottom of this page link to them.\n\nFor Svelte there is no official package. ag-grid.com names React, Angular, Vue and JavaScript and points Svelte users to community-built integrations, so in practice you call the JavaScript API from onMount and manage the DOM yourself. SvGrid is the inverse: a Svelte 5 component with runes and snippets, and a headless core underneath it.\n\n## Looking for an AG Grid alternative for Svelte?\n\nThe usual reason to look is that AG Grid on Svelte means the JavaScript API inside onMount plus a wrapper you do not control. SvGrid gives you a component that speaks runes natively and an MIT core with no row cap and no licence key. What you give up is breadth: AG Grid has more Enterprise modules, more integrations and a far larger body of examples, and if you need one grid across three frameworks it is still the right answer.\n\n## What they have in common\n\n- A full grid with sorting, filtering, grouping, virtualization, editing and selection.\n- Documentation with a runnable example per feature.\n- TypeScript-first APIs with an imperative grid API for code outside the component.\n- An MIT core plus a paid tier with support.\n- An official MCP server so AI assistants read the real docs instead of guessing.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native: runes for state, snippets for cells, no wrapper layer and no DOM imperatives.\n- A smaller bundle on a Svelte page; the At a glance table above carries both measured numbers and the method.\n- Set-style Excel filters, cell-range selection with clipboard, master/detail, tree data, row grouping with aggregation, integrated charts, the column tool panel and the context menu are in the MIT core. AG Grid lists each of these under Enterprise. Flat server-side data (paging, infinite block scroll) is free in both; the server-side row model with lazy grouping, tree, pivot, transactions and selection across unloaded rows is a paid module in both.\n- In-grid AI helpers (natural-language filter, smart fill, summarise, classify, anomalies) ship free in @svgrid/grid with your own model provider.\n- @svgrid/mcp validates props and imports against the real API surface and scaffolds a project, on top of docs and examples.\n- @svgrid/enterprise costs $599 per developer per year for the enterprise grid, unlimited applications; AG Grid Enterprise's per-developer price is in the pricing line above.\n\n## Where AG Grid (community + enterprise) is stronger\n\n- A longer production record across very large enterprise deployments, and the largest body of third-party examples and answers of any grid.\n- The viewport row model, where the server pushes the visible window over a socket; SvGrid has no equivalent.\n- AG Charts, a separately licensed charting product with far more chart types than SvGridChart.\n- One grid API across React, Angular and Vue teams.\n- Enterprise support through Zendesk with one year of updates on the licence.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | AG Grid (community + enterprise) |\n| --- | --- | --- |\n| Svelte 5 native | Yes - runes and snippets | No - JavaScript build; Svelte is pointed to community-built integrations |\n| Sorting and filtering | Yes | Yes |\n| Excel-style set filter | Yes - MIT core | Paid tier - the Set Filter page is marked Enterprise |\n| Row and column virtualization | Yes | Yes |\n| Inline editing | Yes | Yes |\n| Cell-range selection and clipboard | Yes - MIT core, with a fill handle | Paid tier - Range Selection and Clipboard are Enterprise |\n| Master/detail | Yes - MIT core | Paid tier - Enterprise |\n| Tree data | Yes - MIT core | Paid tier - the Tree Data page is marked Enterprise |\n| Row grouping with aggregation | Yes - MIT core | Paid tier - Enterprise |\n| Pivot tables | Paid tier - @svgrid/enterprise | Paid tier - Enterprise |\n| Integrated charts | Yes - SvGridChart in the MIT core | Paid tier - Enterprise, through AG Charts |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes - MIT core: createServerDataSource, page or infinite block scroll | Yes - Infinite Row Model, Community |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise: createServerRowModel | Paid tier - Server-Side Row Model, Enterprise |\n| Global search over server-side data | Yes - filterModel.global goes to the server | No - the Quick Filter is not supported by the Infinite or Server-Side row models |\n| Viewport row model (server pushes the visible window) | No - not built; real-time deltas merge into loaded rows instead | Paid tier - Viewport Row Model, Enterprise |\n| CSV export | Yes - MIT core | Yes - Community, through the API |\n| Excel export | Paid tier - @svgrid/enterprise | Paid tier - Enterprise |\n| Column tool panel | Yes - MIT core | Paid tier - Tool Panels are Enterprise |\n| Context menu | Yes - MIT core | Paid tier - Enterprise |\n| MCP server for AI assistants | Yes - @svgrid/mcp, MIT: docs, examples, code check, scaffold, preview | Yes - ag-mcp, MIT: docs search, version detection, upgrade prompts |\n| In-grid AI helpers (natural-language filter, smart fill) | Yes - MIT core, bring your own model | No - not in the feature list |\n| Paid support | Yes - with @svgrid/enterprise, per developer per year | Yes - Enterprise licence, Zendesk |\n| Core licence | Yes - MIT | Yes - MIT (Community) |\n\n## Measured performance\n\nMeasured 28 Oct 2026 with the harness in the repository (pnpm bench:compare): 100,000 rows, fastest of 3 samples per operation, each grid in a fresh page, Chromium through Playwright, headless, on win32 10.0.26200 x64, Intel(R) Core(TM) Ultra 7 255U; a developer workstation, not a dedicated bench rig. Versions: @svgrid/grid 3.0.25, ag-grid-community 36.1.0. Lower is better. Tick: 1,000 of the rows replaced per tick with a new amount, the grid sorted by amount, on the update path each grid provides, timed until the changed cells paint; 180 ticks after 10 to warm up.\n\n| Operation | SvGrid | AG Grid (community + enterprise) |\n| --- | --- | --- |\n| Mount | 30.7 ms | 67 ms |\n| Sort, text column | 77.4 ms | 213.9 ms |\n| Sort, numeric column | 32.8 ms | 216.7 ms |\n| Filter, one column (indicative only) | 34.5 ms | 148.8 ms |\n| Scroll, p95 frame | 34.3 ms | 34.5 ms |\n| Live tick, 1,000 rows replaced, sorted by that column, p95 | 48.6 ms | 259.2 ms |\n| Ticks over one 60 Hz frame, of 180 | 115 | 180 |\n| Rows kept in the DOM (virtualization on) | 36 | 36 |\n\nThe filter row is not like for like: each grid is driven through its own single-column filter API, and the amount of work differs. The comparison guide explains the method, what is left out and why.\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want a grid shaped by Svelte idioms rather than a wrapper around a JavaScript build.\n- You want the set filter, master/detail, tree data, range selection, grouping and flat server-side infinite scroll without an Enterprise licence.\n- You want a paid support tier priced per developer per year at $599.\n- You want an assistant that can check the code it writes against the real API, not only read the docs.\n\n## When to choose AG Grid (community + enterprise)\n\n- You depend on a specific AG Grid Enterprise module, the viewport row model or server-side pivoting.\n- You run the same grid across React, Angular and Vue teams and want one library.\n- Your procurement already covers AG Grid Enterprise, or you need AG Charts.\n- You want the largest possible pool of existing examples, answers and contractors.\n\n## Migration effort\n\nEstimated effort: 1-3 hours per grid.\n\nWhat you give up:\n\n- AG Charts as a separately licensed charting product with far more chart types than SvGridChart.\n- The viewport row model; SvGrid's server-side row model covers sort, filter, grouping, tree, pivot, transactions and select-all across unloaded rows, but the server cannot push the visible window.\n- One grid API across React, Angular and Vue teams.\n- The largest body of third-party examples, Stack Overflow answers and contractors of any grid.\n\nMigration guide: moving from AG Grid (community + enterprise) to SvGrid: https://svgrid.com/docs/help/migrating-from-ag-grid/\n\n## Live example\n\nMigration From Ag Grid: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/139-migration-from-ag-grid/\n\n## The bottom line\n\nAG Grid is the safer pick for a multi-framework organisation, or when you need its deepest Enterprise modules, AG Charts or the viewport row model. For a Svelte-first team, SvGrid covers the Enterprise surface most teams actually buy AG Grid for, in the MIT core: master/detail, tree data, range selection, grouping, flat server-side infinite scroll and integrated charts, with pivot tables, the server-side row model and Excel, PDF and print output in the @svgrid/enterprise pack. The bundle and benchmark numbers on this page are measured on the dates shown, so read them rather than either vendor's adjectives.\n\n## Frequently asked questions\n\n### Does AG Grid support Svelte?\n\nNot with an official package. ag-grid.com names React, Angular, Vue and JavaScript, and points Svelte, Solid and Python users to community-built integrations. In practice you create the grid from the JavaScript API inside onMount and keep its state in sync with your Svelte state by hand.\n\n### Which AG Grid Enterprise features does SvGrid include in its MIT core?\n\nThe set-style Excel filter, cell-range selection with clipboard, master/detail, tree data, row grouping with aggregation, integrated charts, the column tool panel and the context menu. AG Grid's own pages list each of these under Enterprise. Pivot tables, the server-side row model (grouping, tree, pivot, transactions, selection) and Excel, PDF and print output are in @svgrid/enterprise, as they are in AG Grid Enterprise; flat server-side infinite scroll is free in both.\n\n### What does SvGrid not have that AG Grid Enterprise has?\n\nThe viewport row model, AG Charts as a full charting product, and the breadth of niche Enterprise modules that come from a decade of enterprise customers. SvGrid's server-side row model handles sort, filter, grouping, tree, pivot, transactions and selection across unloaded rows; if your requirement is the viewport model, AG Grid Enterprise is the better fit.\n\n### Does AG Grid have an MCP server too?\n\nYes. ag-mcp (MIT) gives an assistant AG Grid's docs, examples and version-specific upgrade prompts. @svgrid/mcp covers the same ground for SvGrid and adds a code checker that validates props and imports against the real API surface, a scaffold tool and a preview. Neither is an in-grid feature; SvGrid's natural-language filter, smart fill and summarise helpers are, and they ship free in @svgrid/grid with your own model provider.\n\n### How much does AG Grid Enterprise cost compared with SvGrid?\n\nAG Grid Enterprise is sold per developer with one year of updates; the price as ag-grid.com listed it on the date we read it is in the pricing line at the top of this page. SvGrid's core is MIT and @svgrid/enterprise is $599 per developer per year for the enterprise grid or $999 with the spreadsheet and Studio, both for unlimited applications. Compare what each paid tier adds before comparing prices: most of what AG Grid sells as Enterprise is in SvGrid's free core.\n\n### How long does a migration from AG Grid to SvGrid take?\n\nThe migration guide budgets one to three hours per grid. Column definitions map almost one to one, the imperative API has equivalents on SvGridApi, and most of the time goes into moving cell renderers to snippets. The guide also lists what does not map, so you can decide before you start.\n\n### Is SvGrid a good AG Grid alternative for Svelte?\n\nThe usual reason to look is that AG Grid on Svelte means the JavaScript API inside onMount plus a wrapper you do not control. SvGrid gives you a component that speaks runes natively and an MIT core with no row cap and no licence key. What you give up is breadth: AG Grid has more Enterprise modules, more integrations and a far larger body of examples, and if you need one grid across three frameworks it is still the right answer.\n\n## Related comparisons and docs\n\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/)\n- [SvGrid vs Handsontable](https://svgrid.com/compare/handsontable/)\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Performance benchmarks](https://svgrid.com/docs/help/benchmarks/)\n- [Server-side data](https://svgrid.com/docs/help/server-side-data/)\n- [MCP server](https://svgrid.com/docs/help/mcp-server/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- ag-pricing: Enterprise is per developer at $999 USD with one year of updates and Zendesk support; the Enterprise Bundle is $1,498 USD; Community is free under MIT (checked 12 Sep 2026) https://www.ag-grid.com/license-pricing/\n- ag-community-vs-enterprise: Lists Range Selection, Clipboard, Master/Detail, Row Grouping, Pivot Tables and Aggregations, Integrated Charts, Excel export, Server-Side Row Model, Tool Panels and Context Menu under Enterprise Features; no AI feature in either list (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/community-vs-enterprise/\n- ag-set-filter: The Set Filter page is marked enterprise (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/filter-set/\n- ag-tree-data: The Tree Data page is marked enterprise (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/tree-data/\n- ag-csv-export: CSV export is available in Community through the API; the context-menu export and Excel export are Enterprise (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/csv-export/\n- ag-mcp: Official MCP server installed with npx ag-mcp: search_docs, detect_version, set_version and list_versions tools, quick-start and upgrade-grid prompts, docs and examples as resources; described as help for integrating and maintaining grid code, not an in-grid feature. The ag-mcp package on npm is MIT (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/mcp-server/\n- ag-home: Names React, Angular, Vue and JavaScript; says Svelte, Solid and Python users can use community-built integrations; the 36.1.0 release notes list an AI Coding Skill (checked 12 Sep 2026) https://www.ag-grid.com/\n- ag-row-models: The row-models feature table lists the Infinite Row Model as Community and the Server-Side and Viewport row models as Enterprise; the Quick Filter row is marked unsupported for the Infinite, Server-Side and Viewport models (checked 15 Sep 2026) https://www.ag-grid.com/javascript-data-grid/row-models/\n"
8611
+ "markdown": "# SvGrid vs AG Grid (community + enterprise)\n\nCanonical URL: https://svgrid.com/compare/ag-grid/\nFacts verified: 2026-09-12\nCompetitor package: ag-grid-community\nCompetitor site: https://www.ag-grid.com/\n\nAG Grid is the most widely used JavaScript data grid, sold for React, Angular, Vue and plain JavaScript, with Svelte covered only by community-built integrations.\n\n## Verdict\n\nAG Grid has the longer track record and the wider Enterprise catalogue; SvGrid is Svelte-5-native, smaller on a Svelte page, and puts most of what AG Grid sells as Enterprise into its MIT core. Pick AG Grid for parity with React and Angular teams; pick SvGrid for a Svelte-first stack.\n\nMigration guide: moving from AG Grid (community + enterprise) to SvGrid: https://svgrid.com/docs/help/migrating-from-ag-grid/\n\n## At a glance\n\n| Fact | SvGrid | AG Grid (community + enterprise) | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 36.3.0 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 4 Nov 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (JavaScript, React, Angular, Vue) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 16,400,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 346.4 KB JS, no separate stylesheet | 9 Nov 2026 |\n| Frameworks | Svelte 5 | JavaScript, React, Angular, Vue | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | AG Grid Community is free under MIT. AG Grid Enterprise is listed at $999 USD per developer with one year of updates and Zendesk support; the Enterprise Bundle with AG Charts Enterprise is $1,498 USD per developer | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; ag-grid-community 36.3.0 entry `createGrid + AllCommunityModule + themeQuartz` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nAG Grid is the heavyweight of the category: a Community edition under MIT and an Enterprise edition sold per developer that adds the set filter, range selection, master/detail, tree data, row grouping, pivoting, integrated charts, Excel export and the server-side row model. Its own pages mark each of those as Enterprise, and the sources at the bottom of this page link to them.\n\nFor Svelte there is no official package. ag-grid.com names React, Angular, Vue and JavaScript and points Svelte users to community-built integrations, so in practice you call the JavaScript API from onMount and manage the DOM yourself. SvGrid is the inverse: a Svelte 5 component with runes and snippets, and a headless core underneath it.\n\n## Looking for an AG Grid alternative for Svelte?\n\nThe usual reason to look is that AG Grid on Svelte means the JavaScript API inside onMount plus a wrapper you do not control. SvGrid gives you a component that speaks runes natively and an MIT core with no row cap and no licence key. What you give up is breadth: AG Grid has more Enterprise modules, more integrations and a far larger body of examples, and if you need one grid across three frameworks it is still the right answer.\n\n## What they have in common\n\n- A full grid with sorting, filtering, grouping, virtualization, editing and selection.\n- Documentation with a runnable example per feature.\n- TypeScript-first APIs with an imperative grid API for code outside the component.\n- An MIT core plus a paid tier with support.\n- An official MCP server so AI assistants read the real docs instead of guessing.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native: runes for state, snippets for cells, no wrapper layer and no DOM imperatives.\n- A smaller bundle on a Svelte page; the At a glance table above carries both measured numbers and the method.\n- Set-style Excel filters, cell-range selection with clipboard, master/detail, tree data, row grouping with aggregation, integrated charts, the column tool panel and the context menu are in the MIT core. AG Grid lists each of these under Enterprise. Flat server-side data (paging, infinite block scroll) is free in both; the server-side row model with lazy grouping, tree, pivot, transactions and selection across unloaded rows is a paid module in both.\n- In-grid AI helpers (natural-language filter, smart fill, summarise, classify, anomalies) ship free in @svgrid/grid with your own model provider.\n- @svgrid/mcp validates props and imports against the real API surface and scaffolds a project, on top of docs and examples.\n- @svgrid/enterprise costs $599 per developer per year for the enterprise grid, unlimited applications; AG Grid Enterprise's per-developer price is in the pricing line above.\n\n## Where AG Grid (community + enterprise) is stronger\n\n- A longer production record across very large enterprise deployments, and the largest body of third-party examples and answers of any grid.\n- The viewport row model, where the server pushes the visible window over a socket; SvGrid has no equivalent.\n- AG Charts, a separately licensed charting product with far more chart types than SvGridChart.\n- One grid API across React, Angular and Vue teams.\n- Enterprise support through Zendesk with one year of updates on the licence.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | AG Grid (community + enterprise) |\n| --- | --- | --- |\n| Svelte 5 native | Yes - runes and snippets | No - JavaScript build; Svelte is pointed to community-built integrations |\n| Sorting and filtering | Yes | Yes |\n| Excel-style set filter | Yes - MIT core | Paid tier - the Set Filter page is marked Enterprise |\n| Row and column virtualization | Yes | Yes |\n| Inline editing | Yes | Yes |\n| Cell-range selection and clipboard | Yes - MIT core, with a fill handle | Paid tier - Range Selection and Clipboard are Enterprise |\n| Master/detail | Yes - MIT core | Paid tier - Enterprise |\n| Tree data | Yes - MIT core | Paid tier - the Tree Data page is marked Enterprise |\n| Row grouping with aggregation | Yes - MIT core | Paid tier - Enterprise |\n| Pivot tables | Paid tier - @svgrid/enterprise | Paid tier - Enterprise |\n| Integrated charts | Yes - SvGridChart in the MIT core | Paid tier - Enterprise, through AG Charts |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes - MIT core: createServerDataSource, page or infinite block scroll | Yes - Infinite Row Model, Community |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise: createServerRowModel | Paid tier - Server-Side Row Model, Enterprise |\n| Global search over server-side data | Yes - filterModel.global goes to the server | No - the Quick Filter is not supported by the Infinite or Server-Side row models |\n| Viewport row model (server pushes the visible window) | No - not built; real-time deltas merge into loaded rows instead | Paid tier - Viewport Row Model, Enterprise |\n| CSV export | Yes - MIT core | Yes - Community, through the API |\n| Excel export | Paid tier - @svgrid/enterprise | Paid tier - Enterprise |\n| Column tool panel | Yes - MIT core | Paid tier - Tool Panels are Enterprise |\n| Context menu | Yes - MIT core | Paid tier - Enterprise |\n| MCP server for AI assistants | Yes - @svgrid/mcp, MIT: docs, examples, code check, scaffold, preview | Yes - ag-mcp, MIT: docs search, version detection, upgrade prompts |\n| In-grid AI helpers (natural-language filter, smart fill) | Yes - MIT core, bring your own model | No - not in the feature list |\n| Paid support | Yes - with @svgrid/enterprise, per developer per year | Yes - Enterprise licence, Zendesk |\n| Core licence | Yes - MIT | Yes - MIT (Community) |\n\n## Measured performance\n\nMeasured 28 Oct 2026 with the harness in the repository (pnpm bench:compare): 100,000 rows, fastest of 3 samples per operation, each grid in a fresh page, Chromium through Playwright, headless, on win32 10.0.26200 x64, Intel(R) Core(TM) Ultra 7 255U; a developer workstation, not a dedicated bench rig. Versions: @svgrid/grid 3.0.25, ag-grid-community 36.1.0. Lower is better. Tick: 1,000 of the rows replaced per tick with a new amount, the grid sorted by amount, on the update path each grid provides, timed until the changed cells paint; 180 ticks after 10 to warm up.\n\n| Operation | SvGrid | AG Grid (community + enterprise) |\n| --- | --- | --- |\n| Mount | 30.7 ms | 67 ms |\n| Sort, text column | 77.4 ms | 213.9 ms |\n| Sort, numeric column | 32.8 ms | 216.7 ms |\n| Filter, one column (indicative only) | 34.5 ms | 148.8 ms |\n| Scroll, p95 frame | 34.3 ms | 34.5 ms |\n| Live tick, 1,000 rows replaced, sorted by that column, p95 | 48.6 ms | 259.2 ms |\n| Ticks over one 60 Hz frame, of 180 | 115 | 180 |\n| Rows kept in the DOM (virtualization on) | 36 | 36 |\n\nThe filter row is not like for like: each grid is driven through its own single-column filter API, and the amount of work differs. The comparison guide explains the method, what is left out and why.\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want a grid shaped by Svelte idioms rather than a wrapper around a JavaScript build.\n- You want the set filter, master/detail, tree data, range selection, grouping and flat server-side infinite scroll without an Enterprise licence.\n- You want a paid support tier priced per developer per year at $599.\n- You want an assistant that can check the code it writes against the real API, not only read the docs.\n\n## When to choose AG Grid (community + enterprise)\n\n- You depend on a specific AG Grid Enterprise module, the viewport row model or server-side pivoting.\n- You run the same grid across React, Angular and Vue teams and want one library.\n- Your procurement already covers AG Grid Enterprise, or you need AG Charts.\n- You want the largest possible pool of existing examples, answers and contractors.\n\n## Migration effort\n\nEstimated effort: 1-3 hours per grid.\n\nWhat you give up:\n\n- AG Charts as a separately licensed charting product with far more chart types than SvGridChart.\n- The viewport row model; SvGrid's server-side row model covers sort, filter, grouping, tree, pivot, transactions and select-all across unloaded rows, but the server cannot push the visible window.\n- One grid API across React, Angular and Vue teams.\n- The largest body of third-party examples, Stack Overflow answers and contractors of any grid.\n\nMigration guide: moving from AG Grid (community + enterprise) to SvGrid: https://svgrid.com/docs/help/migrating-from-ag-grid/\n\n## Live example\n\nMigration From Ag Grid: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/139-migration-from-ag-grid/\n\n## The bottom line\n\nAG Grid is the safer pick for a multi-framework organisation, or when you need its deepest Enterprise modules, AG Charts or the viewport row model. For a Svelte-first team, SvGrid covers the Enterprise surface most teams actually buy AG Grid for, in the MIT core: master/detail, tree data, range selection, grouping, flat server-side infinite scroll and integrated charts, with pivot tables, the server-side row model and Excel, PDF and print output in the @svgrid/enterprise pack. The bundle and benchmark numbers on this page are measured on the dates shown, so read them rather than either vendor's adjectives.\n\n## Frequently asked questions\n\n### Does AG Grid support Svelte?\n\nNot with an official package. ag-grid.com names React, Angular, Vue and JavaScript, and points Svelte, Solid and Python users to community-built integrations. In practice you create the grid from the JavaScript API inside onMount and keep its state in sync with your Svelte state by hand.\n\n### Which AG Grid Enterprise features does SvGrid include in its MIT core?\n\nThe set-style Excel filter, cell-range selection with clipboard, master/detail, tree data, row grouping with aggregation, integrated charts, the column tool panel and the context menu. AG Grid's own pages list each of these under Enterprise. Pivot tables, the server-side row model (grouping, tree, pivot, transactions, selection) and Excel, PDF and print output are in @svgrid/enterprise, as they are in AG Grid Enterprise; flat server-side infinite scroll is free in both.\n\n### What does SvGrid not have that AG Grid Enterprise has?\n\nThe viewport row model, AG Charts as a full charting product, and the breadth of niche Enterprise modules that come from a decade of enterprise customers. SvGrid's server-side row model handles sort, filter, grouping, tree, pivot, transactions and selection across unloaded rows; if your requirement is the viewport model, AG Grid Enterprise is the better fit.\n\n### Does AG Grid have an MCP server too?\n\nYes. ag-mcp (MIT) gives an assistant AG Grid's docs, examples and version-specific upgrade prompts. @svgrid/mcp covers the same ground for SvGrid and adds a code checker that validates props and imports against the real API surface, a scaffold tool and a preview. Neither is an in-grid feature; SvGrid's natural-language filter, smart fill and summarise helpers are, and they ship free in @svgrid/grid with your own model provider.\n\n### How much does AG Grid Enterprise cost compared with SvGrid?\n\nAG Grid Enterprise is sold per developer with one year of updates; the price as ag-grid.com listed it on the date we read it is in the pricing line at the top of this page. SvGrid's core is MIT and @svgrid/enterprise is $599 per developer per year for the enterprise grid or $999 with the spreadsheet and Studio, both for unlimited applications. Compare what each paid tier adds before comparing prices: most of what AG Grid sells as Enterprise is in SvGrid's free core.\n\n### How long does a migration from AG Grid to SvGrid take?\n\nThe migration guide budgets one to three hours per grid. Column definitions map almost one to one, the imperative API has equivalents on SvGridApi, and most of the time goes into moving cell renderers to snippets. The guide also lists what does not map, so you can decide before you start.\n\n### Is SvGrid a good AG Grid alternative for Svelte?\n\nThe usual reason to look is that AG Grid on Svelte means the JavaScript API inside onMount plus a wrapper you do not control. SvGrid gives you a component that speaks runes natively and an MIT core with no row cap and no licence key. What you give up is breadth: AG Grid has more Enterprise modules, more integrations and a far larger body of examples, and if you need one grid across three frameworks it is still the right answer.\n\n## Related comparisons and docs\n\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/)\n- [SvGrid vs Handsontable](https://svgrid.com/compare/handsontable/)\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Performance benchmarks](https://svgrid.com/docs/help/benchmarks/)\n- [Server-side data](https://svgrid.com/docs/help/server-side-data/)\n- [MCP server](https://svgrid.com/docs/help/mcp-server/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- ag-pricing: Enterprise is per developer at $999 USD with one year of updates and Zendesk support; the Enterprise Bundle is $1,498 USD; Community is free under MIT (checked 12 Sep 2026) https://www.ag-grid.com/license-pricing/\n- ag-community-vs-enterprise: Lists Range Selection, Clipboard, Master/Detail, Row Grouping, Pivot Tables and Aggregations, Integrated Charts, Excel export, Server-Side Row Model, Tool Panels and Context Menu under Enterprise Features; no AI feature in either list (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/community-vs-enterprise/\n- ag-set-filter: The Set Filter page is marked enterprise (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/filter-set/\n- ag-tree-data: The Tree Data page is marked enterprise (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/tree-data/\n- ag-csv-export: CSV export is available in Community through the API; the context-menu export and Excel export are Enterprise (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/csv-export/\n- ag-mcp: Official MCP server installed with npx ag-mcp: search_docs, detect_version, set_version and list_versions tools, quick-start and upgrade-grid prompts, docs and examples as resources; described as help for integrating and maintaining grid code, not an in-grid feature. The ag-mcp package on npm is MIT (checked 12 Sep 2026) https://www.ag-grid.com/javascript-data-grid/mcp-server/\n- ag-home: Names React, Angular, Vue and JavaScript; says Svelte, Solid and Python users can use community-built integrations; the 36.1.0 release notes list an AI Coding Skill (checked 12 Sep 2026) https://www.ag-grid.com/\n- ag-row-models: The row-models feature table lists the Infinite Row Model as Community and the Server-Side and Viewport row models as Enterprise; the Quick Filter row is marked unsupported for the Infinite, Server-Side and Viewport models (checked 15 Sep 2026) https://www.ag-grid.com/javascript-data-grid/row-models/\n"
8612
8612
  },
8613
8613
  {
8614
8614
  "slug": "compare/airtable",
@@ -8636,14 +8636,14 @@ export const docs = [
8636
8636
  "path": "docs/_data/comparisons/glide-data-grid.json",
8637
8637
  "title": "SvGrid vs Glide Data Grid",
8638
8638
  "section": "Comparisons",
8639
- "markdown": "# SvGrid vs Glide Data Grid\n\nCanonical URL: https://svgrid.com/compare/glide-data-grid/\nFacts verified: 2026-09-12\nCompetitor package: @glideapps/glide-data-grid\nCompetitor site: https://grid.glideapps.com/\n\nGlide Data Grid is a canvas-rendered React data grid built for extreme scroll performance on very large datasets.\n\n## Verdict\n\nGlide wins on raw canvas FPS for hundreds of thousands of visible cells. SvGrid renders accessible, selectable DOM with virtualization and is Svelte-5-native - the right call for most apps.\n\nMigration guide: moving from Glide Data Grid to SvGrid: https://svgrid.com/docs/help/migrating-from-glide/\n\n## At a glance\n\n| Fact | SvGrid | Glide Data Grid | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 6.0.3 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 3 Feb 2024 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (React) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 2,010,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 71.1 KB JS + 0.2 KB CSS, lodash + marked + react + react-dom + react-responsive-carousel external | 9 Nov 2026 |\n| Frameworks | Svelte 5 | React | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | Glide Data Grid is MIT and, in its own words, fully free and open source; no licence is sold for the grid | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @glideapps/glide-data-grid 6.0.3 entry `DataEditor component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nGlide Data Grid renders to a <canvas> instead of DOM, which gives it exceptional performance on very wide, very tall datasets at the cost of standard HTML semantics. It is React-only and MIT-licensed.\n\nSvGrid renders virtualized DOM: real table semantics, WAI-ARIA roles, keyboard navigation, and selectable / indexable text. For the vast majority of apps the performance gap is invisible, and you keep accessibility and SEO-friendly markup.\n\n## Looking for a Glide Data Grid alternative for Svelte?\n\nGlide renders to canvas, which is why it is fast and also why text selection, accessibility and custom cells work differently there. SvGrid renders real DOM with virtualization, so a screen reader and browser find behave normally and a cell can be any Svelte snippet. If your bar is a million-row canvas scroll inside React, Glide is still the specialist.\n\n## What they have in common\n\n- MIT-licensed.\n- Virtualized rendering for large datasets.\n- Cell editing and custom cell content.\n- TypeScript-first.\n\n## Where SvGrid is stronger\n\n- Renders real DOM - WAI-ARIA, keyboard nav, screen-reader support, selectable text.\n- Svelte 5 native rather than React-only.\n- Excel-style filter menu, headless engine, and the imperative API out of the box.\n- @svgrid/mcp MCP server for AI assistants.\n\n## Where Glide Data Grid is stronger\n\n- Peak canvas rendering performance on hundreds of thousands of visible cells.\n- Per-pixel custom cell drawing.\n- Proven inside Glide's own product at scale.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | Glide Data Grid |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - React only |\n| Rendering | Yes - Accessible DOM | Partial - canvas |\n| WAI-ARIA + keyboard | Yes | Partial - Limited (canvas) |\n| Selectable / indexable text | Yes | No - Canvas pixels |\n| Peak FPS on 200k+ visible cells | Partial - Good | Yes - Best |\n| Virtualization | Yes | Yes |\n| Excel-style filters | Yes | No - BYO |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You need accessibility, keyboard support, or selectable text.\n- You are on Svelte 5 and want a native grid.\n- You want filters, editing, and an imperative API without building them.\n\n## When to choose Glide Data Grid\n\n- You have measured a need for peak canvas FPS on 200k+ simultaneously visible cells.\n- You require per-pixel custom cell drawing.\n- Your app is React and accessibility is not a requirement.\n\n## Migration effort\n\nEstimated effort: 3-6 hours per grid.\n\nWhat you give up:\n\n- Canvas rendering throughput on very wide, very dense sheets.\n- The React ecosystem around it, including Glide Data Editor.\n\nMigration guide: moving from Glide Data Grid to SvGrid: https://svgrid.com/docs/help/migrating-from-glide/\n\n## Live example\n\nHeadless Virtual: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/187-headless-virtual/\n\n## The bottom line\n\nGlide is the right tool when you have measured a need for canvas-level FPS on hundreds of thousands of visible cells. For everyone else - and especially when accessibility, selectable text, or Svelte matter - the virtualized DOM in SvGrid is the better default.\n\n## Frequently asked questions\n\n### SvGrid vs Glide Data Grid - which should I use?\n\nUse SvGrid unless you have a measured need for canvas rendering at extreme scale. SvGrid renders accessible DOM and is Svelte-native; Glide is React-only and canvas-based.\n\n### Is Glide faster than SvGrid?\n\nAt hundreds of thousands of simultaneously visible cells, canvas rendering gives Glide higher peak FPS. For typical datasets the difference is invisible, and SvGrid virtualizes comfortably to 100k+ rows.\n\n### Does SvGrid handle accessibility better than Glide?\n\nYes. SvGrid renders real DOM with WAI-ARIA roles, keyboard navigation, and selectable text; a canvas grid like Glide cannot offer those natively.\n\n### Is SvGrid a good Glide Data Grid alternative for Svelte?\n\nGlide renders to canvas, which is why it is fast and also why text selection, accessibility and custom cells work differently there. SvGrid renders real DOM with virtualization, so a screen reader and browser find behave normally and a cell can be any Svelte snippet. If your bar is a million-row canvas scroll inside React, Glide is still the specialist.\n\n## Related comparisons and docs\n\n- [SvGrid vs React Data Grid (adazzle)](https://svgrid.com/compare/react-data-grid/)\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/)\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Headless virtualization](https://svgrid.com/docs/help/headless/virtualization/)\n- [Accessibility](https://svgrid.com/docs/help/accessibility/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- glide-site: MIT, fully free and open source; a React data grid that scales to millions of rows with lazy cell rendering and native scrolling (checked 12 Sep 2026) https://grid.glideapps.com/\n- glide-readme: README explains why the grid renders on an HTML canvas; supports React 16 and later (checked 12 Sep 2026) https://github.com/glideapps/glide-data-grid\n"
8639
+ "markdown": "# SvGrid vs Glide Data Grid\n\nCanonical URL: https://svgrid.com/compare/glide-data-grid/\nFacts verified: 2026-09-12\nCompetitor package: @glideapps/glide-data-grid\nCompetitor site: https://grid.glideapps.com/\n\nGlide Data Grid is a canvas-rendered React data grid built for extreme scroll performance on very large datasets.\n\n## Verdict\n\nGlide wins on raw canvas FPS for hundreds of thousands of visible cells. SvGrid renders accessible, selectable DOM with virtualization and is Svelte-5-native - the right call for most apps.\n\nMigration guide: moving from Glide Data Grid to SvGrid: https://svgrid.com/docs/help/migrating-from-glide/\n\n## At a glance\n\n| Fact | SvGrid | Glide Data Grid | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 6.0.3 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 3 Feb 2024 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (React) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 2,010,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 71.1 KB JS + 0.2 KB CSS, lodash + marked + react + react-dom + react-responsive-carousel external | 9 Nov 2026 |\n| Frameworks | Svelte 5 | React | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | Glide Data Grid is MIT and, in its own words, fully free and open source; no licence is sold for the grid | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @glideapps/glide-data-grid 6.0.3 entry `DataEditor component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nGlide Data Grid renders to a <canvas> instead of DOM, which gives it exceptional performance on very wide, very tall datasets at the cost of standard HTML semantics. It is React-only and MIT-licensed.\n\nSvGrid renders virtualized DOM: real table semantics, WAI-ARIA roles, keyboard navigation, and selectable / indexable text. For the vast majority of apps the performance gap is invisible, and you keep accessibility and SEO-friendly markup.\n\n## Looking for a Glide Data Grid alternative for Svelte?\n\nGlide renders to canvas, which is why it is fast and also why text selection, accessibility and custom cells work differently there. SvGrid renders real DOM with virtualization, so a screen reader and browser find behave normally and a cell can be any Svelte snippet. If your bar is a million-row canvas scroll inside React, Glide is still the specialist.\n\n## What they have in common\n\n- MIT-licensed.\n- Virtualized rendering for large datasets.\n- Cell editing and custom cell content.\n- TypeScript-first.\n\n## Where SvGrid is stronger\n\n- Renders real DOM - WAI-ARIA, keyboard nav, screen-reader support, selectable text.\n- Svelte 5 native rather than React-only.\n- Excel-style filter menu, headless engine, and the imperative API out of the box.\n- @svgrid/mcp MCP server for AI assistants.\n\n## Where Glide Data Grid is stronger\n\n- Peak canvas rendering performance on hundreds of thousands of visible cells.\n- Per-pixel custom cell drawing.\n- Proven inside Glide's own product at scale.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | Glide Data Grid |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - React only |\n| Rendering | Yes - Accessible DOM | Partial - canvas |\n| WAI-ARIA + keyboard | Yes | Partial - Limited (canvas) |\n| Selectable / indexable text | Yes | No - Canvas pixels |\n| Peak FPS on 200k+ visible cells | Partial - Good | Yes - Best |\n| Virtualization | Yes | Yes |\n| Excel-style filters | Yes | No - BYO |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You need accessibility, keyboard support, or selectable text.\n- You are on Svelte 5 and want a native grid.\n- You want filters, editing, and an imperative API without building them.\n\n## When to choose Glide Data Grid\n\n- You have measured a need for peak canvas FPS on 200k+ simultaneously visible cells.\n- You require per-pixel custom cell drawing.\n- Your app is React and accessibility is not a requirement.\n\n## Migration effort\n\nEstimated effort: 3-6 hours per grid.\n\nWhat you give up:\n\n- Canvas rendering throughput on very wide, very dense sheets.\n- The React ecosystem around it, including Glide Data Editor.\n\nMigration guide: moving from Glide Data Grid to SvGrid: https://svgrid.com/docs/help/migrating-from-glide/\n\n## Live example\n\nHeadless Virtual: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/187-headless-virtual/\n\n## The bottom line\n\nGlide is the right tool when you have measured a need for canvas-level FPS on hundreds of thousands of visible cells. For everyone else - and especially when accessibility, selectable text, or Svelte matter - the virtualized DOM in SvGrid is the better default.\n\n## Frequently asked questions\n\n### SvGrid vs Glide Data Grid - which should I use?\n\nUse SvGrid unless you have a measured need for canvas rendering at extreme scale. SvGrid renders accessible DOM and is Svelte-native; Glide is React-only and canvas-based.\n\n### Is Glide faster than SvGrid?\n\nAt hundreds of thousands of simultaneously visible cells, canvas rendering gives Glide higher peak FPS. For typical datasets the difference is invisible, and SvGrid virtualizes comfortably to 100k+ rows.\n\n### Does SvGrid handle accessibility better than Glide?\n\nYes. SvGrid renders real DOM with WAI-ARIA roles, keyboard navigation, and selectable text; a canvas grid like Glide cannot offer those natively.\n\n### Is SvGrid a good Glide Data Grid alternative for Svelte?\n\nGlide renders to canvas, which is why it is fast and also why text selection, accessibility and custom cells work differently there. SvGrid renders real DOM with virtualization, so a screen reader and browser find behave normally and a cell can be any Svelte snippet. If your bar is a million-row canvas scroll inside React, Glide is still the specialist.\n\n## Related comparisons and docs\n\n- [SvGrid vs React Data Grid (adazzle)](https://svgrid.com/compare/react-data-grid/)\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/)\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Headless virtualization](https://svgrid.com/docs/help/headless/virtualization/)\n- [Accessibility](https://svgrid.com/docs/help/accessibility/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- glide-site: MIT, fully free and open source; a React data grid that scales to millions of rows with lazy cell rendering and native scrolling (checked 12 Sep 2026) https://grid.glideapps.com/\n- glide-readme: README explains why the grid renders on an HTML canvas; supports React 16 and later (checked 12 Sep 2026) https://github.com/glideapps/glide-data-grid\n"
8640
8640
  },
8641
8641
  {
8642
8642
  "slug": "compare/gridjs",
8643
8643
  "path": "docs/_data/comparisons/gridjs.json",
8644
8644
  "title": "SvGrid vs Grid.js",
8645
8645
  "section": "Comparisons",
8646
- "markdown": "# SvGrid vs Grid.js\n\nCanonical URL: https://svgrid.com/compare/gridjs/\nFacts verified: 2026-09-12\nCompetitor package: gridjs\nCompetitor site: https://gridjs.io/\n\nGrid.js is a small, elegant, framework-agnostic table library focused on the essentials, MIT-licensed.\n\n## Verdict\n\nGrid.js is a lovely lightweight pick for search / sort / paginate. SvGrid is the fuller, Svelte-native grid when you need virtualization, editing, grouping, or large datasets.\n\nMigration guide: moving from Grid.js to SvGrid: https://svgrid.com/docs/help/migrating-from-gridjs/\n\n## At a glance\n\n| Fact | SvGrid | Grid.js | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 6.2.0 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 3 Mar 2024 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (JavaScript) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 393,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 17.8 KB JS + 2.4 KB CSS | 9 Nov 2026 |\n| Frameworks | Svelte 5 | JavaScript | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | Grid.js is MIT, free and open source; no licence is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; gridjs 6.2.0 entry `Grid` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nGrid.js keeps a deliberately small surface: search, sort, pagination, and basic server-side data, rendered through a tiny vanilla-JS core with framework wrappers. It is MIT-licensed and easy to learn.\n\nSvGrid trades that minimalism for depth: a Svelte-5-native render component with virtualization, an Excel-style filter menu, inline editing, grouping, tree data, and an imperative API.\n\n## Looking for a Grid.js alternative for Svelte?\n\nGrid.js is deliberately small and covers the essentials well. People outgrow it at the same points every time: virtualization for large datasets, inline editing, row grouping and server-side data. SvGrid has those, and keeps a headless core so a simple table does not have to carry the weight of the full component.\n\n## What they have in common\n\n- MIT-licensed.\n- Search, sort, and pagination out of the box.\n- Server-side data support.\n- Small, approachable API.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native instead of a vanilla core + wrapper.\n- Row + column virtualization for large datasets.\n- Excel-style filters, inline editing, grouping, tree, master/detail.\n- Headless engine + imperative API + @svgrid/mcp.\n\n## Where Grid.js is stronger\n\n- Tiny footprint if all you need is search / sort / paginate.\n- Framework-agnostic core.\n- Very fast to learn for a simple table.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | Grid.js |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - vanilla JavaScript with React, Angular and Vue wrappers |\n| Search / sort / paginate | Yes | Yes |\n| Excel-style filters | Yes | No |\n| Inline editing | Yes | No |\n| Virtualization | Yes | No |\n| Grouping / tree / master-detail | Yes | No |\n| Footprint for a simple table | Partial - Larger | Yes - Tiny |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You need virtualization, editing, grouping, or tree data.\n- You are on Svelte 5 and want a native component.\n- You expect the table to grow beyond search / sort / paginate.\n\n## When to choose Grid.js\n\n- You only need search / sort / paginate over a modest dataset.\n- You want the smallest possible footprint.\n- You want a framework-agnostic drop-in.\n\n## Migration effort\n\nEstimated effort: 30 minutes to 1 hour per table.\n\nWhat you give up:\n\n- A very small library for a simple table on any page.\n- Wrappers for React, Angular and Vue.\n\nMigration guide: moving from Grid.js to SvGrid: https://svgrid.com/docs/help/migrating-from-gridjs/\n\n## Live example\n\nExcel Filters: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/03-excel-filters/\n\n## The bottom line\n\nGrid.js is a lovely minimalist table for search, sort, and paginate. When you outgrow that - virtualization, editing, grouping, tree - SvGrid is the Svelte-native grid that already covers it.\n\n## Frequently asked questions\n\n### SvGrid vs Grid.js - which should I choose?\n\nGrid.js for a small table that only needs search, sort, and paginate; SvGrid when you need virtualization, editing, Excel-style filters, or grouping.\n\n### Is SvGrid heavier than Grid.js?\n\nYes, because it does much more. For a trivial table Grid.js is lighter; for a real data grid SvGrid pays for itself.\n\n### Does SvGrid support server-side data like Grid.js?\n\nYes - externalSort and externalFilter, plus the onSortingChange and onFiltersChange events.\n\n### Is SvGrid a good Grid.js alternative for Svelte?\n\nGrid.js is deliberately small and covers the essentials well. People outgrow it at the same points every time: virtualization for large datasets, inline editing, row grouping and server-side data. SvGrid has those, and keeps a headless core so a simple table does not have to carry the weight of the full component.\n\n## Related comparisons and docs\n\n- [SvGrid vs Tabulator](https://svgrid.com/compare/tabulator/)\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/)\n- [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Getting started with SvGrid](https://svgrid.com/docs/getting-started/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- gridjs-site: Free and open source under MIT; works with React, Angular, Vue and vanilla JavaScript; an extensible data pipeline (checked 12 Sep 2026) https://gridjs.io/\n"
8646
+ "markdown": "# SvGrid vs Grid.js\n\nCanonical URL: https://svgrid.com/compare/gridjs/\nFacts verified: 2026-09-12\nCompetitor package: gridjs\nCompetitor site: https://gridjs.io/\n\nGrid.js is a small, elegant, framework-agnostic table library focused on the essentials, MIT-licensed.\n\n## Verdict\n\nGrid.js is a lovely lightweight pick for search / sort / paginate. SvGrid is the fuller, Svelte-native grid when you need virtualization, editing, grouping, or large datasets.\n\nMigration guide: moving from Grid.js to SvGrid: https://svgrid.com/docs/help/migrating-from-gridjs/\n\n## At a glance\n\n| Fact | SvGrid | Grid.js | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 6.2.0 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 3 Mar 2024 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (JavaScript) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 393,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 17.8 KB JS + 2.4 KB CSS | 9 Nov 2026 |\n| Frameworks | Svelte 5 | JavaScript | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | Grid.js is MIT, free and open source; no licence is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; gridjs 6.2.0 entry `Grid` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nGrid.js keeps a deliberately small surface: search, sort, pagination, and basic server-side data, rendered through a tiny vanilla-JS core with framework wrappers. It is MIT-licensed and easy to learn.\n\nSvGrid trades that minimalism for depth: a Svelte-5-native render component with virtualization, an Excel-style filter menu, inline editing, grouping, tree data, and an imperative API.\n\n## Looking for a Grid.js alternative for Svelte?\n\nGrid.js is deliberately small and covers the essentials well. People outgrow it at the same points every time: virtualization for large datasets, inline editing, row grouping and server-side data. SvGrid has those, and keeps a headless core so a simple table does not have to carry the weight of the full component.\n\n## What they have in common\n\n- MIT-licensed.\n- Search, sort, and pagination out of the box.\n- Server-side data support.\n- Small, approachable API.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native instead of a vanilla core + wrapper.\n- Row + column virtualization for large datasets.\n- Excel-style filters, inline editing, grouping, tree, master/detail.\n- Headless engine + imperative API + @svgrid/mcp.\n\n## Where Grid.js is stronger\n\n- Tiny footprint if all you need is search / sort / paginate.\n- Framework-agnostic core.\n- Very fast to learn for a simple table.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | Grid.js |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - vanilla JavaScript with React, Angular and Vue wrappers |\n| Search / sort / paginate | Yes | Yes |\n| Excel-style filters | Yes | No |\n| Inline editing | Yes | No |\n| Virtualization | Yes | No |\n| Grouping / tree / master-detail | Yes | No |\n| Footprint for a simple table | Partial - Larger | Yes - Tiny |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You need virtualization, editing, grouping, or tree data.\n- You are on Svelte 5 and want a native component.\n- You expect the table to grow beyond search / sort / paginate.\n\n## When to choose Grid.js\n\n- You only need search / sort / paginate over a modest dataset.\n- You want the smallest possible footprint.\n- You want a framework-agnostic drop-in.\n\n## Migration effort\n\nEstimated effort: 30 minutes to 1 hour per table.\n\nWhat you give up:\n\n- A very small library for a simple table on any page.\n- Wrappers for React, Angular and Vue.\n\nMigration guide: moving from Grid.js to SvGrid: https://svgrid.com/docs/help/migrating-from-gridjs/\n\n## Live example\n\nExcel Filters: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/03-excel-filters/\n\n## The bottom line\n\nGrid.js is a lovely minimalist table for search, sort, and paginate. When you outgrow that - virtualization, editing, grouping, tree - SvGrid is the Svelte-native grid that already covers it.\n\n## Frequently asked questions\n\n### SvGrid vs Grid.js - which should I choose?\n\nGrid.js for a small table that only needs search, sort, and paginate; SvGrid when you need virtualization, editing, Excel-style filters, or grouping.\n\n### Is SvGrid heavier than Grid.js?\n\nYes, because it does much more. For a trivial table Grid.js is lighter; for a real data grid SvGrid pays for itself.\n\n### Does SvGrid support server-side data like Grid.js?\n\nYes - externalSort and externalFilter, plus the onSortingChange and onFiltersChange events.\n\n### Is SvGrid a good Grid.js alternative for Svelte?\n\nGrid.js is deliberately small and covers the essentials well. People outgrow it at the same points every time: virtualization for large datasets, inline editing, row grouping and server-side data. SvGrid has those, and keeps a headless core so a simple table does not have to carry the weight of the full component.\n\n## Related comparisons and docs\n\n- [SvGrid vs Tabulator](https://svgrid.com/compare/tabulator/)\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/)\n- [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Getting started with SvGrid](https://svgrid.com/docs/getting-started/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- gridjs-site: Free and open source under MIT; works with React, Angular, Vue and vanilla JavaScript; an extensible data pipeline (checked 12 Sep 2026) https://gridjs.io/\n"
8647
8647
  },
8648
8648
  {
8649
8649
  "slug": "compare/handsontable",
@@ -8671,7 +8671,7 @@ export const docs = [
8671
8671
  "path": "docs/_data/comparisons/mui-x-datagrid.json",
8672
8672
  "title": "SvGrid vs MUI X DataGrid",
8673
8673
  "section": "Comparisons",
8674
- "markdown": "# SvGrid vs MUI X DataGrid\n\nCanonical URL: https://svgrid.com/compare/mui-x-datagrid/\nFacts verified: 2026-09-12\nCompetitor package: @mui/x-data-grid\nCompetitor site: https://mui.com/x/react-data-grid/\n\nMUI X DataGrid is the data grid of the Material UI ecosystem on React, split into a free Community tier and paid Pro and Premium tiers.\n\n## Verdict\n\nMUI X is the obvious pick if you are already on Material UI and React. SvGrid is Svelte-5-native, ships more in its MIT core (pinning, master/detail, tree data, grouping and cell selection are not Pro or Premium features), and carries no Material Design opinion.\n\nMigration guide: moving from MUI X DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-mui-x/\n\n## At a glance\n\n| Fact | SvGrid | MUI X DataGrid | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 9.17.0 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 30 Oct 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (React) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 13,100,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 158.4 KB JS, no separate stylesheet, @emotion/react + @emotion/styled + @mui/material + @mui/system + react + react-dom external | 9 Nov 2026 |\n| Frameworks | Svelte 5 | React | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | MUI X Community is free under MIT. mui.com lists Pro at $299 per year per developer, Premium at $599 per year per developer and Enterprise at $1,399 per year per developer, with perpetual and annual licence models | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @mui/x-data-grid 9.17.0 entry `DataGrid component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nMUI X DataGrid is what most React teams on Material UI reach for. The Community edition is MIT and covers sorting, filtering, pagination, CSV export and row virtualization. Column pinning, column virtualization, row reordering, tree data and master/detail are Pro; row grouping, aggregation, cell selection, Excel export, pivoting and the AI assistant are Premium. Both paid tiers are sold per developer per year, and the sources below link to the pages that mark each feature.\n\nSvGrid is the inverse trade: one Svelte-5-native grid whose MIT core already includes pinning, master/detail, tree data, grouping with aggregation, cell-range selection with a fill handle, integrated charts and in-grid AI helpers you point at your own model, plus an optional @svgrid/enterprise pack for Excel, PDF and print output, import and pivot tables.\n\n## Looking for a MUI X DataGrid alternative for Svelte?\n\nMUI X DataGrid is React-only, so this is less an alternative than a port. The concepts carry over almost one to one: column definitions, a free tier and a paid tier, server-side rows, an imperative API. What does not carry over is Material Design: SvGrid is themed through CSS custom properties, so you bring your own look, and Emotion and Material UI stay out of the bundle.\n\n## What they have in common\n\n- A Community / paid split with an MIT free tier.\n- Sorting, filtering, pagination, virtualization and inline editing.\n- Server-side data through a row model the app feeds.\n- TypeScript-first with an imperative API for code outside the component.\n- An official MCP server so AI assistants read the real docs.\n- An in-grid AI assistant driven by natural language.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native: runes and snippets, not a React component behind a wrapper.\n- Column pinning, master/detail, tree data, row grouping with aggregation and cell-range selection are in the MIT core; MUI X sells them as Pro or Premium.\n- The in-grid AI helpers (natural-language filter, smart fill, summarise, classify) are free in @svgrid/grid and run against a provider you register; MUI X's AI assistant is Premium and routes prompts through MUI's hosted service unless you build your own.\n- No Material Design opinion: theme through --sg-* tokens, with no Emotion or Material UI in the bundle. The At a glance table has the measured sizes.\n- Integrated charts in the MIT core, without a separate charts package.\n\n## Where MUI X DataGrid is stronger\n\n- First-class fit if the app is already built on Material UI and React: theming, date pickers and charts share one design system.\n- Pro and Premium features are vendor-supported with priority support and guaranteed response times on the Enterprise plan.\n- A very large React community and years of production use.\n- The Premium AI assistant comes with a hosted prompt service and starter credits, so there is nothing to run yourself.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | MUI X DataGrid |\n| --- | --- | --- |\n| Svelte 5 native | Yes - runes and snippets | No - React only |\n| Sorting, filtering, pagination | Yes | Yes - Community |\n| Row virtualization | Yes | Yes - Community |\n| Column virtualization | Yes | Paid tier - Pro |\n| Column pinning | Yes - MIT core, left and right | Paid tier - Pro |\n| Row reordering by drag | Yes - MIT core | Paid tier - Pro |\n| Master/detail | Yes - MIT core | Paid tier - Pro |\n| Tree data | Yes - MIT core | Paid tier - Pro |\n| Row grouping with aggregation | Yes - MIT core | Paid tier - Premium |\n| Cell-range selection and clipboard | Yes - MIT core, with a fill handle | Paid tier - Premium |\n| CSV export | Yes - MIT core | Yes - Community |\n| Excel export | Paid tier - @svgrid/enterprise | Paid tier - Premium |\n| Pivot tables | Paid tier - @svgrid/enterprise | Paid tier - Premium |\n| Integrated charts | Yes - SvGridChart in the MIT core | Partial - MUI X Charts is a separate package |\n| In-grid AI assistant | Yes - MIT core, bring your own model | Paid tier - Premium, through MUI's hosted service or your own proxy |\n| MCP server for AI assistants | Yes - @svgrid/mcp, MIT: docs, examples, code check, scaffold, preview | Yes - @mui/mcp, free: docs and examples |\n| Inline editing | Yes | Yes - Community |\n| Design-system independence | Yes - --sg-* tokens, no UI framework dependency | Partial - Material UI, Emotion and the MUI system are peer dependencies |\n| Core licence | Yes - MIT | Yes - MIT (Community) |\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want a grid shaped by runes, not a React component.\n- You want pinning, master/detail, tree data, grouping and cell selection without buying Pro or Premium.\n- You do not want Material Design, Emotion or the MUI system in your bundle.\n- You want the in-grid AI helpers to talk to a model you choose and host.\n\n## When to choose MUI X DataGrid\n\n- Your app is already built on Material UI and React.\n- You need Premium grouping, aggregation, pivoting or Excel export as vendor-supported features with a support contract.\n- You want one grid that matches React teammates across the MUI X suite, including date pickers and charts.\n- You want a hosted AI assistant with nothing to run yourself.\n\n## Migration effort\n\nEstimated effort: 1-3 hours per grid.\n\nWhat you give up:\n\n- Material Design out of the box; SvGrid ships neutral tokens and you theme it.\n- The rest of MUI X in the same design system: date pickers, charts and the tree view.\n- A hosted AI assistant with starter credits; SvGrid's helpers need a model provider you register.\n- React-specific patterns such as slots and hooks; SvGrid uses snippets and props.\n\nMigration guide: moving from MUI X DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-mui-x/\n\n## Live example\n\nTree And Master Detail: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/08-tree-and-master-detail/\n\n## The bottom line\n\nIf your app already lives in Material UI and React, MUI X is the path of least resistance, and its Premium tier now includes a hosted AI assistant. On Svelte 5, SvGrid is the better fit: it puts pinning, master/detail, tree data, grouping and cell selection in the MIT core that MUI X reserves for Pro and Premium, keeps Material UI and Emotion out of the bundle, and runs its AI helpers against a model you choose.\n\n## Frequently asked questions\n\n### SvGrid vs MUI X DataGrid: which one for Svelte?\n\nSvGrid, if your stack is Svelte. MUI X is React-only and depends on Material UI, Emotion and the MUI system as peer dependencies; SvGrid is Svelte-5-native with no design-system dependency.\n\n### Which MUI X Pro and Premium features are free in SvGrid?\n\nColumn pinning, column virtualization, row reordering, master/detail and tree data (Pro in MUI X), and row grouping, aggregation and cell-range selection (Premium in MUI X) ship in the MIT core of SvGrid. Excel export and pivoting are paid on both sides: Premium in MUI X, @svgrid/enterprise in SvGrid.\n\n### Does MUI X have an AI assistant like SvGrid?\n\nYes, in Premium. mui.com describes a natural-language assistant (sort by name, show amounts larger than a value) that sends prompts through a proxy to MUI's hosted service with an API key, with starter credits included and the option to build your own processing service. SvGrid's natural-language filter, smart fill, summarise and classify helpers are free in @svgrid/grid and only run against a model provider you register, so nothing leaves your infrastructure unless you route it there.\n\n### Does MUI ship an MCP server?\n\nYes. @mui/mcp gives an AI client the official docs and code examples for Material UI and MUI X, including the Data Grid. @svgrid/mcp does the same for SvGrid and adds a code checker that validates props and imports against the real API surface, a scaffold tool and a preview.\n\n### Is SvGrid cheaper than MUI X Pro or Premium?\n\nThe prices mui.com listed on the date we read them are in the pricing line at the top of this page, per developer per year for Pro and Premium. SvGrid's core is MIT and @svgrid/enterprise is $599 per developer per year for the enterprise grid or $999 with the spreadsheet and Studio, both for unlimited applications. The difference is what the free tier holds: most features MUI X sells in Pro and Premium are in SvGrid's MIT core, so many teams never need the paid pack.\n\n### How long does a MUI X to SvGrid migration take?\n\nThe migration guide budgets one to three hours per grid. Column definitions and the row model map directly; the time goes into replacing React slots and renderCell functions with Svelte snippets and re-theming away from Material Design.\n\n### Is SvGrid a good MUI X DataGrid alternative for Svelte?\n\nMUI X DataGrid is React-only, so this is less an alternative than a port. The concepts carry over almost one to one: column definitions, a free tier and a paid tier, server-side rows, an imperative API. What does not carry over is Material Design: SvGrid is themed through CSS custom properties, so you bring your own look, and Emotion and Material UI stay out of the bundle.\n\n## Related comparisons and docs\n\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [SvGrid vs React Data Grid (adazzle)](https://svgrid.com/compare/react-data-grid/)\n- [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [AI assistant](https://svgrid.com/docs/help/ai/)\n- [Column pinning](https://svgrid.com/docs/help/columns/column-pinning/)\n- [Master / detail (nested grids)](https://svgrid.com/docs/help/rows/master-detail/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- mui-overview: Feature list with plan badges: CSV export, cell editing and row virtualization in Community; column pinning, column reordering, row reordering, tree data and column virtualization Pro; row grouping, aggregation, Excel export and pivoting Premium; React package with Material UI peer dependencies (checked 12 Sep 2026) https://mui.com/x/react-data-grid/\n- mui-master-detail: Master-detail row panels carry the Pro plan badge (checked 12 Sep 2026) https://mui.com/x/react-data-grid/master-detail/\n- mui-cell-selection: Cell selection carries the Premium plan badge: with the Data Grid Premium, you can apply the cellSelection prop (checked 12 Sep 2026) https://mui.com/x/react-data-grid/cell-selection/\n- mui-ai-assistant: AI Assistant is Premium: natural-language prompts such as sort by name or show amounts larger than 1000, sent through a server-side proxy to MUI's hosted service with an API key from MUI Console; Premium licence holders get starter credits; a self-built prompt service is optional (checked 12 Sep 2026) https://mui.com/x/react-data-grid/ai-assistant/\n- mui-mcp: Official MCP server (@mui/mcp) that quotes the official docs and code examples for MUI X components including the Data Grid; installed with npx, no payment (checked 12 Sep 2026) https://mui.com/x/introduction/mcp/\n- mui-pricing: Community free forever under MIT; Pro $299 per year per dev; Premium $599 per year per dev; Enterprise $1,399 per year per dev; perpetual and annual models both allow perpetual production use (checked 12 Sep 2026) https://mui.com/pricing/\n"
8674
+ "markdown": "# SvGrid vs MUI X DataGrid\n\nCanonical URL: https://svgrid.com/compare/mui-x-datagrid/\nFacts verified: 2026-09-12\nCompetitor package: @mui/x-data-grid\nCompetitor site: https://mui.com/x/react-data-grid/\n\nMUI X DataGrid is the data grid of the Material UI ecosystem on React, split into a free Community tier and paid Pro and Premium tiers.\n\n## Verdict\n\nMUI X is the obvious pick if you are already on Material UI and React. SvGrid is Svelte-5-native, ships more in its MIT core (pinning, master/detail, tree data, grouping and cell selection are not Pro or Premium features), and carries no Material Design opinion.\n\nMigration guide: moving from MUI X DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-mui-x/\n\n## At a glance\n\n| Fact | SvGrid | MUI X DataGrid | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 9.17.0 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 30 Oct 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (React) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 13,100,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 158.4 KB JS, no separate stylesheet, @emotion/react + @emotion/styled + @mui/material + @mui/system + react + react-dom external | 9 Nov 2026 |\n| Frameworks | Svelte 5 | React | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | MUI X Community is free under MIT. mui.com lists Pro at $299 per year per developer, Premium at $599 per year per developer and Enterprise at $1,399 per year per developer, with perpetual and annual licence models | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @mui/x-data-grid 9.17.0 entry `DataGrid component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nMUI X DataGrid is what most React teams on Material UI reach for. The Community edition is MIT and covers sorting, filtering, pagination, CSV export and row virtualization. Column pinning, column virtualization, row reordering, tree data and master/detail are Pro; row grouping, aggregation, cell selection, Excel export, pivoting and the AI assistant are Premium. Both paid tiers are sold per developer per year, and the sources below link to the pages that mark each feature.\n\nSvGrid is the inverse trade: one Svelte-5-native grid whose MIT core already includes pinning, master/detail, tree data, grouping with aggregation, cell-range selection with a fill handle, integrated charts and in-grid AI helpers you point at your own model, plus an optional @svgrid/enterprise pack for Excel, PDF and print output, import and pivot tables.\n\n## Looking for a MUI X DataGrid alternative for Svelte?\n\nMUI X DataGrid is React-only, so this is less an alternative than a port. The concepts carry over almost one to one: column definitions, a free tier and a paid tier, server-side rows, an imperative API. What does not carry over is Material Design: SvGrid is themed through CSS custom properties, so you bring your own look, and Emotion and Material UI stay out of the bundle.\n\n## What they have in common\n\n- A Community / paid split with an MIT free tier.\n- Sorting, filtering, pagination, virtualization and inline editing.\n- Server-side data through a row model the app feeds.\n- TypeScript-first with an imperative API for code outside the component.\n- An official MCP server so AI assistants read the real docs.\n- An in-grid AI assistant driven by natural language.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native: runes and snippets, not a React component behind a wrapper.\n- Column pinning, master/detail, tree data, row grouping with aggregation and cell-range selection are in the MIT core; MUI X sells them as Pro or Premium.\n- The in-grid AI helpers (natural-language filter, smart fill, summarise, classify) are free in @svgrid/grid and run against a provider you register; MUI X's AI assistant is Premium and routes prompts through MUI's hosted service unless you build your own.\n- No Material Design opinion: theme through --sg-* tokens, with no Emotion or Material UI in the bundle. The At a glance table has the measured sizes.\n- Integrated charts in the MIT core, without a separate charts package.\n\n## Where MUI X DataGrid is stronger\n\n- First-class fit if the app is already built on Material UI and React: theming, date pickers and charts share one design system.\n- Pro and Premium features are vendor-supported with priority support and guaranteed response times on the Enterprise plan.\n- A very large React community and years of production use.\n- The Premium AI assistant comes with a hosted prompt service and starter credits, so there is nothing to run yourself.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | MUI X DataGrid |\n| --- | --- | --- |\n| Svelte 5 native | Yes - runes and snippets | No - React only |\n| Sorting, filtering, pagination | Yes | Yes - Community |\n| Row virtualization | Yes | Yes - Community |\n| Column virtualization | Yes | Paid tier - Pro |\n| Column pinning | Yes - MIT core, left and right | Paid tier - Pro |\n| Row reordering by drag | Yes - MIT core | Paid tier - Pro |\n| Master/detail | Yes - MIT core | Paid tier - Pro |\n| Tree data | Yes - MIT core | Paid tier - Pro |\n| Row grouping with aggregation | Yes - MIT core | Paid tier - Premium |\n| Cell-range selection and clipboard | Yes - MIT core, with a fill handle | Paid tier - Premium |\n| CSV export | Yes - MIT core | Yes - Community |\n| Excel export | Paid tier - @svgrid/enterprise | Paid tier - Premium |\n| Pivot tables | Paid tier - @svgrid/enterprise | Paid tier - Premium |\n| Integrated charts | Yes - SvGridChart in the MIT core | Partial - MUI X Charts is a separate package |\n| In-grid AI assistant | Yes - MIT core, bring your own model | Paid tier - Premium, through MUI's hosted service or your own proxy |\n| MCP server for AI assistants | Yes - @svgrid/mcp, MIT: docs, examples, code check, scaffold, preview | Yes - @mui/mcp, free: docs and examples |\n| Inline editing | Yes | Yes - Community |\n| Design-system independence | Yes - --sg-* tokens, no UI framework dependency | Partial - Material UI, Emotion and the MUI system are peer dependencies |\n| Core licence | Yes - MIT | Yes - MIT (Community) |\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want a grid shaped by runes, not a React component.\n- You want pinning, master/detail, tree data, grouping and cell selection without buying Pro or Premium.\n- You do not want Material Design, Emotion or the MUI system in your bundle.\n- You want the in-grid AI helpers to talk to a model you choose and host.\n\n## When to choose MUI X DataGrid\n\n- Your app is already built on Material UI and React.\n- You need Premium grouping, aggregation, pivoting or Excel export as vendor-supported features with a support contract.\n- You want one grid that matches React teammates across the MUI X suite, including date pickers and charts.\n- You want a hosted AI assistant with nothing to run yourself.\n\n## Migration effort\n\nEstimated effort: 1-3 hours per grid.\n\nWhat you give up:\n\n- Material Design out of the box; SvGrid ships neutral tokens and you theme it.\n- The rest of MUI X in the same design system: date pickers, charts and the tree view.\n- A hosted AI assistant with starter credits; SvGrid's helpers need a model provider you register.\n- React-specific patterns such as slots and hooks; SvGrid uses snippets and props.\n\nMigration guide: moving from MUI X DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-mui-x/\n\n## Live example\n\nTree And Master Detail: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/08-tree-and-master-detail/\n\n## The bottom line\n\nIf your app already lives in Material UI and React, MUI X is the path of least resistance, and its Premium tier now includes a hosted AI assistant. On Svelte 5, SvGrid is the better fit: it puts pinning, master/detail, tree data, grouping and cell selection in the MIT core that MUI X reserves for Pro and Premium, keeps Material UI and Emotion out of the bundle, and runs its AI helpers against a model you choose.\n\n## Frequently asked questions\n\n### SvGrid vs MUI X DataGrid: which one for Svelte?\n\nSvGrid, if your stack is Svelte. MUI X is React-only and depends on Material UI, Emotion and the MUI system as peer dependencies; SvGrid is Svelte-5-native with no design-system dependency.\n\n### Which MUI X Pro and Premium features are free in SvGrid?\n\nColumn pinning, column virtualization, row reordering, master/detail and tree data (Pro in MUI X), and row grouping, aggregation and cell-range selection (Premium in MUI X) ship in the MIT core of SvGrid. Excel export and pivoting are paid on both sides: Premium in MUI X, @svgrid/enterprise in SvGrid.\n\n### Does MUI X have an AI assistant like SvGrid?\n\nYes, in Premium. mui.com describes a natural-language assistant (sort by name, show amounts larger than a value) that sends prompts through a proxy to MUI's hosted service with an API key, with starter credits included and the option to build your own processing service. SvGrid's natural-language filter, smart fill, summarise and classify helpers are free in @svgrid/grid and only run against a model provider you register, so nothing leaves your infrastructure unless you route it there.\n\n### Does MUI ship an MCP server?\n\nYes. @mui/mcp gives an AI client the official docs and code examples for Material UI and MUI X, including the Data Grid. @svgrid/mcp does the same for SvGrid and adds a code checker that validates props and imports against the real API surface, a scaffold tool and a preview.\n\n### Is SvGrid cheaper than MUI X Pro or Premium?\n\nThe prices mui.com listed on the date we read them are in the pricing line at the top of this page, per developer per year for Pro and Premium. SvGrid's core is MIT and @svgrid/enterprise is $599 per developer per year for the enterprise grid or $999 with the spreadsheet and Studio, both for unlimited applications. The difference is what the free tier holds: most features MUI X sells in Pro and Premium are in SvGrid's MIT core, so many teams never need the paid pack.\n\n### How long does a MUI X to SvGrid migration take?\n\nThe migration guide budgets one to three hours per grid. Column definitions and the row model map directly; the time goes into replacing React slots and renderCell functions with Svelte snippets and re-theming away from Material Design.\n\n### Is SvGrid a good MUI X DataGrid alternative for Svelte?\n\nMUI X DataGrid is React-only, so this is less an alternative than a port. The concepts carry over almost one to one: column definitions, a free tier and a paid tier, server-side rows, an imperative API. What does not carry over is Material Design: SvGrid is themed through CSS custom properties, so you bring your own look, and Emotion and Material UI stay out of the bundle.\n\n## Related comparisons and docs\n\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [SvGrid vs React Data Grid (adazzle)](https://svgrid.com/compare/react-data-grid/)\n- [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [AI assistant](https://svgrid.com/docs/help/ai/)\n- [Column pinning](https://svgrid.com/docs/help/columns/column-pinning/)\n- [Master / detail (nested grids)](https://svgrid.com/docs/help/rows/master-detail/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- mui-overview: Feature list with plan badges: CSV export, cell editing and row virtualization in Community; column pinning, column reordering, row reordering, tree data and column virtualization Pro; row grouping, aggregation, Excel export and pivoting Premium; React package with Material UI peer dependencies (checked 12 Sep 2026) https://mui.com/x/react-data-grid/\n- mui-master-detail: Master-detail row panels carry the Pro plan badge (checked 12 Sep 2026) https://mui.com/x/react-data-grid/master-detail/\n- mui-cell-selection: Cell selection carries the Premium plan badge: with the Data Grid Premium, you can apply the cellSelection prop (checked 12 Sep 2026) https://mui.com/x/react-data-grid/cell-selection/\n- mui-ai-assistant: AI Assistant is Premium: natural-language prompts such as sort by name or show amounts larger than 1000, sent through a server-side proxy to MUI's hosted service with an API key from MUI Console; Premium licence holders get starter credits; a self-built prompt service is optional (checked 12 Sep 2026) https://mui.com/x/react-data-grid/ai-assistant/\n- mui-mcp: Official MCP server (@mui/mcp) that quotes the official docs and code examples for MUI X components including the Data Grid; installed with npx, no payment (checked 12 Sep 2026) https://mui.com/x/introduction/mcp/\n- mui-pricing: Community free forever under MIT; Pro $299 per year per dev; Premium $599 per year per dev; Enterprise $1,399 per year per dev; perpetual and annual models both allow perpetual production use (checked 12 Sep 2026) https://mui.com/pricing/\n"
8675
8675
  },
8676
8676
  {
8677
8677
  "slug": "compare/nocodb",
@@ -8692,7 +8692,7 @@ export const docs = [
8692
8692
  "path": "docs/_data/comparisons/react-data-grid.json",
8693
8693
  "title": "SvGrid vs React Data Grid (adazzle)",
8694
8694
  "section": "Comparisons",
8695
- "markdown": "# SvGrid vs React Data Grid (adazzle)\n\nCanonical URL: https://svgrid.com/compare/react-data-grid/\nFacts verified: 2026-09-12\nCompetitor package: react-data-grid\nCompetitor site: https://github.com/adazzle/react-data-grid\n\nreact-data-grid is a fast, Excel-like data grid for React, MIT-licensed.\n\n## Verdict\n\nreact-data-grid is a strong, lightweight React grid with great editing. SvGrid is the Svelte-5-native equivalent, with a similar feature set and no React dependency.\n\nMigration guide: moving from React Data Grid (adazzle) to SvGrid: https://svgrid.com/docs/help/migrating-from-react-data-grid/\n\n## At a glance\n\n| Fact | SvGrid | React Data Grid (adazzle) | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 7.0.0-beta.64 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 5 Nov 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (React) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 2,650,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 25.6 KB JS + 2.3 KB CSS, react + react-dom external | 9 Nov 2026 |\n| Frameworks | Svelte 5 | React | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | react-data-grid is MIT on npm and free; no licence is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; react-data-grid 7.0.0-beta.64 entry `DataGrid component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nreact-data-grid (the adazzle project) is a focused, performant React grid known for spreadsheet-style editing, virtualization, and a clean column API. It is MIT-licensed and React-only.\n\nSvGrid offers the same shape - virtualized DOM, typed columns, inline editing, custom cells - built on Svelte 5 runes and snippets rather than React components.\n\n## Looking for a React Data Grid alternative for Svelte?\n\nreact-data-grid is React-only, so this is a port rather than a swap, and the shapes line up well: both are Excel-like, both MIT, both driven by column definitions. Cell renderers become Svelte snippets and hooks become runes. If you are staying on React, there is no reason to move.\n\n## What they have in common\n\n- MIT-licensed.\n- Virtualized rows and columns.\n- Spreadsheet-style inline editing.\n- Typed column definitions.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native - runes and snippets, not React.\n- Built-in Excel-style filter menu and the imperative API.\n- Grouping, tree data, and master/detail in the core.\n- @svgrid/mcp MCP server + optional Enterprise pack.\n\n## Where React Data Grid (adazzle) is stronger\n\n- First-class fit for a React codebase.\n- Mature, focused, and battle-tested in React apps.\n- Large React community.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | React Data Grid (adazzle) |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - React only |\n| Virtualization | Yes | Yes |\n| Inline editing | Yes | Yes |\n| Excel-style filter menu | Yes | Partial - BYO filter UI |\n| Grouping / tree | Yes | Yes - row grouping and a TreeDataGrid |\n| Master/detail | Yes | Partial - bring your own detail row |\n| Headless engine | Yes | No |\n| AI / MCP integration | Yes | No |\n| Fill handle and copy/paste | Yes | Yes - cell copy, paste and value dragging |\n\n## When to choose SvGrid\n\n- You are on Svelte 5 rather than React.\n- You want filters, grouping, and master/detail without building them.\n- You want a headless engine option and an MCP server.\n\n## When to choose React Data Grid (adazzle)\n\n- Your app is React and you want a focused, proven grid.\n- You want the React community and ecosystem around it.\n- You only need the feature set react-data-grid already covers.\n\n## Migration effort\n\nEstimated effort: 2-4 hours per grid.\n\nWhat you give up:\n\n- A very small React grid with the spreadsheet basics built in.\n- React 19 features such as the use hook in cell renderers.\n\nMigration guide: moving from React Data Grid (adazzle) to SvGrid: https://svgrid.com/docs/help/migrating-from-react-data-grid/\n\n## Live example\n\nFill Handle: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/95-fill-handle/\n\n## The bottom line\n\nreact-data-grid is a strong, focused choice in React. On Svelte 5, SvGrid is the equivalent - similar editing and virtualization - plus a built-in filter menu, grouping, and a headless core, with no React dependency.\n\n## Frequently asked questions\n\n### SvGrid vs react-data-grid - which for Svelte?\n\nSvGrid - it is Svelte-5-native, while react-data-grid is React-only. They share a similar editing and virtualization model.\n\n### Does SvGrid have built-in filtering unlike react-data-grid?\n\nYes - SvGrid ships an Excel-style filter menu and a global filter; react-data-grid leaves the filter UI to you.\n\n### Is SvGrid free like react-data-grid?\n\nYes, the core is MIT. SvGrid adds an optional Enterprise pack for export, import, pivot, and AI.\n\n### Is SvGrid a good React Data Grid alternative for Svelte?\n\nreact-data-grid is React-only, so this is a port rather than a swap, and the shapes line up well: both are Excel-like, both MIT, both driven by column definitions. Cell renderers become Svelte snippets and hooks become runes. If you are staying on React, there is no reason to move.\n\n## Related comparisons and docs\n\n- [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/)\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/)\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Cell text selection](https://svgrid.com/docs/help/cells/cell-text-selection/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- rdg-readme: React 19.2+; virtualization of rows and columns, multi-column sorting, cell editing, row grouping and a TreeDataGrid, frozen columns, cell copy and paste, cell value dragging; light and dark mode through color-scheme (checked 12 Sep 2026) https://github.com/adazzle/react-data-grid\n- rdg-npm: The latest tag is a 7.0 beta line; licence MIT (checked 12 Sep 2026) https://www.npmjs.com/package/react-data-grid\n"
8695
+ "markdown": "# SvGrid vs React Data Grid (adazzle)\n\nCanonical URL: https://svgrid.com/compare/react-data-grid/\nFacts verified: 2026-09-12\nCompetitor package: react-data-grid\nCompetitor site: https://github.com/adazzle/react-data-grid\n\nreact-data-grid is a fast, Excel-like data grid for React, MIT-licensed.\n\n## Verdict\n\nreact-data-grid is a strong, lightweight React grid with great editing. SvGrid is the Svelte-5-native equivalent, with a similar feature set and no React dependency.\n\nMigration guide: moving from React Data Grid (adazzle) to SvGrid: https://svgrid.com/docs/help/migrating-from-react-data-grid/\n\n## At a glance\n\n| Fact | SvGrid | React Data Grid (adazzle) | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 7.0.0-beta.64 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 5 Nov 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (React) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 2,650,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 25.6 KB JS + 2.3 KB CSS, react + react-dom external | 9 Nov 2026 |\n| Frameworks | Svelte 5 | React | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | react-data-grid is MIT on npm and free; no licence is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; react-data-grid 7.0.0-beta.64 entry `DataGrid component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nreact-data-grid (the adazzle project) is a focused, performant React grid known for spreadsheet-style editing, virtualization, and a clean column API. It is MIT-licensed and React-only.\n\nSvGrid offers the same shape - virtualized DOM, typed columns, inline editing, custom cells - built on Svelte 5 runes and snippets rather than React components.\n\n## Looking for a React Data Grid alternative for Svelte?\n\nreact-data-grid is React-only, so this is a port rather than a swap, and the shapes line up well: both are Excel-like, both MIT, both driven by column definitions. Cell renderers become Svelte snippets and hooks become runes. If you are staying on React, there is no reason to move.\n\n## What they have in common\n\n- MIT-licensed.\n- Virtualized rows and columns.\n- Spreadsheet-style inline editing.\n- Typed column definitions.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native - runes and snippets, not React.\n- Built-in Excel-style filter menu and the imperative API.\n- Grouping, tree data, and master/detail in the core.\n- @svgrid/mcp MCP server + optional Enterprise pack.\n\n## Where React Data Grid (adazzle) is stronger\n\n- First-class fit for a React codebase.\n- Mature, focused, and battle-tested in React apps.\n- Large React community.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | React Data Grid (adazzle) |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - React only |\n| Virtualization | Yes | Yes |\n| Inline editing | Yes | Yes |\n| Excel-style filter menu | Yes | Partial - BYO filter UI |\n| Grouping / tree | Yes | Yes - row grouping and a TreeDataGrid |\n| Master/detail | Yes | Partial - bring your own detail row |\n| Headless engine | Yes | No |\n| AI / MCP integration | Yes | No |\n| Fill handle and copy/paste | Yes | Yes - cell copy, paste and value dragging |\n\n## When to choose SvGrid\n\n- You are on Svelte 5 rather than React.\n- You want filters, grouping, and master/detail without building them.\n- You want a headless engine option and an MCP server.\n\n## When to choose React Data Grid (adazzle)\n\n- Your app is React and you want a focused, proven grid.\n- You want the React community and ecosystem around it.\n- You only need the feature set react-data-grid already covers.\n\n## Migration effort\n\nEstimated effort: 2-4 hours per grid.\n\nWhat you give up:\n\n- A very small React grid with the spreadsheet basics built in.\n- React 19 features such as the use hook in cell renderers.\n\nMigration guide: moving from React Data Grid (adazzle) to SvGrid: https://svgrid.com/docs/help/migrating-from-react-data-grid/\n\n## Live example\n\nFill Handle: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/95-fill-handle/\n\n## The bottom line\n\nreact-data-grid is a strong, focused choice in React. On Svelte 5, SvGrid is the equivalent - similar editing and virtualization - plus a built-in filter menu, grouping, and a headless core, with no React dependency.\n\n## Frequently asked questions\n\n### SvGrid vs react-data-grid - which for Svelte?\n\nSvGrid - it is Svelte-5-native, while react-data-grid is React-only. They share a similar editing and virtualization model.\n\n### Does SvGrid have built-in filtering unlike react-data-grid?\n\nYes - SvGrid ships an Excel-style filter menu and a global filter; react-data-grid leaves the filter UI to you.\n\n### Is SvGrid free like react-data-grid?\n\nYes, the core is MIT. SvGrid adds an optional Enterprise pack for export, import, pivot, and AI.\n\n### Is SvGrid a good React Data Grid alternative for Svelte?\n\nreact-data-grid is React-only, so this is a port rather than a swap, and the shapes line up well: both are Excel-like, both MIT, both driven by column definitions. Cell renderers become Svelte snippets and hooks become runes. If you are staying on React, there is no reason to move.\n\n## Related comparisons and docs\n\n- [SvGrid vs Glide Data Grid](https://svgrid.com/compare/glide-data-grid/)\n- [SvGrid vs MUI X DataGrid](https://svgrid.com/compare/mui-x-datagrid/)\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Cell text selection](https://svgrid.com/docs/help/cells/cell-text-selection/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- rdg-readme: React 19.2+; virtualization of rows and columns, multi-column sorting, cell editing, row grouping and a TreeDataGrid, frozen columns, cell copy and paste, cell value dragging; light and dark mode through color-scheme (checked 12 Sep 2026) https://github.com/adazzle/react-data-grid\n- rdg-npm: The latest tag is a 7.0 beta line; licence MIT (checked 12 Sep 2026) https://www.npmjs.com/package/react-data-grid\n"
8696
8696
  },
8697
8697
  {
8698
8698
  "slug": "compare/smart-grid",
@@ -8706,14 +8706,14 @@ export const docs = [
8706
8706
  "path": "docs/_data/comparisons/svar-svelte-datagrid.json",
8707
8707
  "title": "SvGrid vs SVAR Svelte DataGrid",
8708
8708
  "section": "Comparisons",
8709
- "markdown": "# SvGrid vs SVAR Svelte DataGrid\n\nCanonical URL: https://svgrid.com/compare/svar-svelte-datagrid/\nFacts verified: 2026-09-12\nCompetitor package: @svar-ui/svelte-grid\nCompetitor site: https://svar.dev/svelte/datagrid/\n\nSVAR Svelte DataGrid (@svar-ui/svelte-grid, formerly wx-svelte-grid) is a Svelte-native, MIT-licensed data grid from XB Software, the team behind Webix, with React and Vue versions of the same grid.\n\n## Verdict\n\nSVAR is the closest genuinely Svelte-native rival: both grids are MIT and both target Svelte 5. SvGrid has the wider capability surface (grouping with aggregation, master/detail, Excel-style range editing with a fill handle, integrated charts, a headless core, in-grid AI helpers and an MCP server); SVAR is the smaller grid with print in the free package, React and Vue versions, and the Webix team behind it.\n\nMigration guide: moving from SVAR Svelte DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-svar-svelte-datagrid/\n\n## At a glance\n\n| Fact | SvGrid | SVAR Svelte DataGrid | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 2.7.6 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 7 Nov 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Yes (peer svelte ^5.0.0) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 106,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 53.1 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte, React, Vue | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | The DataGrid is MIT and free for commercial use with no paid tier; svar.dev sells PRO editions of its Gantt, Calendar and Kanban components and offers paid consulting and custom development, with no price list for the grid | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @svar-ui/svelte-grid 2.7.6 entry `Grid component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nSVAR DataGrid (npm: @svar-ui/svelte-grid; the older wx-svelte-grid package is deprecated and points to it) is part of the SVAR suite from XB Software, the makers of Webix, and is one of the few grids built for Svelte rather than wrapped from another framework. Its docs list in-cell editing, header and advanced filtering, single and multi-column sorting, tree data, virtual scrolling for rows and columns, columns pinned to the left, column and row spans in the header and footer, auto-sized rows, undo and redo, a context menu, row reordering, CSV export, a print API and a REST data provider, all under MIT.\n\nSvGrid overlaps on the Svelte-native angle and reaches further: row grouping with aggregation, master/detail, Excel-style cell-range selection with clipboard and a fill handle, pinned rows and left and right column pinning, integrated charts, a server-side row model, a headless engine, in-grid AI helpers and the @svgrid/mcp server, all in the MIT package. Excel, PDF and print output, import, pivot tables and the Kanban and Scheduler views of the grid live in the paid @svgrid/enterprise pack, whereas SVAR keeps its grid entirely free and sells PRO editions of its Gantt, Calendar and Kanban instead.\n\n## Looking for a SVAR Svelte DataGrid alternative for Svelte?\n\nThe closest comparison on this list, since both are Svelte-native and both are MIT, so licensing does not decide it. The differences are depth and direction: SvGrid ships a larger feature surface, a UI component suite, in-grid AI helpers and an MCP server, while SVAR spans Svelte, React and Vue and keeps a smaller core. The At a glance table above has both measured bundle sizes; if a compact MIT grid with sorting, filtering, editing, tree data and CSV export is all you need, SVAR is a perfectly good answer.\n\n## What they have in common\n\n- Purpose-built for Svelte 5, not a cross-framework wrapper.\n- MIT-licensed and free for commercial use.\n- Sorting, filtering, in-cell editing, virtualization on rows and columns, tree data.\n- Auto-sized row heights, undo and redo, a context menu and drag-and-drop row reordering.\n- Themeable through CSS variables with light and dark themes, localization, WAI-ARIA markup and keyboard navigation.\n\n## Where SvGrid is stronger\n\n- Row grouping with aggregation and a drag-and-drop group panel; SVAR's docs list neither.\n- Master/detail and expandable detail rows; SVAR offers tree data only.\n- Excel-style cell-range selection with clipboard copy and paste and a fill handle; SVAR selects rows.\n- Pinned top and bottom rows, and columns pinned left and right; SVAR pins columns to the left only.\n- Integrated charts with no dependency, plus a chart wizard.\n- A server-side row model with cursor pagination, GraphQL, websocket and optimistic updates, beyond a REST data provider.\n- A headless engine (createSvGrid) you can compose, in addition to the render component.\n- In-grid AI helpers free in @svgrid/grid, and @svgrid/mcp with llms.txt so assistants answer and write code accurately.\n- A paid pack with Excel, PDF and styled-HTML export, print, pivot tables, Excel and CSV import and support. CSV, TSV and JSON export are free in the grid.\n\n## Where SVAR Svelte DataGrid is stronger\n\n- The whole grid is free, print API included; SvGrid puts print and Excel and PDF export in the paid pack. Both export CSV for free.\n- The same grid for React and Vue as well as Svelte; SvGrid is Svelte-only by design.\n- A smaller bundle: the At a glance table has both measured numbers.\n- XB Software and Webix behind it, a team that has built data grids for a long time.\n- Part of the wider SVAR suite, with paid consulting and custom development on offer.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | SVAR Svelte DataGrid |\n| --- | --- | --- |\n| Svelte 5 native | Yes | Yes - Svelte 5 compatible |\n| Licence | Yes - MIT core, paid enterprise pack | Yes - MIT, the grid is entirely free |\n| Headless engine | Yes - createSvGrid | No - component only |\n| Excel-style filter menu | Yes - built in | Yes - header filters plus the SVAR Filter component |\n| Cell-range selection and clipboard | Yes - built in | No - single and multiple row selection |\n| Fill handle | Yes | No - not in the feature list |\n| Row grouping with aggregation | Yes - with a group panel | No - not in the feature list |\n| Master/detail rows | Yes | No - tree data only |\n| Tree data | Yes | Yes |\n| Pinned columns | Yes - left and right | Partial - to the left |\n| Pinned rows (top and bottom) | Yes | No - not in the feature list |\n| Header and footer column spans | Yes | Yes |\n| Auto-sized row height | Yes - autoRowHeight | Yes |\n| Virtualization (rows and columns) | Yes | Yes |\n| Undo and redo, context menu, row drag | Yes | Yes |\n| Integrated charts | Yes - no dependency, plus a chart wizard | No - not in the feature list |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes - createServerDataSource: page or infinite block scroll, cursor, GraphQL, websocket | Partial - RestDataProvider and dynamic loading while scrolling |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise | No - not in the feature list |\n| CSV export | Yes - free, with TSV and JSON | Yes - free |\n| Excel and PDF export | Paid tier - @svgrid/enterprise | No - CSV only in the feature list |\n| Print | Paid tier - @svgrid/enterprise | Yes - print API, free |\n| Pivot tables | Paid tier - @svgrid/enterprise | No - not in the SVAR Svelte product list |\n| React and Vue versions | No - Svelte only | Yes |\n| In-grid AI helpers and MCP server | Yes - helpers free in @svgrid/grid; @svgrid/mcp, MIT | No - not mentioned in the docs |\n| Paid support | Yes - with @svgrid/enterprise, per developer per year | Partial - paid consulting and custom development; no support plan for the grid listed |\n\n## When to choose SvGrid\n\n- You need grouping with aggregation, master/detail, Excel-style range editing or pivot tables, which SVAR's docs do not list.\n- You want a headless engine plus a render component, not a component-only API.\n- You want integrated charts and a full server-side row model in one library.\n- You want in-grid AI helpers, an MCP server that checks generated code, and a paid support tier.\n\n## When to choose SVAR Svelte DataGrid\n\n- You want a fully free grid with print included and no paid tier at all.\n- You ship to React and Vue as well as Svelte and want one grid across them.\n- A smaller bundle matters more than grouping with aggregation, master/detail, range editing and the other features SvGrid adds.\n- You are already adopting the wider SVAR suite: Gantt, Calendar, Kanban, File Manager.\n\n## Migration effort\n\nEstimated effort: 2-4 hours per grid.\n\nWhat you give up:\n\n- Free printing: SVAR's print API is in the MIT grid, SvGrid's print view is in @svgrid/enterprise.\n- The React and Vue versions of the same grid.\n- A smaller bundle, by the measurements in the At a glance table.\n- The SVAR suite around the grid: Gantt, Calendar, Kanban and File Manager share its core.\n\nMigration guide: moving from SVAR Svelte DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-svar-svelte-datagrid/\n\n## Live example\n\nGrouping Aggregation: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/07-grouping-aggregation/\n\n## The bottom line\n\nBoth grids are Svelte-5-native and MIT, so this is not a licensing call. On capability SvGrid is the more complete grid: grouping, master/detail, range editing with a fill handle, charts, a server-side row model, a headless core, AI helpers and an MCP server. SVAR is the smaller, fully free, multi-framework option from the Webix team, and it prints for free where SvGrid charges for print and Excel and PDF export. Choose SvGrid when you need the deeper feature set or support; choose SVAR when a compact free grid covers your needs or you also ship React and Vue.\n\n## Frequently asked questions\n\n### SvGrid vs SVAR Svelte DataGrid: what is the difference?\n\nBoth are MIT-licensed, Svelte-5-native grids. SvGrid is the broader product: row grouping with aggregation, master/detail, Excel-style range selection with a fill handle, integrated charts, a server-side row model, a headless core, in-grid AI helpers and an MCP server. SVAR is a smaller, fully free grid with React and Vue versions from XB Software, with CSV export and a print API included.\n\n### Are SvGrid and SVAR both free?\n\nBoth are free under MIT, including commercial use. SVAR keeps the whole grid free, CSV export and print included, and sells PRO editions of its Gantt, Calendar and Kanban. SvGrid's grid is free too, including CSV, TSV and JSON export, charts and the AI helpers; Excel, PDF and print output, import, pivot tables, the Kanban and Scheduler views and support are in @svgrid/enterprise at $599 per developer per year.\n\n### Are both built for Svelte 5?\n\nYes. SvGrid is written in Svelte 5 runes, and svar.dev states the DataGrid is Svelte 5 compatible. Neither is a cross-framework wrapper, although SVAR also ships React and Vue versions of its grid.\n\n### Which grid is smaller?\n\nSVAR, by the measurements in the At a glance table: both packages were built alone with Vite, minified and gzipped with Svelte external, on the dates shown. SvGrid's extra size buys grouping, master/detail, range editing, charts and the server-side row model; features you never import, such as the date editors, menus and charts, load as separate chunks.\n\n### Does SVAR have row grouping, master/detail or cell-range selection?\n\nNot according to its docs overview on the date we read it, which lists tree data, single and multiple row selection, header and advanced filters, sorting, in-cell editing, columns pinned to the left, CSV export and a print API. SvGrid has all three in the MIT core, and the demos on this page show them.\n\n### How long does a SVAR to SvGrid migration take?\n\nThe migration guide budgets two to four hours per grid: a prop and event rename pass, since both use array-of-object column definitions, plus re-theming through --sg-* tokens.\n\n### Is SvGrid a good SVAR Svelte DataGrid alternative for Svelte?\n\nThe closest comparison on this list, since both are Svelte-native and both are MIT, so licensing does not decide it. The differences are depth and direction: SvGrid ships a larger feature surface, a UI component suite, in-grid AI helpers and an MCP server, while SVAR spans Svelte, React and Vue and keeps a smaller core. The At a glance table above has both measured bundle sizes; if a compact MIT grid with sorting, filtering, editing, tree data and CSV export is all you need, SVAR is a perfectly good answer.\n\n## Related comparisons and docs\n\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/)\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Grouping & aggregation](https://svgrid.com/docs/help/grouping-aggregation/)\n- [Master / detail (nested grids)](https://svgrid.com/docs/help/rows/master-detail/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- svar-datagrid: Licensed under MIT; Svelte 5 compatible; React and Vue versions available; lists multi-column sorting, header filters, in-cell editing, tree data, virtual scrolling, columns pinned to the left, context menu, row drag-and-drop, checkbox selection, pagination, hotkeys, WAI-ARIA, print, undo/redo (checked 12 Sep 2026) https://svar.dev/svelte/datagrid/\n- svar-overview: Feature list: in-cell editing, header and advanced filtering, sorting, single and multiple row selection, row reordering, context menu, undo/redo, tree data, pagination, colspans and rowspans in header and footer, collapsible column groups, pinned columns to the left, auto-sizing of rows and columns, localization, themes, export to CSV, print API, RestDataProvider, virtual scrolling for rows and columns, WAI-ARIA, keyboard navigation; no Excel export, row grouping, aggregation, master/detail, charts, AI or MCP (checked 12 Sep 2026) https://docs.svar.dev/svelte/grid/overview/\n- svar-svelte: DataGrid, Filter, Editor, File Manager and Core are MIT; Gantt, Calendar and Kanban have MIT cores with paid PRO editions; no pivot component in the list; no price list on the page (checked 12 Sep 2026) https://svar.dev/svelte/\n- svar-gantt: Offers paid consulting and custom development services; PRO edition pricing is on a separate page (checked 12 Sep 2026) https://svar.dev/svelte/gantt/\n"
8709
+ "markdown": "# SvGrid vs SVAR Svelte DataGrid\n\nCanonical URL: https://svgrid.com/compare/svar-svelte-datagrid/\nFacts verified: 2026-09-12\nCompetitor package: @svar-ui/svelte-grid\nCompetitor site: https://svar.dev/svelte/datagrid/\n\nSVAR Svelte DataGrid (@svar-ui/svelte-grid, formerly wx-svelte-grid) is a Svelte-native, MIT-licensed data grid from XB Software, the team behind Webix, with React and Vue versions of the same grid.\n\n## Verdict\n\nSVAR is the closest genuinely Svelte-native rival: both grids are MIT and both target Svelte 5. SvGrid has the wider capability surface (grouping with aggregation, master/detail, Excel-style range editing with a fill handle, integrated charts, a headless core, in-grid AI helpers and an MCP server); SVAR is the smaller grid with print in the free package, React and Vue versions, and the Webix team behind it.\n\nMigration guide: moving from SVAR Svelte DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-svar-svelte-datagrid/\n\n## At a glance\n\n| Fact | SvGrid | SVAR Svelte DataGrid | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 2.7.6 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 7 Nov 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Yes (peer svelte ^5.0.0) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 106,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 53.1 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte, React, Vue | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | The DataGrid is MIT and free for commercial use with no paid tier; svar.dev sells PRO editions of its Gantt, Calendar and Kanban components and offers paid consulting and custom development, with no price list for the grid | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @svar-ui/svelte-grid 2.7.6 entry `Grid component` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nSVAR DataGrid (npm: @svar-ui/svelte-grid; the older wx-svelte-grid package is deprecated and points to it) is part of the SVAR suite from XB Software, the makers of Webix, and is one of the few grids built for Svelte rather than wrapped from another framework. Its docs list in-cell editing, header and advanced filtering, single and multi-column sorting, tree data, virtual scrolling for rows and columns, columns pinned to the left, column and row spans in the header and footer, auto-sized rows, undo and redo, a context menu, row reordering, CSV export, a print API and a REST data provider, all under MIT.\n\nSvGrid overlaps on the Svelte-native angle and reaches further: row grouping with aggregation, master/detail, Excel-style cell-range selection with clipboard and a fill handle, pinned rows and left and right column pinning, integrated charts, a server-side row model, a headless engine, in-grid AI helpers and the @svgrid/mcp server, all in the MIT package. Excel, PDF and print output, import, pivot tables and the Kanban and Scheduler views of the grid live in the paid @svgrid/enterprise pack, whereas SVAR keeps its grid entirely free and sells PRO editions of its Gantt, Calendar and Kanban instead.\n\n## Looking for a SVAR Svelte DataGrid alternative for Svelte?\n\nThe closest comparison on this list, since both are Svelte-native and both are MIT, so licensing does not decide it. The differences are depth and direction: SvGrid ships a larger feature surface, a UI component suite, in-grid AI helpers and an MCP server, while SVAR spans Svelte, React and Vue and keeps a smaller core. The At a glance table above has both measured bundle sizes; if a compact MIT grid with sorting, filtering, editing, tree data and CSV export is all you need, SVAR is a perfectly good answer.\n\n## What they have in common\n\n- Purpose-built for Svelte 5, not a cross-framework wrapper.\n- MIT-licensed and free for commercial use.\n- Sorting, filtering, in-cell editing, virtualization on rows and columns, tree data.\n- Auto-sized row heights, undo and redo, a context menu and drag-and-drop row reordering.\n- Themeable through CSS variables with light and dark themes, localization, WAI-ARIA markup and keyboard navigation.\n\n## Where SvGrid is stronger\n\n- Row grouping with aggregation and a drag-and-drop group panel; SVAR's docs list neither.\n- Master/detail and expandable detail rows; SVAR offers tree data only.\n- Excel-style cell-range selection with clipboard copy and paste and a fill handle; SVAR selects rows.\n- Pinned top and bottom rows, and columns pinned left and right; SVAR pins columns to the left only.\n- Integrated charts with no dependency, plus a chart wizard.\n- A server-side row model with cursor pagination, GraphQL, websocket and optimistic updates, beyond a REST data provider.\n- A headless engine (createSvGrid) you can compose, in addition to the render component.\n- In-grid AI helpers free in @svgrid/grid, and @svgrid/mcp with llms.txt so assistants answer and write code accurately.\n- A paid pack with Excel, PDF and styled-HTML export, print, pivot tables, Excel and CSV import and support. CSV, TSV and JSON export are free in the grid.\n\n## Where SVAR Svelte DataGrid is stronger\n\n- The whole grid is free, print API included; SvGrid puts print and Excel and PDF export in the paid pack. Both export CSV for free.\n- The same grid for React and Vue as well as Svelte; SvGrid is Svelte-only by design.\n- A smaller bundle: the At a glance table has both measured numbers.\n- XB Software and Webix behind it, a team that has built data grids for a long time.\n- Part of the wider SVAR suite, with paid consulting and custom development on offer.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | SVAR Svelte DataGrid |\n| --- | --- | --- |\n| Svelte 5 native | Yes | Yes - Svelte 5 compatible |\n| Licence | Yes - MIT core, paid enterprise pack | Yes - MIT, the grid is entirely free |\n| Headless engine | Yes - createSvGrid | No - component only |\n| Excel-style filter menu | Yes - built in | Yes - header filters plus the SVAR Filter component |\n| Cell-range selection and clipboard | Yes - built in | No - single and multiple row selection |\n| Fill handle | Yes | No - not in the feature list |\n| Row grouping with aggregation | Yes - with a group panel | No - not in the feature list |\n| Master/detail rows | Yes | No - tree data only |\n| Tree data | Yes | Yes |\n| Pinned columns | Yes - left and right | Partial - to the left |\n| Pinned rows (top and bottom) | Yes | No - not in the feature list |\n| Header and footer column spans | Yes | Yes |\n| Auto-sized row height | Yes - autoRowHeight | Yes |\n| Virtualization (rows and columns) | Yes | Yes |\n| Undo and redo, context menu, row drag | Yes | Yes |\n| Integrated charts | Yes - no dependency, plus a chart wizard | No - not in the feature list |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes - createServerDataSource: page or infinite block scroll, cursor, GraphQL, websocket | Partial - RestDataProvider and dynamic loading while scrolling |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise | No - not in the feature list |\n| CSV export | Yes - free, with TSV and JSON | Yes - free |\n| Excel and PDF export | Paid tier - @svgrid/enterprise | No - CSV only in the feature list |\n| Print | Paid tier - @svgrid/enterprise | Yes - print API, free |\n| Pivot tables | Paid tier - @svgrid/enterprise | No - not in the SVAR Svelte product list |\n| React and Vue versions | No - Svelte only | Yes |\n| In-grid AI helpers and MCP server | Yes - helpers free in @svgrid/grid; @svgrid/mcp, MIT | No - not mentioned in the docs |\n| Paid support | Yes - with @svgrid/enterprise, per developer per year | Partial - paid consulting and custom development; no support plan for the grid listed |\n\n## When to choose SvGrid\n\n- You need grouping with aggregation, master/detail, Excel-style range editing or pivot tables, which SVAR's docs do not list.\n- You want a headless engine plus a render component, not a component-only API.\n- You want integrated charts and a full server-side row model in one library.\n- You want in-grid AI helpers, an MCP server that checks generated code, and a paid support tier.\n\n## When to choose SVAR Svelte DataGrid\n\n- You want a fully free grid with print included and no paid tier at all.\n- You ship to React and Vue as well as Svelte and want one grid across them.\n- A smaller bundle matters more than grouping with aggregation, master/detail, range editing and the other features SvGrid adds.\n- You are already adopting the wider SVAR suite: Gantt, Calendar, Kanban, File Manager.\n\n## Migration effort\n\nEstimated effort: 2-4 hours per grid.\n\nWhat you give up:\n\n- Free printing: SVAR's print API is in the MIT grid, SvGrid's print view is in @svgrid/enterprise.\n- The React and Vue versions of the same grid.\n- A smaller bundle, by the measurements in the At a glance table.\n- The SVAR suite around the grid: Gantt, Calendar, Kanban and File Manager share its core.\n\nMigration guide: moving from SVAR Svelte DataGrid to SvGrid: https://svgrid.com/docs/help/migrating-from-svar-svelte-datagrid/\n\n## Live example\n\nGrouping Aggregation: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/07-grouping-aggregation/\n\n## The bottom line\n\nBoth grids are Svelte-5-native and MIT, so this is not a licensing call. On capability SvGrid is the more complete grid: grouping, master/detail, range editing with a fill handle, charts, a server-side row model, a headless core, AI helpers and an MCP server. SVAR is the smaller, fully free, multi-framework option from the Webix team, and it prints for free where SvGrid charges for print and Excel and PDF export. Choose SvGrid when you need the deeper feature set or support; choose SVAR when a compact free grid covers your needs or you also ship React and Vue.\n\n## Frequently asked questions\n\n### SvGrid vs SVAR Svelte DataGrid: what is the difference?\n\nBoth are MIT-licensed, Svelte-5-native grids. SvGrid is the broader product: row grouping with aggregation, master/detail, Excel-style range selection with a fill handle, integrated charts, a server-side row model, a headless core, in-grid AI helpers and an MCP server. SVAR is a smaller, fully free grid with React and Vue versions from XB Software, with CSV export and a print API included.\n\n### Are SvGrid and SVAR both free?\n\nBoth are free under MIT, including commercial use. SVAR keeps the whole grid free, CSV export and print included, and sells PRO editions of its Gantt, Calendar and Kanban. SvGrid's grid is free too, including CSV, TSV and JSON export, charts and the AI helpers; Excel, PDF and print output, import, pivot tables, the Kanban and Scheduler views and support are in @svgrid/enterprise at $599 per developer per year.\n\n### Are both built for Svelte 5?\n\nYes. SvGrid is written in Svelte 5 runes, and svar.dev states the DataGrid is Svelte 5 compatible. Neither is a cross-framework wrapper, although SVAR also ships React and Vue versions of its grid.\n\n### Which grid is smaller?\n\nSVAR, by the measurements in the At a glance table: both packages were built alone with Vite, minified and gzipped with Svelte external, on the dates shown. SvGrid's extra size buys grouping, master/detail, range editing, charts and the server-side row model; features you never import, such as the date editors, menus and charts, load as separate chunks.\n\n### Does SVAR have row grouping, master/detail or cell-range selection?\n\nNot according to its docs overview on the date we read it, which lists tree data, single and multiple row selection, header and advanced filters, sorting, in-cell editing, columns pinned to the left, CSV export and a print API. SvGrid has all three in the MIT core, and the demos on this page show them.\n\n### How long does a SVAR to SvGrid migration take?\n\nThe migration guide budgets two to four hours per grid: a prop and event rename pass, since both use array-of-object column definitions, plus re-theming through --sg-* tokens.\n\n### Is SvGrid a good SVAR Svelte DataGrid alternative for Svelte?\n\nThe closest comparison on this list, since both are Svelte-native and both are MIT, so licensing does not decide it. The differences are depth and direction: SvGrid ships a larger feature surface, a UI component suite, in-grid AI helpers and an MCP server, while SVAR spans Svelte, React and Vue and keeps a smaller core. The At a glance table above has both measured bundle sizes; if a compact MIT grid with sorting, filtering, editing, tree data and CSV export is all you need, SVAR is a perfectly good answer.\n\n## Related comparisons and docs\n\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/)\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Grouping & aggregation](https://svgrid.com/docs/help/grouping-aggregation/)\n- [Master / detail (nested grids)](https://svgrid.com/docs/help/rows/master-detail/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- svar-datagrid: Licensed under MIT; Svelte 5 compatible; React and Vue versions available; lists multi-column sorting, header filters, in-cell editing, tree data, virtual scrolling, columns pinned to the left, context menu, row drag-and-drop, checkbox selection, pagination, hotkeys, WAI-ARIA, print, undo/redo (checked 12 Sep 2026) https://svar.dev/svelte/datagrid/\n- svar-overview: Feature list: in-cell editing, header and advanced filtering, sorting, single and multiple row selection, row reordering, context menu, undo/redo, tree data, pagination, colspans and rowspans in header and footer, collapsible column groups, pinned columns to the left, auto-sizing of rows and columns, localization, themes, export to CSV, print API, RestDataProvider, virtual scrolling for rows and columns, WAI-ARIA, keyboard navigation; no Excel export, row grouping, aggregation, master/detail, charts, AI or MCP (checked 12 Sep 2026) https://docs.svar.dev/svelte/grid/overview/\n- svar-svelte: DataGrid, Filter, Editor, File Manager and Core are MIT; Gantt, Calendar and Kanban have MIT cores with paid PRO editions; no pivot component in the list; no price list on the page (checked 12 Sep 2026) https://svar.dev/svelte/\n- svar-gantt: Offers paid consulting and custom development services; PRO edition pricing is on a separate page (checked 12 Sep 2026) https://svar.dev/svelte/gantt/\n"
8710
8710
  },
8711
8711
  {
8712
8712
  "slug": "compare/svelte-headless-table",
8713
8713
  "path": "docs/_data/comparisons/svelte-headless-table.json",
8714
8714
  "title": "SvGrid vs svelte-headless-table",
8715
8715
  "section": "Comparisons",
8716
- "markdown": "# SvGrid vs svelte-headless-table\n\nCanonical URL: https://svgrid.com/compare/svelte-headless-table/\nFacts verified: 2026-09-12\nCompetitor package: svelte-headless-table\nCompetitor site: https://svelte-headless-table.bryanmylee.com/\n\nsvelte-headless-table is the Svelte-native headless table library by Bryan Lee, one of the earliest Svelte-specific table options, now deprecated on Svelte 4.\n\n## Verdict\n\nsvelte-headless-table popularised headless tables in Svelte and still works on Svelte 4. Its author states there are no plans for a Svelte 5 port, so a Svelte 5 upgrade forces the question. SvGrid keeps the headless approach on Svelte 5 runes, adds virtualization and a rendered grid, and the @humanspeak fork is the option if you want to stay headless with the same API.\n\nMigration guide: moving from svelte-headless-table to SvGrid: https://svgrid.com/docs/help/migrating-from-svelte-headless-table/\n\n## At a glance\n\n| Fact | SvGrid | svelte-headless-table | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 0.18.3 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 28 Oct 2024 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Svelte 4 range (peer svelte ^4.0.0) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 96,400 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 5.7 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | svelte-headless-table is MIT and free; no paid licence or support is offered, and the author states there are no plans for a Svelte 5 port | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; svelte-headless-table 0.18.3 entry `whole package` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nsvelte-headless-table exposes a createTable function with composable plugins for sorting, filtering, column ordering and hiding, pagination, row selection, expansion, grouping with aggregation and column resizing, and leaves the markup to you. The npm package's peer range is Svelte 4, and the repository README states there are no current plans for a Svelte 5 port. A community fork, @humanspeak/svelte-headless-table, carries the same API on Svelte 5 runes; the At a glance table shows the original's registry facts, and the migration guide has the fork's.\n\nSvGrid takes the same headless approach but is written in Svelte 5 runes, adds a production render component, and covers virtualization, Excel-style filters, cell-range selection with a fill handle, inline editing, master/detail, a server-side row model and integrated charts natively.\n\n## Looking for a svelte-headless-table alternative for Svelte?\n\nThis is the clearest migration case on the list: the original package's peer range is Svelte 4, its last release was in 2024 and the README says there are no plans for a Svelte 5 port nor a maintenance schedule for Svelte 3 and 4. SvGrid keeps the headless approach you already chose and adds runes, virtualization and a render component if you want one. If you would rather keep the plugin API, the maintained community fork @humanspeak/svelte-headless-table targets Svelte 5, and its facts are in the guide.\n\n## What they have in common\n\n- Svelte-specific, not a port from another framework.\n- Headless-first composition: plugins in svelte-headless-table, a features object in SvGrid.\n- Sorting, filtering, pagination, row selection, expansion and grouping with aggregation in the engine.\n- MIT-licensed with a strong TypeScript story.\n\n## Where SvGrid is stronger\n\n- Built on Svelte 5 runes and released continuously; the original package is Svelte 4 only and its author has stopped.\n- A bundled <SvGrid /> component, so nobody assembles headers, body and virtualization by hand.\n- Virtualization on rows and columns built in; svelte-headless-table leaves it to a virtual-list library.\n- Excel-style filter menu, cell-range selection with a fill handle, inline editing, master/detail, a server-side row model and charts in the MIT package.\n- In-grid AI helpers free in @svgrid/grid and @svgrid/mcp for assistants, plus a paid support tier.\n\n## Where svelte-headless-table is stronger\n\n- A smaller bundle if all you want is the engine; the At a glance table has both measured sizes.\n- A plugin API many Svelte 4 codebases already know, and a documentation site that still reads well.\n- The @humanspeak fork keeps that API on Svelte 5 without a rewrite.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | svelte-headless-table |\n| --- | --- | --- |\n| Svelte 5 runes | Yes - native | No - Svelte 4 stores; no Svelte 5 port planned by the author |\n| Maintained release line | Yes | Partial - original deprecated; the @humanspeak fork is maintained on Svelte 5 |\n| Render component | Yes - <SvGrid /> | No - bring your own markup |\n| Headless engine | Yes - createSvGrid and a features object | Yes - createTable and plugins |\n| Sorting and column filtering | Yes | Yes - engine only |\n| Excel-style filter menu | Yes - built in | No - bring your own UI |\n| Cell-range selection and clipboard | Yes - built in, with a fill handle | No - row selection only |\n| Inline editing | Yes - built in | No - not a plugin |\n| Virtualization | Yes - rows and columns, built in | Partial - pair with a virtual-list library |\n| Pagination | Yes | Yes |\n| Row grouping with aggregation | Yes - engine and rendered group rows | Yes - engine only |\n| Row expansion and tree | Yes | Yes - expansion plugin |\n| Column ordering, hiding, resizing | Yes | Yes |\n| Master/detail | Yes | No - not a plugin |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes | No - not a plugin |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise | No - not a plugin |\n| Integrated charts | Yes - SvGridChart | No - not part of a headless engine |\n| In-grid AI helpers and MCP server | Yes - helpers free in @svgrid/grid; @svgrid/mcp, MIT | No - not in the docs |\n| Paid support | Yes - with @svgrid/enterprise, per developer per year | No - none offered |\n\n## When to choose SvGrid\n\n- You are moving to Svelte 5 and want a maintained grid that keeps the headless approach.\n- You want virtualization, an Excel-style filter menu, editing and cell selection without building them.\n- You want a rendered component for the common case and a headless core for the rest.\n- You want a support tier behind the library.\n\n## When to choose svelte-headless-table\n\n- You are staying on Svelte 4 for now and the plugin set already covers you.\n- You want the smallest possible engine and will keep writing the markup yourself.\n- You want the same plugin API on Svelte 5 and are happy to move to the @humanspeak fork.\n\n## Migration effort\n\nEstimated effort: 1-3 hours per grid.\n\nWhat you give up:\n\n- The plugin API and its documentation site, if your team knows it well.\n- The smallest engine-only bundle; the At a glance table has both measured sizes.\n- Nothing in maintenance terms on the original package, which is where most migrations start.\n\nMigration guide: moving from svelte-headless-table to SvGrid: https://svgrid.com/docs/help/migrating-from-svelte-headless-table/\n\n## Live example\n\nHeadless Table: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/186-headless-table/\n\n## The bottom line\n\nFor a Svelte 4 codebase that is not moving, svelte-headless-table still works. For a Svelte 5 codebase, the original package is not an option and its author says so; the choice is between the @humanspeak fork, which keeps the plugin API, and SvGrid, which keeps the headless approach and adds virtualization, filters, editing, selection and a rendered grid in the same MIT package.\n\n## Frequently asked questions\n\n### Does svelte-headless-table work with Svelte 5?\n\nThe original package's peer range is Svelte 4 and its README states there are no current plans for a Svelte 5 port nor a maintenance schedule for Svelte 3 and 4. The community fork @humanspeak/svelte-headless-table targets Svelte 5 with runes; SvGrid is written in Svelte 5 from the first line.\n\n### Can I keep a headless table with SvGrid?\n\nYes. createSvGrid plus the row-model factories is a headless engine; you can drive your own markup as before and adopt the SvGrid component only where you want it. The headless demos on this page show a plain table and a virtualized one.\n\n### Is SvGrid heavier than svelte-headless-table?\n\nYes, by the measurements in the At a glance table, because it ships a rendered grid, a virtualizer, filter menus and editors that a headless engine leaves to you. If you only need the engine and will write the rest, the difference is what you would otherwise write.\n\n### How long does a migration take?\n\nThe migration guide budgets one to three hours per grid; most of the time goes into deleting the markup and plugins the rendered grid replaces. Column definitions and the plugin concepts (sort, filter, page, select, expand, group) map one to one onto SvGrid's features.\n\n### Should I move to the @humanspeak fork instead?\n\nIf you want the same plugin API on Svelte 5 and are happy to keep writing the markup, yes, and its registry facts are in the migration guide. If the upgrade is also a chance to stop maintaining a table layer, SvGrid gives you the rendered grid and keeps the headless core for the cases that need it.\n\n### Is SvGrid a good svelte-headless-table alternative for Svelte?\n\nThis is the clearest migration case on the list: the original package's peer range is Svelte 4, its last release was in 2024 and the README says there are no plans for a Svelte 5 port nor a maintenance schedule for Svelte 3 and 4. SvGrid keeps the headless approach you already chose and adds runes, virtualization and a render component if you want one. If you would rather keep the plugin API, the maintained community fork @humanspeak/svelte-headless-table targets Svelte 5, and its facts are in the guide.\n\n## Related comparisons and docs\n\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/)\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n- [Headless virtualization](https://svgrid.com/docs/help/headless/virtualization/)\n- [Grouping & aggregation](https://svgrid.com/docs/help/grouping-aggregation/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- sht-github: README: there are no current plans to work on a Svelte 5 port of this project nor is there a maintenance schedule for Svelte 3/4; the project is flagged as deprecated and users are encouraged to seek alternatives or fork (checked 12 Sep 2026) https://github.com/bryanmylee/svelte-headless-table\n- sht-docs: Advertised plugins: multi-sorting, column filtering, global filtering, column reordering, hiding columns, pagination, row grouping and aggregation, row expansion, row selection, column resizing; headless, you write the markup (checked 12 Sep 2026) https://svelte-headless-table.bryanmylee.com/\n- sht-fork: MIT; part of the Humanspeak family of runes-native Svelte 5 packages; the npm package declares a Svelte 5 peer range (checked 12 Sep 2026) https://github.com/humanspeak/svelte-headless-table\n"
8716
+ "markdown": "# SvGrid vs svelte-headless-table\n\nCanonical URL: https://svgrid.com/compare/svelte-headless-table/\nFacts verified: 2026-09-12\nCompetitor package: svelte-headless-table\nCompetitor site: https://svelte-headless-table.bryanmylee.com/\n\nsvelte-headless-table is the Svelte-native headless table library by Bryan Lee, one of the earliest Svelte-specific table options, now deprecated on Svelte 4.\n\n## Verdict\n\nsvelte-headless-table popularised headless tables in Svelte and still works on Svelte 4. Its author states there are no plans for a Svelte 5 port, so a Svelte 5 upgrade forces the question. SvGrid keeps the headless approach on Svelte 5 runes, adds virtualization and a rendered grid, and the @humanspeak fork is the option if you want to stay headless with the same API.\n\nMigration guide: moving from svelte-headless-table to SvGrid: https://svgrid.com/docs/help/migrating-from-svelte-headless-table/\n\n## At a glance\n\n| Fact | SvGrid | svelte-headless-table | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 0.18.3 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 28 Oct 2024 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Svelte 4 range (peer svelte ^4.0.0) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 96,400 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 5.7 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | svelte-headless-table is MIT and free; no paid licence or support is offered, and the author states there are no plans for a Svelte 5 port | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; svelte-headless-table 0.18.3 entry `whole package` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nsvelte-headless-table exposes a createTable function with composable plugins for sorting, filtering, column ordering and hiding, pagination, row selection, expansion, grouping with aggregation and column resizing, and leaves the markup to you. The npm package's peer range is Svelte 4, and the repository README states there are no current plans for a Svelte 5 port. A community fork, @humanspeak/svelte-headless-table, carries the same API on Svelte 5 runes; the At a glance table shows the original's registry facts, and the migration guide has the fork's.\n\nSvGrid takes the same headless approach but is written in Svelte 5 runes, adds a production render component, and covers virtualization, Excel-style filters, cell-range selection with a fill handle, inline editing, master/detail, a server-side row model and integrated charts natively.\n\n## Looking for a svelte-headless-table alternative for Svelte?\n\nThis is the clearest migration case on the list: the original package's peer range is Svelte 4, its last release was in 2024 and the README says there are no plans for a Svelte 5 port nor a maintenance schedule for Svelte 3 and 4. SvGrid keeps the headless approach you already chose and adds runes, virtualization and a render component if you want one. If you would rather keep the plugin API, the maintained community fork @humanspeak/svelte-headless-table targets Svelte 5, and its facts are in the guide.\n\n## What they have in common\n\n- Svelte-specific, not a port from another framework.\n- Headless-first composition: plugins in svelte-headless-table, a features object in SvGrid.\n- Sorting, filtering, pagination, row selection, expansion and grouping with aggregation in the engine.\n- MIT-licensed with a strong TypeScript story.\n\n## Where SvGrid is stronger\n\n- Built on Svelte 5 runes and released continuously; the original package is Svelte 4 only and its author has stopped.\n- A bundled <SvGrid /> component, so nobody assembles headers, body and virtualization by hand.\n- Virtualization on rows and columns built in; svelte-headless-table leaves it to a virtual-list library.\n- Excel-style filter menu, cell-range selection with a fill handle, inline editing, master/detail, a server-side row model and charts in the MIT package.\n- In-grid AI helpers free in @svgrid/grid and @svgrid/mcp for assistants, plus a paid support tier.\n\n## Where svelte-headless-table is stronger\n\n- A smaller bundle if all you want is the engine; the At a glance table has both measured sizes.\n- A plugin API many Svelte 4 codebases already know, and a documentation site that still reads well.\n- The @humanspeak fork keeps that API on Svelte 5 without a rewrite.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | svelte-headless-table |\n| --- | --- | --- |\n| Svelte 5 runes | Yes - native | No - Svelte 4 stores; no Svelte 5 port planned by the author |\n| Maintained release line | Yes | Partial - original deprecated; the @humanspeak fork is maintained on Svelte 5 |\n| Render component | Yes - <SvGrid /> | No - bring your own markup |\n| Headless engine | Yes - createSvGrid and a features object | Yes - createTable and plugins |\n| Sorting and column filtering | Yes | Yes - engine only |\n| Excel-style filter menu | Yes - built in | No - bring your own UI |\n| Cell-range selection and clipboard | Yes - built in, with a fill handle | No - row selection only |\n| Inline editing | Yes - built in | No - not a plugin |\n| Virtualization | Yes - rows and columns, built in | Partial - pair with a virtual-list library |\n| Pagination | Yes | Yes |\n| Row grouping with aggregation | Yes - engine and rendered group rows | Yes - engine only |\n| Row expansion and tree | Yes | Yes - expansion plugin |\n| Column ordering, hiding, resizing | Yes | Yes |\n| Master/detail | Yes | No - not a plugin |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes | No - not a plugin |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise | No - not a plugin |\n| Integrated charts | Yes - SvGridChart | No - not part of a headless engine |\n| In-grid AI helpers and MCP server | Yes - helpers free in @svgrid/grid; @svgrid/mcp, MIT | No - not in the docs |\n| Paid support | Yes - with @svgrid/enterprise, per developer per year | No - none offered |\n\n## When to choose SvGrid\n\n- You are moving to Svelte 5 and want a maintained grid that keeps the headless approach.\n- You want virtualization, an Excel-style filter menu, editing and cell selection without building them.\n- You want a rendered component for the common case and a headless core for the rest.\n- You want a support tier behind the library.\n\n## When to choose svelte-headless-table\n\n- You are staying on Svelte 4 for now and the plugin set already covers you.\n- You want the smallest possible engine and will keep writing the markup yourself.\n- You want the same plugin API on Svelte 5 and are happy to move to the @humanspeak fork.\n\n## Migration effort\n\nEstimated effort: 1-3 hours per grid.\n\nWhat you give up:\n\n- The plugin API and its documentation site, if your team knows it well.\n- The smallest engine-only bundle; the At a glance table has both measured sizes.\n- Nothing in maintenance terms on the original package, which is where most migrations start.\n\nMigration guide: moving from svelte-headless-table to SvGrid: https://svgrid.com/docs/help/migrating-from-svelte-headless-table/\n\n## Live example\n\nHeadless Table: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/186-headless-table/\n\n## The bottom line\n\nFor a Svelte 4 codebase that is not moving, svelte-headless-table still works. For a Svelte 5 codebase, the original package is not an option and its author says so; the choice is between the @humanspeak fork, which keeps the plugin API, and SvGrid, which keeps the headless approach and adds virtualization, filters, editing, selection and a rendered grid in the same MIT package.\n\n## Frequently asked questions\n\n### Does svelte-headless-table work with Svelte 5?\n\nThe original package's peer range is Svelte 4 and its README states there are no current plans for a Svelte 5 port nor a maintenance schedule for Svelte 3 and 4. The community fork @humanspeak/svelte-headless-table targets Svelte 5 with runes; SvGrid is written in Svelte 5 from the first line.\n\n### Can I keep a headless table with SvGrid?\n\nYes. createSvGrid plus the row-model factories is a headless engine; you can drive your own markup as before and adopt the SvGrid component only where you want it. The headless demos on this page show a plain table and a virtualized one.\n\n### Is SvGrid heavier than svelte-headless-table?\n\nYes, by the measurements in the At a glance table, because it ships a rendered grid, a virtualizer, filter menus and editors that a headless engine leaves to you. If you only need the engine and will write the rest, the difference is what you would otherwise write.\n\n### How long does a migration take?\n\nThe migration guide budgets one to three hours per grid; most of the time goes into deleting the markup and plugins the rendered grid replaces. Column definitions and the plugin concepts (sort, filter, page, select, expand, group) map one to one onto SvGrid's features.\n\n### Should I move to the @humanspeak fork instead?\n\nIf you want the same plugin API on Svelte 5 and are happy to keep writing the markup, yes, and its registry facts are in the migration guide. If the upgrade is also a chance to stop maintaining a table layer, SvGrid gives you the rendered grid and keeps the headless core for the cases that need it.\n\n### Is SvGrid a good svelte-headless-table alternative for Svelte?\n\nThis is the clearest migration case on the list: the original package's peer range is Svelte 4, its last release was in 2024 and the README says there are no plans for a Svelte 5 port nor a maintenance schedule for Svelte 3 and 4. SvGrid keeps the headless approach you already chose and adds runes, virtualization and a render component if you want one. If you would rather keep the plugin API, the maintained community fork @humanspeak/svelte-headless-table targets Svelte 5, and its facts are in the guide.\n\n## Related comparisons and docs\n\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/)\n- [SvGrid vs @vincjo/datatables](https://svgrid.com/compare/vincjo-datatables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n- [Headless virtualization](https://svgrid.com/docs/help/headless/virtualization/)\n- [Grouping & aggregation](https://svgrid.com/docs/help/grouping-aggregation/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- sht-github: README: there are no current plans to work on a Svelte 5 port of this project nor is there a maintenance schedule for Svelte 3/4; the project is flagged as deprecated and users are encouraged to seek alternatives or fork (checked 12 Sep 2026) https://github.com/bryanmylee/svelte-headless-table\n- sht-docs: Advertised plugins: multi-sorting, column filtering, global filtering, column reordering, hiding columns, pagination, row grouping and aggregation, row expansion, row selection, column resizing; headless, you write the markup (checked 12 Sep 2026) https://svelte-headless-table.bryanmylee.com/\n- sht-fork: MIT; part of the Humanspeak family of runes-native Svelte 5 packages; the npm package declares a Svelte 5 peer range (checked 12 Sep 2026) https://github.com/humanspeak/svelte-headless-table\n"
8717
8717
  },
8718
8718
  {
8719
8719
  "slug": "compare/svelte-ui-kit-tables",
@@ -8734,21 +8734,21 @@ export const docs = [
8734
8734
  "path": "docs/_data/comparisons/tabulator.json",
8735
8735
  "title": "SvGrid vs Tabulator",
8736
8736
  "section": "Comparisons",
8737
- "markdown": "# SvGrid vs Tabulator\n\nCanonical URL: https://svgrid.com/compare/tabulator/\nFacts verified: 2026-09-12\nCompetitor package: tabulator-tables\nCompetitor site: https://tabulator.info/\n\nTabulator is a feature-rich, framework-agnostic vanilla-JS interactive table library, MIT-licensed.\n\n## Verdict\n\nTabulator is enormously capable and free, but it is vanilla JS you wire into Svelte imperatively. SvGrid delivers a similar feature surface as a Svelte-5-native component with reactive data binding.\n\nMigration guide: moving from Tabulator to SvGrid: https://svgrid.com/docs/help/migrating-from-tabulator/\n\n## At a glance\n\n| Fact | SvGrid | Tabulator | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 6.7.1 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 22 Oct 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (JavaScript) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 826,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 109.5 KB JS + 3.8 KB CSS | 9 Nov 2026 |\n| Frameworks | Svelte 5 | JavaScript | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | Tabulator is MIT and free; no licence or support plan is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; tabulator-tables 6.7.1 entry `TabulatorFull` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nTabulator packs a huge feature set - sorting, filtering, grouping, editing, tree data, formatters, and more - into a framework-agnostic vanilla-JS library. You instantiate it against a DOM element and drive it through its imperative API.\n\nSvGrid covers the same ground but as a Svelte 5 component: data is a reactive `$state` array, cells are snippets, and the imperative API is there when you want it - no manual mount/teardown or DOM bridging.\n\n## Looking for a Tabulator alternative for Svelte?\n\nTabulator is mature, MIT and framework-agnostic, and the reason people move is usually reactivity: it owns its own DOM, so Svelte state and table state have to be kept in sync by hand. SvGrid is driven by your runes directly. Tabulator still wins on breadth of built-in features and on being usable from plain HTML with no build step.\n\n## What they have in common\n\n- Broad feature set: sort, filter, group, edit, tree, format.\n- MIT-licensed / free for commercial use.\n- Virtualized rendering for large datasets.\n- Rich, mature documentation.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native - reactive data binding, no imperative mount/teardown.\n- Custom cells are Svelte snippets, not HTML-string formatters.\n- Headless engine available in addition to the component.\n- @svgrid/mcp MCP server for AI assistants.\n\n## Where Tabulator is stronger\n\n- Framework-agnostic - the same library across vanilla, React, Vue, Angular.\n- Very large built-in formatter / editor catalogue.\n- Long track record and a big community.\n- No build step required to drop into a plain HTML page.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | Tabulator |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - vanilla JavaScript |\n| Reactive data binding | Yes - $state | Partial - Imperative setData() |\n| Custom cells | Yes - Snippets / components | Partial - HTML-string formatters |\n| Sort / filter / group | Yes | Yes |\n| Inline editing | Yes | Yes |\n| Tree data | Yes | Yes |\n| Virtualization | Yes | Yes |\n| Headless engine | Yes | No |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want reactive data, not imperative setData calls.\n- You want Svelte snippets for custom cells instead of HTML strings.\n- You want a headless engine option and an MCP server.\n\n## When to choose Tabulator\n\n- You need one table library across vanilla JS, React, Vue, and Angular.\n- You want a no-build-step drop-in for a plain HTML page.\n- You rely on a specific Tabulator formatter / module.\n\n## Migration effort\n\nEstimated effort: 2-4 hours per grid.\n\nWhat you give up:\n\n- One library for plain JavaScript pages as well as React, Angular and Vue.\n- Built-in download to CSV, XLSX and PDF without a paid pack; SvGrid has CSV free and Excel and PDF in @svgrid/enterprise.\n\nMigration guide: moving from Tabulator to SvGrid: https://svgrid.com/docs/help/migrating-from-tabulator/\n\n## Live example\n\nExcel Filters: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/03-excel-filters/\n\n## The bottom line\n\nTabulator is a free, feature-packed table that runs anywhere. If your app is Svelte 5, SvGrid gives you a comparable feature set with reactive data binding and snippet cells instead of imperative setData and HTML-string formatters.\n\n## Frequently asked questions\n\n### SvGrid vs Tabulator for Svelte - which is better?\n\nFor Svelte 5, SvGrid: reactive $state data and Svelte snippet cells, with no imperative mount or teardown. Tabulator wins if you need one library across vanilla JS, React, Vue, and Angular.\n\n### Is SvGrid free like Tabulator?\n\nYes - the core is MIT, like Tabulator. Only the optional @svgrid/enterprise pack is paid.\n\n### Can I reuse Tabulator formatters in SvGrid?\n\nYou re-create them as Svelte snippets via renderSnippet, which is more flexible than HTML-string formatters.\n\n### Is SvGrid a good Tabulator alternative for Svelte?\n\nTabulator is mature, MIT and framework-agnostic, and the reason people move is usually reactivity: it owns its own DOM, so Svelte state and table state have to be kept in sync by hand. SvGrid is driven by your runes directly. Tabulator still wins on breadth of built-in features and on being usable from plain HTML with no build step.\n\n## Related comparisons and docs\n\n- [SvGrid vs Grid.js](https://svgrid.com/compare/gridjs/)\n- [SvGrid vs Smart.Grid (Smart Web Components)](https://svgrid.com/compare/smart-grid/)\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Cell components](https://svgrid.com/docs/help/cells/cell-components/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- tabulator-readme: MIT; a vanilla JavaScript library that builds interactive tables from HTML tables, arrays or JSON, with support for React, Angular and Vue; installed as tabulator-tables (checked 12 Sep 2026) https://github.com/olifolkerd/tabulator\n"
8737
+ "markdown": "# SvGrid vs Tabulator\n\nCanonical URL: https://svgrid.com/compare/tabulator/\nFacts verified: 2026-09-12\nCompetitor package: tabulator-tables\nCompetitor site: https://tabulator.info/\n\nTabulator is a feature-rich, framework-agnostic vanilla-JS interactive table library, MIT-licensed.\n\n## Verdict\n\nTabulator is enormously capable and free, but it is vanilla JS you wire into Svelte imperatively. SvGrid delivers a similar feature surface as a Svelte-5-native component with reactive data binding.\n\nMigration guide: moving from Tabulator to SvGrid: https://svgrid.com/docs/help/migrating-from-tabulator/\n\n## At a glance\n\n| Fact | SvGrid | Tabulator | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 6.7.1 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 22 Oct 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | No Svelte package (JavaScript) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 826,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 109.5 KB JS + 3.8 KB CSS | 9 Nov 2026 |\n| Frameworks | Svelte 5 | JavaScript | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | Tabulator is MIT and free; no licence or support plan is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; tabulator-tables 6.7.1 entry `TabulatorFull` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nTabulator packs a huge feature set - sorting, filtering, grouping, editing, tree data, formatters, and more - into a framework-agnostic vanilla-JS library. You instantiate it against a DOM element and drive it through its imperative API.\n\nSvGrid covers the same ground but as a Svelte 5 component: data is a reactive `$state` array, cells are snippets, and the imperative API is there when you want it - no manual mount/teardown or DOM bridging.\n\n## Looking for a Tabulator alternative for Svelte?\n\nTabulator is mature, MIT and framework-agnostic, and the reason people move is usually reactivity: it owns its own DOM, so Svelte state and table state have to be kept in sync by hand. SvGrid is driven by your runes directly. Tabulator still wins on breadth of built-in features and on being usable from plain HTML with no build step.\n\n## What they have in common\n\n- Broad feature set: sort, filter, group, edit, tree, format.\n- MIT-licensed / free for commercial use.\n- Virtualized rendering for large datasets.\n- Rich, mature documentation.\n\n## Where SvGrid is stronger\n\n- Svelte 5 native - reactive data binding, no imperative mount/teardown.\n- Custom cells are Svelte snippets, not HTML-string formatters.\n- Headless engine available in addition to the component.\n- @svgrid/mcp MCP server for AI assistants.\n\n## Where Tabulator is stronger\n\n- Framework-agnostic - the same library across vanilla, React, Vue, Angular.\n- Very large built-in formatter / editor catalogue.\n- Long track record and a big community.\n- No build step required to drop into a plain HTML page.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | Tabulator |\n| --- | --- | --- |\n| Svelte 5 native | Yes | No - vanilla JavaScript |\n| Reactive data binding | Yes - $state | Partial - Imperative setData() |\n| Custom cells | Yes - Snippets / components | Partial - HTML-string formatters |\n| Sort / filter / group | Yes | Yes |\n| Inline editing | Yes | Yes |\n| Tree data | Yes | Yes |\n| Virtualization | Yes | Yes |\n| Headless engine | Yes | No |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want reactive data, not imperative setData calls.\n- You want Svelte snippets for custom cells instead of HTML strings.\n- You want a headless engine option and an MCP server.\n\n## When to choose Tabulator\n\n- You need one table library across vanilla JS, React, Vue, and Angular.\n- You want a no-build-step drop-in for a plain HTML page.\n- You rely on a specific Tabulator formatter / module.\n\n## Migration effort\n\nEstimated effort: 2-4 hours per grid.\n\nWhat you give up:\n\n- One library for plain JavaScript pages as well as React, Angular and Vue.\n- Built-in download to CSV, XLSX and PDF without a paid pack; SvGrid has CSV free and Excel and PDF in @svgrid/enterprise.\n\nMigration guide: moving from Tabulator to SvGrid: https://svgrid.com/docs/help/migrating-from-tabulator/\n\n## Live example\n\nExcel Filters: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/03-excel-filters/\n\n## The bottom line\n\nTabulator is a free, feature-packed table that runs anywhere. If your app is Svelte 5, SvGrid gives you a comparable feature set with reactive data binding and snippet cells instead of imperative setData and HTML-string formatters.\n\n## Frequently asked questions\n\n### SvGrid vs Tabulator for Svelte - which is better?\n\nFor Svelte 5, SvGrid: reactive $state data and Svelte snippet cells, with no imperative mount or teardown. Tabulator wins if you need one library across vanilla JS, React, Vue, and Angular.\n\n### Is SvGrid free like Tabulator?\n\nYes - the core is MIT, like Tabulator. Only the optional @svgrid/enterprise pack is paid.\n\n### Can I reuse Tabulator formatters in SvGrid?\n\nYou re-create them as Svelte snippets via renderSnippet, which is more flexible than HTML-string formatters.\n\n### Is SvGrid a good Tabulator alternative for Svelte?\n\nTabulator is mature, MIT and framework-agnostic, and the reason people move is usually reactivity: it owns its own DOM, so Svelte state and table state have to be kept in sync by hand. SvGrid is driven by your runes directly. Tabulator still wins on breadth of built-in features and on being usable from plain HTML with no build step.\n\n## Related comparisons and docs\n\n- [SvGrid vs Grid.js](https://svgrid.com/compare/gridjs/)\n- [SvGrid vs Smart.Grid (Smart Web Components)](https://svgrid.com/compare/smart-grid/)\n- [SvGrid vs AG Grid (community + enterprise)](https://svgrid.com/compare/ag-grid/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Cell components](https://svgrid.com/docs/help/cells/cell-components/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- tabulator-readme: MIT; a vanilla JavaScript library that builds interactive tables from HTML tables, arrays or JSON, with support for React, Angular and Vue; installed as tabulator-tables (checked 12 Sep 2026) https://github.com/olifolkerd/tabulator\n"
8738
8738
  },
8739
8739
  {
8740
8740
  "slug": "compare/tanstack-table",
8741
8741
  "path": "docs/_data/comparisons/tanstack-table.json",
8742
8742
  "title": "SvGrid vs TanStack Table (Svelte)",
8743
8743
  "section": "Comparisons",
8744
- "markdown": "# SvGrid vs TanStack Table (Svelte)\n\nCanonical URL: https://svgrid.com/compare/tanstack-table/\nFacts verified: 2026-09-12\nCompetitor package: @tanstack/svelte-table\nCompetitor site: https://tanstack.com/table/latest\n\nTanStack Table is the headless table engine with adapters for React, Vue, Svelte, Solid, Qwik and Lit; you bring the markup.\n\n## Verdict\n\nThe closest peer to SvGrid's headless core. SvGrid ships a rendered component on top of the same kind of engine, in Svelte 5 runes; TanStack stays framework-agnostic and asks you to build and maintain the DOM, the virtualizer, the filter menus and the editors yourself.\n\nMigration guide: moving from TanStack Table (Svelte) to SvGrid: https://svgrid.com/docs/help/migrating-from-tanstack-table/\n\n## At a glance\n\n| Fact | SvGrid | TanStack Table (Svelte) | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 9.3.1 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 12 Oct 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Yes (peer svelte ^5.0.0) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 539,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 36.5 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte, React, Vue, Solid, Qwik, Lit | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @tanstack/svelte-table 9.3.1 entry `whole package` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nTanStack Table is the canonical headless table library: it provides the data pipeline (filter, sort, group, expand, paginate, select) and leaves the markup to you. The v9 line ships a Svelte adapter, @tanstack/svelte-table, whose peer dependency range is Svelte 5, with createTable and getters for reactive data instead of the stores the v8 adapter used.\n\nSvGrid covers the same headless surface (createSvGrid, row models, a features object) and adds a full render component with virtualization, Excel-style filters, inline editing, cell-range selection with a fill handle, grouping with aggregation, master/detail, a server-side row model and integrated charts, all in the MIT package.\n\n## Looking for a TanStack Table alternative for Svelte?\n\nPeople arrive here wanting less glue code, and that is the trade: you keep a headless engine with row models and composable features, and you stop hand-building the DOM, the virtualizer, the filter menus and the editors. If you need one table engine across React and Svelte, stay where you are; that portability is exactly what SvGrid gives up.\n\n## What they have in common\n\n- Headless-first architecture with row-model factories and a features object.\n- Composable features: sorting, filtering, grouping, pagination, expansion, selection.\n- TypeScript-first, MIT-licensed, with paid support available from the maintainers.\n- A createTable / createSvGrid entry point that takes columns, data and features.\n- Svelte 5 in both: runes in SvGrid, a Svelte 5 peer range in the TanStack adapter.\n\n## Where SvGrid is stronger\n\n- Ships <SvGrid />: a production render component with virtualization, Excel-style filters, cell-range selection, inline editing, grouping, master/detail and pagination. With TanStack Table you build all of this.\n- The headless core is written in Svelte 5 runes; the TanStack adapter wraps a framework-agnostic core.\n- An imperative API (SvGridApi) for adding rows, setting filters and mutating columns from outside the component.\n- A gallery of runnable examples that covers every feature, and per-feature docs pages with embedded demos.\n- In-grid AI helpers free in @svgrid/grid, and @svgrid/mcp so assistants fetch real example sources and can check the code they write.\n- A paid pack with Excel, PDF and print output, import, pivot tables and support.\n\n## Where TanStack Table (Svelte) is stronger\n\n- A far larger community; the At a glance table has the download counts side by side.\n- Multi-framework parity: if your shop has React, Vue and Svelte apps, the API stays consistent.\n- A smaller bundle when all you want is the engine; the At a glance table has both measured sizes.\n- Framework-specific documentation for each adapter and years of production use in large companies.\n- Enterprise Support as private consulting, and sponsor perks such as priority issue requests.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | TanStack Table (Svelte) |\n| --- | --- | --- |\n| Headless engine | Yes - createSvGrid | Yes - createTable |\n| Render component (drop-in) | Yes - <SvGrid /> | No - bring your own markup |\n| Svelte 5 | Yes - runes throughout | Yes - the adapter's peer range is Svelte 5 |\n| Row virtualization | Yes - built in | Partial - pair with @tanstack/svelte-virtual |\n| Column virtualization | Yes - built in | Partial - pair with @tanstack/svelte-virtual |\n| Excel-style filter menu | Yes - built in | No - filtering logic only; the UI is yours |\n| Inline editing | Yes - enableInlineEditing | No - bring your own |\n| Cell-range selection and clipboard | Yes - enableCellSelection, fill handle | No - bring your own |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes - createServerDataSource: page or infinite block scroll | Yes - manual sorting, filtering and pagination flags |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise | No - the engine has no server row model; the app fetches per group |\n| Master/detail and tree | Yes | Yes - expansion and subRows |\n| Grouping with aggregation | Yes - engine and rendered group rows | Yes - engine only |\n| Integrated charts | Yes - SvGridChart | No - not part of a headless engine |\n| WAI-ARIA helpers | Yes - exported, and applied by <SvGrid /> | Partial - your markup, your ARIA |\n| In-grid AI helpers and MCP server | Yes - helpers free in @svgrid/grid; @svgrid/mcp, MIT | No - not in the docs |\n| Multi-framework support | No - Svelte only | Yes - React, Vue, Svelte, Solid, Qwik, Lit |\n| Paid support tier | Yes - with @svgrid/enterprise, per developer per year | Yes - Enterprise Support (private consulting) and sponsor perks such as priority issues |\n\n## Measured performance\n\nMeasured 28 Oct 2026 with the harness in the repository (pnpm bench:compare): 100,000 rows, fastest of 3 samples per operation, each grid in a fresh page, Chromium through Playwright, headless, on win32 10.0.26200 x64, Intel(R) Core(TM) Ultra 7 255U; a developer workstation, not a dedicated bench rig. Versions: @svgrid/grid 3.0.25, @tanstack/svelte-table 9.2.4. Lower is better. Tick: 1,000 of the rows replaced per tick with a new amount, the grid sorted by amount, on the update path each grid provides, timed until the changed cells paint; 180 ticks after 10 to warm up.\n\n| Operation | SvGrid | TanStack Table (Svelte) |\n| --- | --- | --- |\n| Mount | 30.7 ms | 170.2 ms |\n| Sort, text column | 77.4 ms | 360.9 ms |\n| Sort, numeric column | 32.8 ms | 178.2 ms |\n| Filter, one column (indicative only) | 34.5 ms | 51.1 ms |\n| Scroll, p95 frame | 34.3 ms | 34.3 ms |\n| Live tick, 1,000 rows replaced, sorted by that column, p95 | 48.6 ms | 563.4 ms |\n| Ticks over one 60 Hz frame, of 180 | 115 | 180 |\n| Rows kept in the DOM (virtualization on) | 36 | 27 |\n\nThe filter row is not like for like: each grid is driven through its own single-column filter API, and the amount of work differs. The comparison guide explains the method, what is left out and why.\n\nTanStack Table renders nothing, so its adapter draws the row model through a minimal fixed-height windowed table with no virtualizer library, filter UI or editors; the numbers are the engine plus that DOM.\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want a grid that uses runes idiomatically, headless or rendered.\n- You want a drop-in <SvGrid /> with virtualization, filters, editing and selection working in one prop pass.\n- You want AI tools to give accurate answers about your grid through MCP, and to have their code checked.\n- You want paid support available without switching libraries later.\n\n## When to choose TanStack Table (Svelte)\n\n- You ship to multiple frameworks (React, Vue, Svelte) and want one API across them.\n- You explicitly want a headless library and intend to build your own render layer.\n- You want the smallest possible engine and the largest possible community.\n- A shadcn-svelte data table built on TanStack already does what you need.\n\n## Migration effort\n\nEstimated effort: About 30 minutes per read-only grid, 2-4 hours for editing-heavy grids.\n\nWhat you give up:\n\n- One table engine across React, Vue, Svelte, Solid, Qwik and Lit.\n- The smallest engine-only bundle; the At a glance table has both measured sizes.\n- The TanStack ecosystem around the table: Virtual, Query and the shadcn-svelte data table recipe.\n\nMigration guide: moving from TanStack Table (Svelte) to SvGrid: https://svgrid.com/docs/help/migrating-from-tanstack-table/\n\n## Live example\n\nHeadless Table: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/186-headless-table/\n\n## The bottom line\n\nIf you build across React, Vue and Svelte and want one headless API, TanStack Table is the safer bet. If you are committed to Svelte 5 and want virtualization, Excel-style filters, editing and selection out of the box, SvGrid gets you there with far less code, and you keep a headless core when you need it.\n\n## Frequently asked questions\n\n### Is SvGrid better than TanStack Table for Svelte?\n\nFor a Svelte 5 app, SvGrid is usually the faster path: it ships a virtualized render component with built-in filters, editing and accessibility, so there is no DOM layer to write. The TanStack Svelte adapter is Svelte 5 in its v9 line, but it is headless by design; you still build and maintain the table, virtualization and editing UI. TanStack wins if you need one API across several frameworks, or if owning that markup is the point.\n\n### Can I keep a headless setup like TanStack Table?\n\nYes. createSvGrid plus the row-model factories is a headless engine; you can drive your own markup exactly as with TanStack and adopt the SvGrid component only where you want it. The headless docs and the headless demos on this page show both.\n\n### Does the TanStack Svelte adapter support Svelte 5?\n\nYes. The @tanstack/svelte-table package on npm declares a Svelte 5 peer range in its v9 line and its examples pass data through getters rather than stores. The At a glance table shows the version and the peer range as read from the registry.\n\n### How long is a TanStack Table to SvGrid migration?\n\nAbout thirty minutes for a read-only grid and two to four hours for editing-heavy ones. The row-model concepts map almost one to one; the migration guide walks through createTable to createSvGrid, the features object and the rendering options.\n\n### Which is smaller, TanStack Table or SvGrid?\n\nThe TanStack adapter, as an engine with no rendering. The At a glance table has both packages measured the same way on the same date. Add what you would write on top of TanStack (a virtualizer, filter UI, editors) before comparing, since that is what the extra size in SvGrid is.\n\n### Is SvGrid a good TanStack Table alternative for Svelte?\n\nPeople arrive here wanting less glue code, and that is the trade: you keep a headless engine with row models and composable features, and you stop hand-building the DOM, the virtualizer, the filter menus and the editors. If you need one table engine across React and Svelte, stay where you are; that portability is exactly what SvGrid gives up.\n\n## Related comparisons and docs\n\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/)\n- [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/)\n- [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n- [Server-side data (load on demand)](https://svgrid.com/docs/help/headless/server-side/)\n- [Headless virtualization](https://svgrid.com/docs/help/headless/virtualization/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- tanstack-svelte: The Svelte adapter is headless: build your own markup from the table instance's state and APIs; createTable takes features, columns and data through a getter; adapters exist for React, Vue, Solid, Qwik, Lit and Svelte (checked 12 Sep 2026) https://tanstack.com/table/latest/docs/framework/svelte/overview\n- tanstack-npm: The latest version's peerDependencies declare svelte ^5.0.0; licence MIT (checked 12 Sep 2026) https://www.npmjs.com/package/@tanstack/svelte-table\n- tanstack-home: Sponsor-supported with gold, silver and bronze sponsors; sponsors get private discord channels, priority issue requests and direct support; an Enterprise Support tier offers private consulting and expert support; no licence for the table is sold (checked 12 Sep 2026) https://tanstack.com/table/latest\n"
8744
+ "markdown": "# SvGrid vs TanStack Table (Svelte)\n\nCanonical URL: https://svgrid.com/compare/tanstack-table/\nFacts verified: 2026-09-12\nCompetitor package: @tanstack/svelte-table\nCompetitor site: https://tanstack.com/table/latest\n\nTanStack Table is the headless table engine with adapters for React, Vue, Svelte, Solid, Qwik and Lit; you bring the markup.\n\n## Verdict\n\nThe closest peer to SvGrid's headless core. SvGrid ships a rendered component on top of the same kind of engine, in Svelte 5 runes; TanStack stays framework-agnostic and asks you to build and maintain the DOM, the virtualizer, the filter menus and the editors yourself.\n\nMigration guide: moving from TanStack Table (Svelte) to SvGrid: https://svgrid.com/docs/help/migrating-from-tanstack-table/\n\n## At a glance\n\n| Fact | SvGrid | TanStack Table (Svelte) | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 9.3.1 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 12 Oct 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Yes (peer svelte ^5.0.0) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 539,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 36.5 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte, React, Vue, Solid, Qwik, Lit | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | TanStack Table is MIT and free; tanstack.com is sponsor-supported and offers Enterprise Support as private consulting and expert support, with no licence sold for the table | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @tanstack/svelte-table 9.3.1 entry `whole package` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\nTanStack Table is the canonical headless table library: it provides the data pipeline (filter, sort, group, expand, paginate, select) and leaves the markup to you. The v9 line ships a Svelte adapter, @tanstack/svelte-table, whose peer dependency range is Svelte 5, with createTable and getters for reactive data instead of the stores the v8 adapter used.\n\nSvGrid covers the same headless surface (createSvGrid, row models, a features object) and adds a full render component with virtualization, Excel-style filters, inline editing, cell-range selection with a fill handle, grouping with aggregation, master/detail, a server-side row model and integrated charts, all in the MIT package.\n\n## Looking for a TanStack Table alternative for Svelte?\n\nPeople arrive here wanting less glue code, and that is the trade: you keep a headless engine with row models and composable features, and you stop hand-building the DOM, the virtualizer, the filter menus and the editors. If you need one table engine across React and Svelte, stay where you are; that portability is exactly what SvGrid gives up.\n\n## What they have in common\n\n- Headless-first architecture with row-model factories and a features object.\n- Composable features: sorting, filtering, grouping, pagination, expansion, selection.\n- TypeScript-first, MIT-licensed, with paid support available from the maintainers.\n- A createTable / createSvGrid entry point that takes columns, data and features.\n- Svelte 5 in both: runes in SvGrid, a Svelte 5 peer range in the TanStack adapter.\n\n## Where SvGrid is stronger\n\n- Ships <SvGrid />: a production render component with virtualization, Excel-style filters, cell-range selection, inline editing, grouping, master/detail and pagination. With TanStack Table you build all of this.\n- The headless core is written in Svelte 5 runes; the TanStack adapter wraps a framework-agnostic core.\n- An imperative API (SvGridApi) for adding rows, setting filters and mutating columns from outside the component.\n- A gallery of runnable examples that covers every feature, and per-feature docs pages with embedded demos.\n- In-grid AI helpers free in @svgrid/grid, and @svgrid/mcp so assistants fetch real example sources and can check the code they write.\n- A paid pack with Excel, PDF and print output, import, pivot tables and support.\n\n## Where TanStack Table (Svelte) is stronger\n\n- A far larger community; the At a glance table has the download counts side by side.\n- Multi-framework parity: if your shop has React, Vue and Svelte apps, the API stays consistent.\n- A smaller bundle when all you want is the engine; the At a glance table has both measured sizes.\n- Framework-specific documentation for each adapter and years of production use in large companies.\n- Enterprise Support as private consulting, and sponsor perks such as priority issue requests.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | TanStack Table (Svelte) |\n| --- | --- | --- |\n| Headless engine | Yes - createSvGrid | Yes - createTable |\n| Render component (drop-in) | Yes - <SvGrid /> | No - bring your own markup |\n| Svelte 5 | Yes - runes throughout | Yes - the adapter's peer range is Svelte 5 |\n| Row virtualization | Yes - built in | Partial - pair with @tanstack/svelte-virtual |\n| Column virtualization | Yes - built in | Partial - pair with @tanstack/svelte-virtual |\n| Excel-style filter menu | Yes - built in | No - filtering logic only; the UI is yours |\n| Inline editing | Yes - enableInlineEditing | No - bring your own |\n| Cell-range selection and clipboard | Yes - enableCellSelection, fill handle | No - bring your own |\n| Server-side data, flat (sort, filter, paging, infinite scroll) | Yes - createServerDataSource: page or infinite block scroll | Yes - manual sorting, filtering and pagination flags |\n| Server-side row model (lazy grouping, tree, pivot, transactions, select-all across unloaded rows) | Paid tier - @svgrid/enterprise | No - the engine has no server row model; the app fetches per group |\n| Master/detail and tree | Yes | Yes - expansion and subRows |\n| Grouping with aggregation | Yes - engine and rendered group rows | Yes - engine only |\n| Integrated charts | Yes - SvGridChart | No - not part of a headless engine |\n| WAI-ARIA helpers | Yes - exported, and applied by <SvGrid /> | Partial - your markup, your ARIA |\n| In-grid AI helpers and MCP server | Yes - helpers free in @svgrid/grid; @svgrid/mcp, MIT | No - not in the docs |\n| Multi-framework support | No - Svelte only | Yes - React, Vue, Svelte, Solid, Qwik, Lit |\n| Paid support tier | Yes - with @svgrid/enterprise, per developer per year | Yes - Enterprise Support (private consulting) and sponsor perks such as priority issues |\n\n## Measured performance\n\nMeasured 28 Oct 2026 with the harness in the repository (pnpm bench:compare): 100,000 rows, fastest of 3 samples per operation, each grid in a fresh page, Chromium through Playwright, headless, on win32 10.0.26200 x64, Intel(R) Core(TM) Ultra 7 255U; a developer workstation, not a dedicated bench rig. Versions: @svgrid/grid 3.0.25, @tanstack/svelte-table 9.2.4. Lower is better. Tick: 1,000 of the rows replaced per tick with a new amount, the grid sorted by amount, on the update path each grid provides, timed until the changed cells paint; 180 ticks after 10 to warm up.\n\n| Operation | SvGrid | TanStack Table (Svelte) |\n| --- | --- | --- |\n| Mount | 30.7 ms | 170.2 ms |\n| Sort, text column | 77.4 ms | 360.9 ms |\n| Sort, numeric column | 32.8 ms | 178.2 ms |\n| Filter, one column (indicative only) | 34.5 ms | 51.1 ms |\n| Scroll, p95 frame | 34.3 ms | 34.3 ms |\n| Live tick, 1,000 rows replaced, sorted by that column, p95 | 48.6 ms | 563.4 ms |\n| Ticks over one 60 Hz frame, of 180 | 115 | 180 |\n| Rows kept in the DOM (virtualization on) | 36 | 27 |\n\nThe filter row is not like for like: each grid is driven through its own single-column filter API, and the amount of work differs. The comparison guide explains the method, what is left out and why.\n\nTanStack Table renders nothing, so its adapter draws the row model through a minimal fixed-height windowed table with no virtualizer library, filter UI or editors; the numbers are the engine plus that DOM.\n\n## When to choose SvGrid\n\n- You are on Svelte 5 and want a grid that uses runes idiomatically, headless or rendered.\n- You want a drop-in <SvGrid /> with virtualization, filters, editing and selection working in one prop pass.\n- You want AI tools to give accurate answers about your grid through MCP, and to have their code checked.\n- You want paid support available without switching libraries later.\n\n## When to choose TanStack Table (Svelte)\n\n- You ship to multiple frameworks (React, Vue, Svelte) and want one API across them.\n- You explicitly want a headless library and intend to build your own render layer.\n- You want the smallest possible engine and the largest possible community.\n- A shadcn-svelte data table built on TanStack already does what you need.\n\n## Migration effort\n\nEstimated effort: About 30 minutes per read-only grid, 2-4 hours for editing-heavy grids.\n\nWhat you give up:\n\n- One table engine across React, Vue, Svelte, Solid, Qwik and Lit.\n- The smallest engine-only bundle; the At a glance table has both measured sizes.\n- The TanStack ecosystem around the table: Virtual, Query and the shadcn-svelte data table recipe.\n\nMigration guide: moving from TanStack Table (Svelte) to SvGrid: https://svgrid.com/docs/help/migrating-from-tanstack-table/\n\n## Live example\n\nHeadless Table: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/186-headless-table/\n\n## The bottom line\n\nIf you build across React, Vue and Svelte and want one headless API, TanStack Table is the safer bet. If you are committed to Svelte 5 and want virtualization, Excel-style filters, editing and selection out of the box, SvGrid gets you there with far less code, and you keep a headless core when you need it.\n\n## Frequently asked questions\n\n### Is SvGrid better than TanStack Table for Svelte?\n\nFor a Svelte 5 app, SvGrid is usually the faster path: it ships a virtualized render component with built-in filters, editing and accessibility, so there is no DOM layer to write. The TanStack Svelte adapter is Svelte 5 in its v9 line, but it is headless by design; you still build and maintain the table, virtualization and editing UI. TanStack wins if you need one API across several frameworks, or if owning that markup is the point.\n\n### Can I keep a headless setup like TanStack Table?\n\nYes. createSvGrid plus the row-model factories is a headless engine; you can drive your own markup exactly as with TanStack and adopt the SvGrid component only where you want it. The headless docs and the headless demos on this page show both.\n\n### Does the TanStack Svelte adapter support Svelte 5?\n\nYes. The @tanstack/svelte-table package on npm declares a Svelte 5 peer range in its v9 line and its examples pass data through getters rather than stores. The At a glance table shows the version and the peer range as read from the registry.\n\n### How long is a TanStack Table to SvGrid migration?\n\nAbout thirty minutes for a read-only grid and two to four hours for editing-heavy ones. The row-model concepts map almost one to one; the migration guide walks through createTable to createSvGrid, the features object and the rendering options.\n\n### Which is smaller, TanStack Table or SvGrid?\n\nThe TanStack adapter, as an engine with no rendering. The At a glance table has both packages measured the same way on the same date. Add what you would write on top of TanStack (a virtualizer, filter UI, editors) before comparing, since that is what the extra size in SvGrid is.\n\n### Is SvGrid a good TanStack Table alternative for Svelte?\n\nPeople arrive here wanting less glue code, and that is the trade: you keep a headless engine with row models and composable features, and you stop hand-building the DOM, the virtualizer, the filter menus and the editors. If you need one table engine across React and Svelte, stay where you are; that portability is exactly what SvGrid gives up.\n\n## Related comparisons and docs\n\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/)\n- [SvGrid vs SVAR Svelte DataGrid](https://svgrid.com/compare/svar-svelte-datagrid/)\n- [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n- [Server-side data (load on demand)](https://svgrid.com/docs/help/headless/server-side/)\n- [Headless virtualization](https://svgrid.com/docs/help/headless/virtualization/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- tanstack-svelte: The Svelte adapter is headless: build your own markup from the table instance's state and APIs; createTable takes features, columns and data through a getter; adapters exist for React, Vue, Solid, Qwik, Lit and Svelte (checked 12 Sep 2026) https://tanstack.com/table/latest/docs/framework/svelte/overview\n- tanstack-npm: The latest version's peerDependencies declare svelte ^5.0.0; licence MIT (checked 12 Sep 2026) https://www.npmjs.com/package/@tanstack/svelte-table\n- tanstack-home: Sponsor-supported with gold, silver and bronze sponsors; sponsors get private discord channels, priority issue requests and direct support; an Enterprise Support tier offers private consulting and expert support; no licence for the table is sold (checked 12 Sep 2026) https://tanstack.com/table/latest\n"
8745
8745
  },
8746
8746
  {
8747
8747
  "slug": "compare/vincjo-datatables",
8748
8748
  "path": "docs/_data/comparisons/vincjo-datatables.json",
8749
8749
  "title": "SvGrid vs @vincjo/datatables",
8750
8750
  "section": "Comparisons",
8751
- "markdown": "# SvGrid vs @vincjo/datatables\n\nCanonical URL: https://svgrid.com/compare/vincjo-datatables/\nFacts verified: 2026-09-12\nCompetitor package: @vincjo/datatables\nCompetitor site: https://github.com/vincjo/datatables\n\n@vincjo/datatables is a lightweight, headless Svelte datatable helper for sorting, filtering, and paginating your own markup.\n\n## Verdict\n\n@vincjo/datatables is a great minimal pick when you want to keep your own table markup and just need sort / filter / paginate. SvGrid is the fuller grid - virtualization, Excel filters, editing, a render component - for heavier apps.\n\nMigration guide: moving from @vincjo/datatables to SvGrid: https://svgrid.com/docs/help/migrating-from-vincjo-datatables/\n\n## At a glance\n\n| Fact | SvGrid | @vincjo/datatables | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 2.8.1 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 4 Jun 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Yes (peer svelte ^5.56.1) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 107,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.9 KB JS + 11.3 KB CSS | 10.2 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | @vincjo/datatables is MIT on npm and free; no licence is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.33 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @vincjo/datatables 2.8.1 entry `whole package` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\n@vincjo/datatables gives you a small, ergonomic handler that wraps your data with reactive sorting, filtering, and pagination while you keep full control of the markup. It is MIT-licensed and popular for its simplicity.\n\nSvGrid scales further up the curve: it keeps a headless engine but also ships a virtualized render component, an Excel-style filter menu, inline editing, cell selection, and an imperative API - for when a hand-rolled table stops being enough.\n\n## Looking for a @vincjo/datatables alternative for Svelte?\n\nIt is a helper for your own markup rather than a grid, which is exactly right until you need virtualization, inline editing or column pinning and start building them yourself. Moving to SvGrid keeps the headless option and adds those. If your table is a few hundred rows of your own HTML, staying put is the smaller dependency.\n\n## What they have in common\n\n- Svelte-native, not a port.\n- Headless-friendly: you can drive your own markup.\n- MIT-licensed.\n- Client- and server-side data modes.\n\n## Where SvGrid is stronger\n\n- Built-in virtualized render component - no need to author the table.\n- Excel-style filter menu, inline editing, and cell-range selection.\n- Row + column virtualization for large datasets.\n- Imperative SvGridApi + @svgrid/mcp MCP server.\n- Optional Enterprise pack for export, import, pivot, and AI.\n\n## Where @vincjo/datatables is stronger\n\n- Much smaller surface and bundle if all you need is sort / filter / paginate.\n- You keep 100% control of the table markup with zero render opinions.\n- Faster to learn for a simple table.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | @vincjo/datatables |\n| --- | --- | --- |\n| Svelte native | Yes | Yes - Svelte 5 peer range |\n| Render component | Yes - <SvGrid /> | No - headless, you render the markup |\n| Virtualization | Yes | No |\n| Excel-style filters | Yes | Partial - Basic filtering |\n| Inline editing | Yes | No - BYO |\n| Cell selection + copy | Yes | No - BYO |\n| Grouping / tree / master-detail | Yes | No |\n| Footprint for a simple table | Partial - Larger | Yes - Tiny |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You need virtualization, editing, or Excel-style filtering.\n- You want a render component instead of hand-authoring the table.\n- You expect the grid to grow into grouping, tree, or master/detail.\n\n## When to choose @vincjo/datatables\n\n- You only need sort / filter / paginate over a small dataset.\n- You want to keep full control of your own table markup.\n- You want the smallest possible bundle for a simple table.\n\n## Migration effort\n\nEstimated effort: 30 minutes to 2 hours per table.\n\nWhat you give up:\n\n- A tiny dependency-free helper when a sortable, filterable list is all you need.\n- Full control of the markup without opting into a headless mode.\n\nMigration guide: moving from @vincjo/datatables to SvGrid: https://svgrid.com/docs/help/migrating-from-vincjo-datatables/\n\n## Live example\n\nHeadless Table: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/186-headless-table/\n\n## The bottom line\n\nFor a small sort/filter/paginate list you keep full control over, @vincjo/datatables is delightfully light. The moment you need virtualization, Excel-style filters, editing, or grouping, SvGrid is the grid that already has them.\n\n## Frequently asked questions\n\n### SvGrid vs @vincjo/datatables - which to pick?\n\nPick @vincjo/datatables for a small, markup-controlled table; pick SvGrid when you need virtualization, editing, filters, or grouping. Both are MIT and Svelte-native.\n\n### Is @vincjo/datatables lighter than SvGrid?\n\nYes, for a simple table it is smaller. SvGrid carries more because it ships a virtualized render component and a full feature set.\n\n### Can SvGrid be headless like @vincjo/datatables?\n\nYes - its headless core lets you keep your own markup, the same way @vincjo/datatables does.\n\n### Is SvGrid a good @vincjo/datatables alternative for Svelte?\n\nIt is a helper for your own markup rather than a grid, which is exactly right until you need virtualization, inline editing or column pinning and start building them yourself. Moving to SvGrid keeps the headless option and adds those. If your table is a few hundred rows of your own HTML, staying put is the smaller dependency.\n\n## Related comparisons and docs\n\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/)\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- vincjo-docs: A headless, SSR-friendly toolkit with no dependency: sort, filter, search and pagination in client-side and server-side modes; you render the markup (checked 12 Sep 2026) https://vincjo.fr/datatables/\n- vincjo-npm: MIT; the latest version declares a Svelte 5 peer range (checked 12 Sep 2026) https://www.npmjs.com/package/@vincjo/datatables\n"
8751
+ "markdown": "# SvGrid vs @vincjo/datatables\n\nCanonical URL: https://svgrid.com/compare/vincjo-datatables/\nFacts verified: 2026-09-12\nCompetitor package: @vincjo/datatables\nCompetitor site: https://github.com/vincjo/datatables\n\n@vincjo/datatables is a lightweight, headless Svelte datatable helper for sorting, filtering, and paginating your own markup.\n\n## Verdict\n\n@vincjo/datatables is a great minimal pick when you want to keep your own table markup and just need sort / filter / paginate. SvGrid is the fuller grid - virtualization, Excel filters, editing, a render component - for heavier apps.\n\nMigration guide: moving from @vincjo/datatables to SvGrid: https://svgrid.com/docs/help/migrating-from-vincjo-datatables/\n\n## At a glance\n\n| Fact | SvGrid | @vincjo/datatables | Checked |\n| --- | --- | --- | --- |\n| Latest version | 3.0.32 | 2.8.1 | 9 Nov 2026 |\n| Licence field on npm | MIT | MIT | 9 Nov 2026 |\n| Last published | 3 Nov 2026 | 4 Jun 2026 | 9 Nov 2026 |\n| Svelte 5 support | Native (Svelte 5 runes) | Yes (peer svelte ^5.56.1) | 9 Nov 2026 |\n| npm downloads, 30 days to 8 Nov 2026 | 44,600 | 107,000 | 9 Nov 2026 |\n| Bundle, minified + gzip | 99.6 KB JS + 11.4 KB CSS | 10.2 KB JS, no separate stylesheet, svelte external | 9 Nov 2026 |\n| Engine alone, minified + gzip | 3.2 KB JS, no separate stylesheet (`createSvGrid` from `@svgrid/grid/core`) | the row above; the package is the engine | 10 Nov 2026 |\n| Frameworks | Svelte 5 | Svelte | 12 Sep 2026 |\n| Pricing | MIT core; @svgrid/enterprise from $299 per developer per year (Indie, under $1M revenue), $599 for Enterprise | @vincjo/datatables is MIT on npm and free; no licence is sold | 12 Sep 2026 |\n\nBundle sizes: each package built alone with Vite in library mode, minified, gzip level 9, Svelte kept external where a package uses it (it is a peer dependency). SvGrid 3.0.34 measured 10 Nov 2026 with packages/grid/scripts/measure-size.mjs; @vincjo/datatables 2.8.1 entry `whole package` measured 9 Nov 2026 with packages/grid/scripts/measure-competitor-bundles.mjs.\n\n## Overview\n\n@vincjo/datatables gives you a small, ergonomic handler that wraps your data with reactive sorting, filtering, and pagination while you keep full control of the markup. It is MIT-licensed and popular for its simplicity.\n\nSvGrid scales further up the curve: it keeps a headless engine but also ships a virtualized render component, an Excel-style filter menu, inline editing, cell selection, and an imperative API - for when a hand-rolled table stops being enough.\n\n## Looking for a @vincjo/datatables alternative for Svelte?\n\nIt is a helper for your own markup rather than a grid, which is exactly right until you need virtualization, inline editing or column pinning and start building them yourself. Moving to SvGrid keeps the headless option and adds those. If your table is a few hundred rows of your own HTML, staying put is the smaller dependency.\n\n## What they have in common\n\n- Svelte-native, not a port.\n- Headless-friendly: you can drive your own markup.\n- MIT-licensed.\n- Client- and server-side data modes.\n\n## Where SvGrid is stronger\n\n- Built-in virtualized render component - no need to author the table.\n- Excel-style filter menu, inline editing, and cell-range selection.\n- Row + column virtualization for large datasets.\n- Imperative SvGridApi + @svgrid/mcp MCP server.\n- Optional Enterprise pack for export, import, pivot, and AI.\n\n## Where @vincjo/datatables is stronger\n\n- Much smaller surface and bundle if all you need is sort / filter / paginate.\n- You keep 100% control of the table markup with zero render opinions.\n- Faster to learn for a simple table.\n\n## Feature by feature\n\nYes: shipped in the package. Partial: needs extra work or a companion package. Paid tier: only in a paid edition. No: not available. n/a: does not apply.\n\n| Feature | SvGrid | @vincjo/datatables |\n| --- | --- | --- |\n| Svelte native | Yes | Yes - Svelte 5 peer range |\n| Render component | Yes - <SvGrid /> | No - headless, you render the markup |\n| Virtualization | Yes | No |\n| Excel-style filters | Yes | Partial - Basic filtering |\n| Inline editing | Yes | No - BYO |\n| Cell selection + copy | Yes | No - BYO |\n| Grouping / tree / master-detail | Yes | No |\n| Footprint for a simple table | Partial - Larger | Yes - Tiny |\n| AI / MCP integration | Yes | No |\n\n## When to choose SvGrid\n\n- You need virtualization, editing, or Excel-style filtering.\n- You want a render component instead of hand-authoring the table.\n- You expect the grid to grow into grouping, tree, or master/detail.\n\n## When to choose @vincjo/datatables\n\n- You only need sort / filter / paginate over a small dataset.\n- You want to keep full control of your own table markup.\n- You want the smallest possible bundle for a simple table.\n\n## Migration effort\n\nEstimated effort: 30 minutes to 2 hours per table.\n\nWhat you give up:\n\n- A tiny dependency-free helper when a sortable, filterable list is all you need.\n- Full control of the markup without opting into a headless mode.\n\nMigration guide: moving from @vincjo/datatables to SvGrid: https://svgrid.com/docs/help/migrating-from-vincjo-datatables/\n\n## Live example\n\nHeadless Table: open it in the gallery for the source and the full-size version. https://svgrid.com/demos/186-headless-table/\n\n## The bottom line\n\nFor a small sort/filter/paginate list you keep full control over, @vincjo/datatables is delightfully light. The moment you need virtualization, Excel-style filters, editing, or grouping, SvGrid is the grid that already has them.\n\n## Frequently asked questions\n\n### SvGrid vs @vincjo/datatables - which to pick?\n\nPick @vincjo/datatables for a small, markup-controlled table; pick SvGrid when you need virtualization, editing, filters, or grouping. Both are MIT and Svelte-native.\n\n### Is @vincjo/datatables lighter than SvGrid?\n\nYes, for a simple table it is smaller. SvGrid carries more because it ships a virtualized render component and a full feature set.\n\n### Can SvGrid be headless like @vincjo/datatables?\n\nYes - its headless core lets you keep your own markup, the same way @vincjo/datatables does.\n\n### Is SvGrid a good @vincjo/datatables alternative for Svelte?\n\nIt is a helper for your own markup rather than a grid, which is exactly right until you need virtualization, inline editing or column pinning and start building them yourself. Moving to SvGrid keeps the headless option and adds those. If your table is a few hundred rows of your own HTML, staying put is the smaller dependency.\n\n## Related comparisons and docs\n\n- [SvGrid vs svelte-headless-table](https://svgrid.com/compare/svelte-headless-table/)\n- [SvGrid vs TanStack Table (Svelte)](https://svgrid.com/compare/tanstack-table/)\n- [SvGrid vs Flowbite / Skeleton / shadcn-svelte tables](https://svgrid.com/compare/svelte-ui-kit-tables/)\n- [Comparison: SvGrid vs AG Grid vs TanStack Table](https://svgrid.com/docs/help/comparison/)\n- [Build a table from scratch](https://svgrid.com/docs/help/headless/build-a-table/)\n\n## Sources and last verified\n\nEvery claim on this page was last checked on 12 Sep 2026. Versions, licences and download counts come from the npm registry; feature and pricing statements from the pages listed below.\n\n- vincjo-docs: A headless, SSR-friendly toolkit with no dependency: sort, filter, search and pagination in client-side and server-side modes; you render the markup (checked 12 Sep 2026) https://vincjo.fr/datatables/\n- vincjo-npm: MIT; the latest version declares a Svelte 5 peer range (checked 12 Sep 2026) https://www.npmjs.com/package/@vincjo/datatables\n"
8752
8752
  }
8753
8753
  ];
8754
8754
  export const apiReference = {
@@ -8825,8 +8825,8 @@ export const apiReference = {
8825
8825
  ]
8826
8826
  };
8827
8827
  export const apiSurface = {
8828
- "gridVersion": "3.0.34",
8829
- "enterpriseVersion": "3.1.38",
8828
+ "gridVersion": "3.0.35",
8829
+ "enterpriseVersion": "3.1.40",
8830
8830
  "grid": {
8831
8831
  "values": [
8832
8832
  "ALL_FILTER_OPERATORS",