@oar-sh/oar 0.9.24 → 0.9.26

package/.github/extensions/web-relay/skills/reasoning-stream.mjs

@@ -168,7 +168,9 @@ export function createReasoningStreamHandlers({
     const reasoningId = reasoningIdFor(event);
     const deltaContent = String(event?.data?.deltaContent ?? event?.data?.delta ?? "");
     if (!deltaContent) return;
-    const accumulated = (reasoningAccum.get(reasoningId) || "") + deltaContent;
+    // Capped before the gate: past the cap the text stops changing, so the
+    // same 16 KiB is not posted again and the accumulator stops growing.
+    const accumulated = capThought((reasoningAccum.get(reasoningId) || "") + deltaContent);
     reasoningAccum.set(reasoningId, accumulated);
     const lastSent = reasoningLastSent.get(reasoningId) || "";
     if (!shouldEmitDelta(accumulated, lastSent)) return;

package/CHANGELOG.md

@@ -3,6 +3,290 @@
 All notable changes to OAR are documented here. The format follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versions follow semver.
 
+## [0.9.26] — 2026-11-11
+
+Highlights: formulas from their delimiters only, and Settings sorted by
+where they live. Replies render a formula only when it is written as
+`$…$`, `\(…\)`, `$$…$$`, `\[…\]` or a fenced `math` block, so an id in
+parentheses, a Windows path, a price or a shell variable stays text, a
+formula KaTeX cannot read is shown as written instead of in red, and
+the agent is told the notation that renders, except in Claude Cloud
+chats and on the Copilot Extension engine. KaTeX moves to 0.19.1.
+Theme, text scaling and the live-turn settings get an Appearance tab of
+their own, and the relay name and the two restart settings move to
+Relays, where an empty name shows the host name the relay goes by and a
+rename made on another device shows at once. In Claude, Claude API and
+DeepSeek chats a plan card comes only from a plan the agent hands over,
+and a plan card shows up to 200,000 characters of the plan instead of
+12,000. Archive and delete no longer fail on Linux and macOS for a
+worker that has already ended, a session an agent on a paired relay
+started can always report back to it, and the tmux console draws a real
+terminal.
+
+### Added
+
+- **Fenced `math` blocks.** A code block with the language `math` is a
+  displayed formula once it is closed.
+- **Agents are told how to write formulas.** Once per session, the agent
+  of a Claude, Claude API, DeepSeek, Cursor or Grok chat, and of a
+  Copilot or OpenAI chat on the SDK engine, learns that `$…$` in a
+  sentence and `$$…$$` on lines of their own render (`\(…\)` and `\[…\]`
+  too), that TeX in bare parentheses or brackets does not, and that a
+  dollar amount stays text. Claude Cloud chats and the Copilot Extension
+  engine do not get the note. The prompt a transcript shows leaves the
+  note out.
+
+### Changed
+
+- **Settings: an Appearance tab, and the relay's own settings on
+  Relays.** Theme, Text scaling, Expand thoughts while an agent works,
+  Tool lines shown while an agent works, Stream view and Show Suspend
+  host action move from General to the new Appearance tab right after
+  it, under one line that says they are kept in this browser; Jump to
+  the end when I send a message and Messages kept per conversation stay
+  in General's Reading block. Relay name moves to the top of Relays,
+  above Public address, and After a relay restart, tell interrupted
+  sessions to carry on and Ask after a queued relay restart has waited
+  follow right below Public address; what stays of General's "This
+  relay" block (Default CWD, Autostart, the Windows helper) is headed
+  "This computer". Every setting keeps its value and default, and a
+  remembered tab or a link to General still opens General.
+- **Formulas are read from their delimiters only.** `$…$` and `\(…\)`
+  inside a sentence, `$$…$$` and `\[…\]` as a display, and a fenced
+  `math` block; nothing is guessed from TeX in bare parentheses or
+  brackets any more. On one line `\[ … \]` needs a space inside both
+  ends, so `\[1\]` and `\[x\]` are escaped brackets. A `$` opens a
+  formula only where maths is written that way, so prices (`$5 and $9`,
+  `$5/$10`), shell variables (`$HOME/$PATH`) and Windows paths stay
+  text, and `\$` is always a dollar sign. Code spans and code blocks
+  never hold a formula. A formula KaTeX cannot parse is shown as
+  written, in the normal text colour, and a percent sign in a one-line
+  formula is shown. A message over 100,000 characters renders no
+  formulas. The rules are in
+  [Formulas in replies](docs/web-ui.md#formulas-in-replies).
+- **KaTeX 0.19.1.** The page loads KaTeX 0.19.1 (was 0.16.22) and its
+  chemistry extension from jsDelivr, as before, now with integrity hashes;
+  its auto-render script is no longer loaded. KaTeX is not in the npm
+  package: a browser that cannot reach jsDelivr shows every formula as
+  written.
+- **A plan card comes from a plan hand-over only** in Claude, Claude API
+  and DeepSeek chats. A plan-mode reply with two or more list lines no
+  longer gets a "Plan ready for review" card, so a session that stays in
+  plan mode no longer puts the buttons under every status reply. When the
+  plan of a hand-over cannot be read, the reply stands in for it. Cursor,
+  Grok, Copilot and OpenAI (BYOK) chats keep the card for such replies.
+
+### Fixed
+
+- **The whole plan on the plan card.** The card cut every plan at 12,000
+  characters, without a mark. It now keeps up to 200,000; a longer plan
+  is cut, at a line break where it can, and ends with a line that says
+  how much is missing and, where the agent named it, the file that holds
+  the whole plan. A cut inside a code block closes the block first, also
+  in a plan file with Windows line ends.
+- **Archive and delete no longer fail for a worker that has ended**
+  (Linux and macOS). Archiving or deleting an idle conversation could
+  answer "Couldn't stop this conversation's session" again and again:
+  the relay counted the old process id of a worker that had ended, by
+  then another program's, and processes that were ending by themselves.
+  Those no longer count, and a process the relay may not signal no longer
+  keeps it from stopping the others. A worker that really survives the
+  stop still leaves the conversation as it was.
+- **A session an agent on a paired relay started can always report
+  back.** Its send to the session that started it was refused with
+  `REMOTE_RELAY_HOP_LIMIT` when its turn also acted on a prompt from a
+  session on its own relay. Another session on that relay, a third relay
+  and `create_session` are still held to the limit.
+- **The relay name in Settings.** An empty Relay name showed "OAR" as its
+  placeholder; it now shows the host name, the name paired relays know
+  the relay by then. A rename made on another device shows in an open
+  Settings without reopening it.
+- **The tmux console draws a terminal.** 🖥️ Inspect tmux console asked
+  jsDelivr and unpkg for xterm.js versions that were never published, so
+  it always fell back to plain text. It now loads `@xterm/xterm` 5.5.0
+  and `@xterm/addon-fit` 0.10.0 from jsDelivr, with integrity hashes,
+  and only the two scripts from unpkg when jsDelivr fails; a browser
+  that reaches neither still gets the plain text.
+- **A long line in a code block no longer runs under its Copy button**,
+  which has a strip of its own above the first line.
+
+### Upgrade notes
+
+- **No migration.** As before, the relay copies its database to the
+  backups folder before the first start of a new version; nothing else
+  waits.
+- **Sessions still on 0.9.25's worker wait one worker start.** A session
+  whose worker still runs the old code is given a new worker between
+  turns, as in 0.9.25: a Claude, Claude API, DeepSeek or Cursor session
+  outside a git checkout before its next message, any other once its
+  agent has been idle for about 10 minutes. Its agent learns the formula
+  notation with the new worker, and Claude, Claude API and DeepSeek
+  chats get the new plan-card rule with it.
+- **Your settings stay as they were.** The Appearance settings are kept
+  in the browser under the same keys, and the relay's own settings in its
+  database; only their place in Settings changes.
+- **Older replies follow the new formula rules** when they are shown
+  again: a parenthesis 0.9.25 drew as a formula is text now, TeX
+  written in bare parentheses shows as written, and a one-line
+  `\[x^2\]` without a space inside its ends is read as escaped brackets.
+- **KaTeX and xterm.js load in the browser,** not from the relay: from
+  jsDelivr, and the two xterm.js scripts, not its stylesheet, also from
+  unpkg; neither is in the npm package. A browser without access to them
+  shows formulas as written and the tmux console as plain text, whatever
+  the relay host can reach.
+- **Mixed versions work together.** The report-back rule is checked by
+  the relay the started session runs on: that one needs 0.9.26, the relay
+  of the session that started it may run 0.9.25.
+- **Going back to 0.9.25** needs no database step. Install it with
+  `oar update --to 0.9.25` or `npm install -g @oar-sh/oar@0.9.25` (in a
+  git checkout: `git checkout v0.9.25`, then a relay restart). The
+  settings go back to their old places with their values; a browser that
+  last had Appearance open opens General. Formulas follow 0.9.25's rules
+  again, Claude plan-mode replies with list lines get cards again, and
+  plan cards made on 0.9.26 keep their whole plan.
+
+## [0.9.25] — 2026-11-09
+
+Highlights: a media gallery, and agents that make images. Every picture,
+video and file of a session is now in one place, the Media tab of the 🧠
+modal, and 🖼️ Media in the ︙ menu shows those of every session of the
+relay, with filters and a search; the file viewer moves through the whole
+gallery. Agents drive OpenAI Image sessions through `relay_tool`: they send
+pictures to edit and set the size, quality, count, format and background.
+While gpt-image works, the reply shows partial previews; the composer gets
+a custom size, a count, a format, a background and presets, and the
+gpt-image-2.5 and 2.6 models take xhigh and max quality. Every image turn
+shows what it cost, a daily cap of USD 5 per relay unless you change it
+keeps image spend in check, and a failed image turn costs at most two calls
+instead of up to five. The relay's memory no longer grows while several
+agents stream long thinking at once.
+
+### Added
+
+- **Media gallery.** The 🧠 modal has two tabs, Context (where it opens)
+  and Media: the session's pictures and videos as a square grid under month
+  headers, three columns on a phone, with a Files tab for attached PDFs,
+  audio and documents. **🖼️ Media** in the ︙ menu shows the media of every
+  session of the relay, with filters (All, Mine, Agent, Generated, a
+  provider, a session, Hide archived) and a search. It collects uploads,
+  `@file:` pictures, generated images and the pictures agents put in their
+  replies; a picture shown in several messages appears once, at its newest
+  message, and a deleted local file is left out. A tap opens the file
+  viewer, which now moves through the whole gallery (it loads more at the
+  end) and offers **Open in session** and **Copy path / link**. The pictures
+  on screen that could not load while the relay was away load again once it
+  is back; one scrolled out of view waits for the next reconnect.
+  Owner only: not in shared views, and no agent reads it through
+  `relay_tool`, on this relay or a paired one.
+- **Media index.** The relay keeps the gallery's index in its database
+  (migration 0039) and fills it for older messages in the background after
+  the upgrade, in short slices.
+- **Agents drive OpenAI Image sessions.** `relay_tool`'s `send` and
+  `create_session` take `images`, up to 4 absolute paths of PNG, JPEG or
+  WebP files on the target relay, and `image`, the turn's size (any W×H
+  within the model's limits), quality, count (1 to 4), output format and
+  background; `update_session` with `image` sets a session's own defaults
+  (migration 0040). The target relay takes only files inside the session's
+  working directory or images one of its sessions generated, follows links
+  to the real file, and checks each file's first bytes. Replies list the
+  pictures as `images` with their paths there, and a reply without text
+  reads like `1 image · 1536x1024 · gpt-image-2`. A paired relay must run
+  this version (`imageInputs` in `relay_info`); an older one is refused
+  before anything is sent. A new image session without a model starts on
+  gpt-image-2.5-flare (else its dated snapshot, else gpt-image-2, else the
+  first gpt-image model listed), and is refused when none is listed.
+
+- **Image previews.** While a gpt-image model makes an image, the reply's
+  bubble shows a timer with the size and quality, then two partial previews
+  as OpenAI streams them, then the final image. Previews reach your open
+  clients only, never the transcript, shared views, the media gallery or
+  agents, and their files go when the turn ends.
+- **Image composer.** An image chat's composer has a Custom… size (any
+  W×H the model takes, checked as you type), a Count (1 to 4), a Format
+  and a Background for gpt-image models.
+
+- **What image turns cost.** Every try of an OpenAI image turn is priced
+  as it ends, from the usage OpenAI reports and its published rates, and
+  estimated where OpenAI reports no usage for a try that may have been
+  billed; a model the relay has no price for counts at the dearest listed
+  price. The reply line ends with the cost, the chat menu shows what a
+  chat's images cost, and Check Usage has an OpenAI images card.
+- **A daily cap for OpenAI images.** USD 5 per relay and day unless you
+  change it (Settings → Providers → OpenAI; 0 is no cap). A request that
+  would pass it is not sent, and once it is reached new image messages are
+  refused, from you and from agents alike; estimates count toward it. A
+  warning shows from 80 %. Allow for today lifts it until midnight.
+- **Image presets.** Five built-in bundles of size, quality, format and
+  background, and your own (Save as preset… in the composer); agents name
+  one as image.preset.
+- **xhigh and max image quality.** The gpt-image-2.5 and 2.6 models take
+  the two higher qualities, in the composer and for agents; other models
+  keep auto, low, medium and high. Sizes for gpt-image-2 and later need at
+  least 655,360 pixels (1024x640), as OpenAI requires.
+
+### Changed
+
+- **OpenAI Image turns.** All attached images go to the edit, in order,
+  not only the first, and the format and background reach OpenAI. Quality
+  is checked per model family before the call (gpt-image: auto, low,
+  medium, high). OpenAI gets 3 minutes for a new image and 5 for
+  an edit (`openaiImageGenerateTimeoutMs`, `openaiImageEditTimeoutMs`).
+- **Copilot hides image models.** Copilot's `gpt-image-*` and `dall-e-*`
+  ids, which its runtime cannot serve, are left out of its model list; a
+  Copilot chat on one is refused with a pointer to OpenAI Image.
+
+### Fixed
+
+- **An image turn costs at most two calls.** A timeout, a server error or
+  a broken connection is tried once more, then the turn ends with a note
+  that says whether the request may have been billed; before, the turn went
+  back into the queue up to five times and ended as "Relay timeout after 5
+  attempts", each attempt a new billed request. When the second try works,
+  the reply notes that the first may have been billed as well. A relay
+  restart during the call no longer repeats it either.
+- **The relay's own image route is local only.** A paired relay's token can
+  no longer call it directly, past the queue.
+- **An agent's prompt header no longer reaches the image model.**
+- **Relay memory no longer grows while several agents stream long thinking
+  at the same time.** A few sessions thinking at length at once could make
+  the relay hold more and more memory until it ran out. Thinking still shows
+  live; a slow connection may skip some of the updates in between, and the
+  whole thought arrives when it ends and after a reconnect, as before.
+
+### Upgrade notes
+
+- **The first start takes a little longer on a big database.** Before it
+  listens, the relay copies its database to the backups folder, as before,
+  and migration 0039 creates the media index and queues every message for
+  it once: about 0.2 s per 200,000 messages on a small server. Migration
+  0040 adds two columns. Nothing else waits.
+- **The gallery fills in the background.** After the start the relay
+  indexes the older messages in slices of about 20 ms every half second,
+  in no particular order: about 3 minutes per 200,000 messages on a small
+  server, longer on a phone or with long messages. Until it ends, the
+  gallery shows what is indexed and "Indexing older messages… (N left)". A
+  restart picks up where it stopped.
+- **The daily cap for OpenAI images is on after the update**: USD 5 per
+  relay and day, in Settings → Providers → OpenAI (0 is no cap). Without an
+  OpenAI key nothing changes, and Check Usage shows no OpenAI images card
+  until something was booked in the month.
+- **Sessions still on 0.9.24's worker wait one worker start.** A session
+  whose worker still runs the old code is given a new worker before its
+  next message, as in 0.9.24. A Claude session's agent gets `images` and
+  `image` in `relay_tool` with its new worker.
+- **Mixed versions work together**, except for images across relays: a
+  0.9.25 relay refuses to send `images` or `image` to a paired 0.9.24 relay
+  and sends nothing, and a 0.9.24 relay cannot send them.
+- **Going back to 0.9.24** needs no database step. Install it with
+  `npm install -g @oar-sh/oar@0.9.24` (in a git checkout:
+  `git checkout v0.9.24`, then a relay restart). 0.9.24 ignores the media
+  index and the image defaults and has no daily image cap, so image turns
+  are not limited there. The index's triggers stay in the database and
+  queue the messages written on 0.9.24, which a later 0.9.25 indexes. To
+  remove the index, drop its two triggers first, then its three tables, as
+  `server/migrations/0039-media-index.mjs` describes; a table dropped
+  before its trigger breaks every new message.
+
 ## [0.9.24] — 2026-11-08
 
 Highlights: a much smaller install. The Claude and Cursor agent SDKs leave

package/README.md

@@ -45,29 +45,29 @@ flowchart LR
 <table>
 <tr>
 <td width="42%" rowspan="4" valign="top">
-<a href="https://oar.sh/assets/shots/cfmail-mobile.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-mobile.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/cfmail-mobile.png?v=0.9.24" alt="OAR on a phone: a finished agent turn in a conversation, with the composer below"></picture></a>
+<a href="https://oar.sh/assets/shots/cfmail-mobile.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-mobile.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/cfmail-mobile.png?v=0.9.26" alt="OAR on a phone: a finished agent turn in a conversation, with the composer below"></picture></a>
 <br><sub>On a phone: the same conversations, installable as an app.</sub>
 </td>
 <td width="58%" valign="top">
-<a href="https://oar.sh/assets/shots/cfmail-agents.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-agents.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/cfmail-agents.png?v=0.9.24" alt="Background subagents running in parallel in the task panel, each with live token counts, elapsed time, and its own Stop button"></picture></a>
+<a href="https://oar.sh/assets/shots/cfmail-agents.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-agents.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/cfmail-agents.png?v=0.9.26" alt="Background subagents running in parallel in the task panel, each with live token counts, elapsed time, and its own Stop button"></picture></a>
 <br><sub>Agents working in parallel, each with live token counts and its own Stop.</sub>
 </td>
 </tr>
 <tr>
 <td valign="top">
-<a href="https://oar.sh/assets/shots/cfmail-question.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-question.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/cfmail-question.png?v=0.9.24" alt="A question card in the middle of a turn: the agent offers several choices and a free-text reply"></picture></a>
+<a href="https://oar.sh/assets/shots/cfmail-question.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-question.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/cfmail-question.png?v=0.9.26" alt="A question card in the middle of a turn: the agent offers several choices and a free-text reply"></picture></a>
 <br><sub>When the agent needs a decision, it asks, and waits for your answer.</sub>
 </td>
 </tr>
 <tr>
 <td valign="top">
-<a href="https://oar.sh/assets/shots/pingcf-thoughts.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/pingcf-thoughts.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/pingcf-thoughts.png?v=0.9.24" alt="An expanded Thoughts block showing the agent's reasoning above its reply"></picture></a>
+<a href="https://oar.sh/assets/shots/pingcf-thoughts.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/pingcf-thoughts.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/pingcf-thoughts.png?v=0.9.26" alt="An expanded Thoughts block showing the agent's reasoning above its reply"></picture></a>
 <br><sub>Read the reasoning, not just the answer.</sub>
 </td>
 </tr>
 <tr>
 <td valign="top">
-<a href="https://oar.sh/assets/shots/pagewright-watching.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/pagewright-watching.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/pagewright-watching.png?v=0.9.24" alt="A session watching two other sessions: a finished reply arrives as a compact relay-event row, the watching 2 sessions list is open with Unsubscribe, and a request to also watch a session on a paired relay waits on an Allow / Deny card"></picture></a>
+<a href="https://oar.sh/assets/shots/pagewright-watching.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/pagewright-watching.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/pagewright-watching.png?v=0.9.26" alt="A session watching two other sessions: a finished reply arrives as a compact relay-event row, the watching 2 sessions list is open with Unsubscribe, and a request to also watch a session on a paired relay waits on an Allow / Deny card"></picture></a>
 <br><sub>Agents that watch agents: woken by events, approving nothing.</sub>
 </td>
 </tr>
@@ -100,7 +100,7 @@ The runtime (the *provider* in the UI) is chosen in **New Chat** and stays fixed
 | ------- | ---------- | ------------- |
 | [**GitHub Copilot**](docs/providers.md#github-copilot--engine-choice) (default) | GitHub's coding agent, with every model your Copilot plan offers | The Copilot CLI, signed in on the relay host |
 | [**OpenAI (BYOK)**](docs/providers.md#openai-byok-and-openai-image) | Copilot's agent against OpenAI or a compatible endpoint | Your own API key |
-| [**OpenAI Image**](docs/providers.md#openai-byok-and-openai-image) | A chat whose replies are images | The same key |
+| [**OpenAI Image**](docs/providers.md#openai-byok-and-openai-image) | A chat whose replies are images, with a daily spending cap; agents can use it too | The same key |
 | [**Claude**](docs/providers.md#claude-agent-sdk) | Claude Code on your machine | A Claude login, made from **⚙️ Settings** |
 | [**Claude API**](docs/providers.md#claude-api) | Claude Code on your machine, billed to a Claude Console API key; a Claude chat can switch between the two | A Claude Console API key |
 | [**Claude Cloud**](docs/providers.md#claude-cloud) (experimental) | Claude Code in a sandbox at Anthropic, on a GitHub repository | A claude.ai account with GitHub connected |
@@ -118,11 +118,11 @@ API keys are entered in **⚙️ Settings** and stay on the relay host. The Clau
 <table>
 <tr>
 <td width="62%" valign="top">
-<a href="https://oar.sh/assets/shots/cfmail-steering.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-steering.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/cfmail-steering.png?v=0.9.24" alt="A running Claude turn with a message steered into it mid-turn, the reply streaming with its tool calls, and the Steer button in the composer"></picture></a>
+<a href="https://oar.sh/assets/shots/cfmail-steering.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/cfmail-steering.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/cfmail-steering.png?v=0.9.26" alt="A running Claude turn with a message steered into it mid-turn, the reply streaming with its tool calls, and the Steer button in the composer"></picture></a>
 <br><sub>Send a message while the agent works, and it steers the running turn.</sub>
 </td>
 <td width="38%" valign="top">
-<a href="https://oar.sh/assets/shots/usage-copilot-compact.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/usage-copilot-compact.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/usage-copilot-compact.png?v=0.9.24" alt="The GitHub Copilot card in Check Usage: premium requests and plan quota with reset dates, and the cost of the last turn"></picture></a>
+<a href="https://oar.sh/assets/shots/usage-copilot-compact.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/usage-copilot-compact.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/usage-copilot-compact.png?v=0.9.26" alt="The GitHub Copilot card in Check Usage: premium requests and plan quota with reset dates, and the cost of the last turn"></picture></a>
 <br><sub>What each plan has left, in one place.</sub>
 </td>
 </tr>
@@ -134,12 +134,13 @@ API keys are entered in **⚙️ Settings** and stay on the relay host. The Clau
 - **Background work**: a task panel lists the commands and sub-agents that outlive a reply, each with its own **Stop** and its full record; a command's console shows its output live, says when it has gone quiet and counts a test run's passes and failures; the panel and its consoles come back after a relay restart, and a queued restart shows what it waits for and asks after a maximum wait ([Background tasks](docs/web-ui.md#background-tasks), [After a relay restart](docs/web-ui.md#after-a-relay-restart)).
 - **Choose how the agent works**: `ask`, `plan`, `agent` or `autopilot`, plus model and reasoning effort, per message, from the composer's **⚙** ([The composer](docs/web-ui.md#the-composer), [Relay modes](docs/web-ui.md#relay-modes)).
 - **Files and git**: browse the workspace, preview files, review git changes in a diff viewer, and publish a dev server as a preview link ([Files, git, and previews](docs/web-ui.md#files-git-and-previews), [preview servers](docs/preview-servers.md)).
-- **Usage and context**: plan limits and credits of every provider, the estimated spend of DeepSeek and Claude API keys, and what fills the model's context window ([Usage and context](docs/web-ui.md#usage-and-context)).
+- **Media**: the pictures and videos of a session, or of every session, in one gallery, with a file viewer that moves through all of them ([Media](docs/web-ui.md#media)).
+- **Usage and context**: plan limits and credits of every provider, the estimated spend of DeepSeek and Claude API keys and of OpenAI images, and what fills the model's context window ([Usage and context](docs/web-ui.md#usage-and-context)).
 - **Your conversations everywhere**: the history lives on the relay, drafts follow you from device to device, and conversations can be searched, archived, grouped and pinned ([The conversation list](docs/web-ui.md#the-conversation-list), [drafts](docs/web-ui.md#drafts-across-devices), [session groups](docs/web-ui.md#session-groups), [pins](docs/web-ui.md#pinned-messages)).
 - **Share** a conversation as a read-only link, hide single messages and the host's paths from it, and see when it was opened ([Sharing](docs/web-ui.md#sharing)).
 - **On your phone**: install OAR as an app and get push notifications; back walks through the conversations you opened, a pinch zooms the whole app, and held sideways it keeps a phone's layout ([Notifications and the app](docs/web-ui.md#notifications-and-the-app), [back and forward](docs/web-ui.md#back-and-forward), [zoom](docs/web-ui.md#zoom), [a phone held sideways](docs/web-ui.md#a-phone-held-sideways)).
 
-The full tour is in [Using the web UI](docs/web-ui.md); every setting and config key is in [Settings and configuration](docs/configuration.md). What changed in each release is in the [changelog](CHANGELOG.md), with [0.9.24](CHANGELOG.md#0924--2026-11-08) as the latest.
+The full tour is in [Using the web UI](docs/web-ui.md); every setting and config key is in [Settings and configuration](docs/configuration.md). What changed in each release is in the [changelog](CHANGELOG.md), with [0.9.26](CHANGELOG.md#0926--2026-11-11) as the latest.
 
 ## Remote access
 
@@ -190,7 +191,7 @@ flowchart LR
 
 You stay in charge: per paired relay you decide whether agents may only read, also prompt, or do everything, and in the `ask` and `plan` modes every write asks you first. A subscription only notifies; it approves nothing.
 
-<p><a href="https://oar.sh/assets/shots/pagewright-cross-relay.png?v=0.9.24"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/pagewright-cross-relay.light.png?v=0.9.24"><img src="https://oar.sh/assets/shots/pagewright-cross-relay.png?v=0.9.24" alt="A session watching two sessions on a paired relay: a reply arrives as a relay-event row, and after a short outage the events the other relay kept arrive as one replayed row" width="720"></picture></a></p>
+<p><a href="https://oar.sh/assets/shots/pagewright-cross-relay.png?v=0.9.26"><picture><source media="(prefers-color-scheme: light)" srcset="https://oar.sh/assets/shots/pagewright-cross-relay.light.png?v=0.9.26"><img src="https://oar.sh/assets/shots/pagewright-cross-relay.png?v=0.9.26" alt="A session watching two sessions on a paired relay: a reply arrives as a relay-event row, and after a short outage the events the other relay kept arrive as one replayed row" width="720"></picture></a></p>
 
 How pairing, permissions and limits work is in [Agents working together](docs/agents-together.md).
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oar-sh/oar",
-  "version": "0.9.24",
+  "version": "0.9.26",
   "type": "module",
   "description": "OAR — Open Agent Relay: drive your local coding agents from any browser",
   "license": "MIT",
@@ -25,6 +25,7 @@
     "server/",
     "!server/**/*.test.mjs",
     "!server/**/*-test-harness.mjs",
+    "!server/**/*-test-fixtures.mjs",
     "!server/**/fixtures/",
     "!server/README.md",
     "!server/data/",
@@ -36,6 +37,7 @@
     "shared/",
     "!shared/**/*.test.mjs",
     "!shared/**/*-test-harness.mjs",
+    "!shared/**/*-test-fixtures.mjs",
     "!shared/**/fixtures/",
     ".github/extensions/web-relay/",
     "!.github/extensions/web-relay/**/*.test.mjs",
@@ -72,7 +74,9 @@
     "@modelcontextprotocol/sdk": "1.32.2",
     "@playwright/test": "^1.60.0",
     "jsdom": "^30.0.1",
+    "katex": "0.19.1",
     "koffi": "3.3.4",
+    "marked": "15.0.12",
     "zod": "4.6.7"
   },
   "overrides": {

package/server/claude-worker/claude-remote-relay-tool.mjs

@@ -15,6 +15,7 @@ import {
   REMOTE_RELAY_THINKING_DISPLAYS,
 } from '../../shared/remote-relay-contract.mjs';
 import { RELAY_ASK_URGENCIES } from '../../shared/relay-asks.mjs';
+import { OPENAI_IMAGE_BACKGROUNDS, OPENAI_IMAGE_OUTPUT_FORMATS } from '../../shared/openai-image-options.mjs';
 import {
   REMOTE_RELAY_CALL_FAILED,
   REMOTE_RELAY_TOOL_DESCRIPTION,
@@ -80,6 +81,20 @@ export const REMOTE_RELAY_TOOL_ZOD_SHAPE = {
   thinking_enabled: z.boolean().optional().describe(FIELDS.thinking_enabled.description),
   planOnly: z.boolean().optional().describe(FIELDS.planOnly.description),
   thinking_display: choice('thinking_display', REMOTE_RELAY_THINKING_DISPLAYS),
+  // A list, or one path as text, which the contract takes as well.
+  images: z.union([z.array(z.string()), z.string()]).optional().describe(FIELDS.images.description),
+  // Loose like quiet_hours: the contract's validator checks the values per
+  // model family and names an unknown key; the two fixed lists are enums as
+  // in the JSON schema.
+  image: z.looseObject({
+    size: z.string().optional().describe(FIELDS.image.properties.size.description),
+    quality: z.string().optional().describe(FIELDS.image.properties.quality.description),
+    count: z.number().optional().describe(FIELDS.image.properties.count.description),
+    output_format: z.enum([...OPENAI_IMAGE_OUTPUT_FORMATS]).optional().describe(FIELDS.image.properties.output_format.description),
+    background: z.enum([...OPENAI_IMAGE_BACKGROUNDS]).optional().describe(FIELDS.image.properties.background.description),
+    // Free text: the target relay knows its presets and resolves the name.
+    preset: z.string().optional().describe(FIELDS.image.properties.preset.description),
+  }).optional().describe(FIELDS.image.description),
   question_id: text('question_id'),
   // ask: the contract's validator refuses a question or detail that is too
   // long, or more than eight options, with a message the model can act on.

package/server/claude-worker/claude-sdk-adapter.mjs

@@ -3,6 +3,7 @@ import { loadAgentSdk } from '../agent-sdk-loader.mjs';
 import { formatClaudeBackgroundCommandLimitLabel } from '../../shared/claude-background-command-limit.mjs';
 import { parseThinkingDisplay } from '../../shared/claude-thinking.mjs';
 import { PERSON_ONLY_QUESTION_SOURCE } from '../../shared/relay-asks.mjs';
+import { MATH_INSTRUCTION_TEXT } from '../../shared/math-instructions.mjs';
 import { MEDIA_EMBED_INSTRUCTION_TEXT } from '../../shared/media-embed-instructions.mjs';
 import { REMOTE_RELAY_TOOL_CALL_TIMEOUT_MS } from '../../shared/remote-relay-tool-core.mjs';
 import { createPreviewToolDefinition } from './claude-preview-tool.mjs';
@@ -205,10 +206,12 @@ export function longCommandInstructionText(backgroundCommandLimitMs = null) {
 export function systemPromptForRelayMode(relayMode, { backgroundCommandLimitMs = null } = {}) {
   const mode = String(relayMode || 'agent').trim().toLowerCase();
   // The media-embed capability rides every session — unlike the mode nudges it
-  // is renderer knowledge the model cannot discover any other way. The long
-  // command note rides every session too: the limit is the relay's setting.
+  // is renderer knowledge the model cannot discover any other way; so is the
+  // notation the web app renders formulas from. The long command note rides
+  // every session too: the limit is the relay's setting.
   const append = [
     MEDIA_EMBED_INSTRUCTION_TEXT,
+    MATH_INSTRUCTION_TEXT,
     longCommandInstructionText(backgroundCommandLimitMs),
     MODE_SYSTEM_PROMPT_APPEND[mode],
   ]
@@ -792,8 +795,9 @@ const EXIT_PLAN_DENY_MESSAGE = 'Plan received — it is shown to the user in the
   + 'End the turn now with a short closing message and do not start implementing; '
   + 'the user\'s decision will arrive as a new message.';
 // When the handler could not surface a board (no plan text reached it), the
-// model must restate the plan as its final message: plan-shaped final text
-// triggers the runner's fallback board, so the user still gets the review UI.
+// model must restate the plan as its final message: after a hand-over without
+// a card, the turn's reply is posted as the plan (publishCompletedTurn), so
+// the user still gets the review UI.
 const EXIT_PLAN_DENY_NO_BOARD_MESSAGE = 'Plan received, but it could not be shown for review. '
   + 'Restate the complete plan as your final message, then end the turn without implementing; '
   + 'the user\'s decision will arrive as a new message.';

package/server/claude-worker/claude-session-process.mjs

@@ -876,6 +876,8 @@ export function createClaudeSessionRunner({
         modelUsage: null,
       },
       planBoardPosted: false,
+      // The CLI handed a plan over (ExitPlanMode), whether or not a card came of it.
+      planBoardRequested: false,
       // The CLI process the turn ran in, as its usage report names it: taken
       // from the process here and again when the context is made active, and
       // read from the context at finalize — a late continuation flush
@@ -1569,6 +1571,7 @@ export function createClaudeSessionRunner({
         state: ctx.state,
         responseModel,
         planBoardPosted: ctx.planBoardPosted,
+        planBoardRequested: ctx.planBoardRequested,
         consumedSteerIds,
       });
     }
@@ -3789,6 +3792,7 @@ export function createClaudeSessionRunner({
       onExitPlanMode: async (input) => {
         const ctx = processRef.activeCtx || processRef.pendingDelivered[0]?.ctx || null;
         if (!ctx?.message?.id) return false;
+        ctx.planBoardRequested = true;
         const posted = await publisher.publishPlanBoard(ctx.message, input);
         if (posted) ctx.planBoardPosted = true;
         return posted;

package/server/claude-worker/claude-turn-publisher.mjs

@@ -6,7 +6,6 @@ import {
   classifyClaudeResultFailure,
   claudeFailureGuidance,
 } from './claude-turn-failure.mjs';
-import { countPlanLikeLines } from '../../shared/plan-lines.mjs';
 import { PLAN_BOARD_ACTIONS } from '../../shared/plan-board-actions.mjs';
 import { EMPTY_TURN_COMPLETION_NOTE } from '../../shared/empty-turn-completion.mjs';
 import { buildSteerSettleFailure, steerSettleFailureText } from '../../shared/steer-settle-failure.mjs';
@@ -34,9 +33,10 @@ export function isStaleAttemptError(error) {
 export { buildSteerSettleFailure, steerSettleFailureText };
 export const STEER_SETTLE_FAILED_TEXT = steerSettleFailureText('folded');
 
-export function buildClaudePlanReadyBoardPayload({ message, planText = '' } = {}) {
+export function buildClaudePlanReadyBoardPayload({ message, planText = '', planFilePath = '' } = {}) {
   const summary = String(planText || '').trim();
   if (!summary) return null;
+  const filePath = String(planFilePath || '').trim();
   return {
     queueId: message?.id,
     messageId: message?.id,
@@ -52,6 +52,8 @@ export function buildClaudePlanReadyBoardPayload({ message, planText = '' } = {}
       queueMessageId: message?.id || null,
       conversationId: message?.conversationId || null,
       relayMode: message?.relayMode || 'agent',
+      // Named on the card when the relay has to cut a very long plan.
+      ...(filePath ? { planFilePath: filePath } : {}),
     },
   };
 }
@@ -797,13 +799,13 @@ export function createClaudeTurnPublisher({
    * Post the plan-ready review board for an ExitPlanMode interception. The CLI
    * sends both `plan` and `planFilePath`; the inline plan is authoritative, the
    * plan file covers CLI versions that stop inlining it. Returns false when no
-   * plan text could be recovered so the caller's plan-shaped-final-text
-   * fallback still gets a chance to post the board.
+   * plan text could be recovered, so the turn's reply can stand in for it
+   * (see publishCompletedTurn).
    */
   async function publishPlanBoard(message, input) {
     let planText = String(input?.plan || input?.summary || '').trim();
+    const planFilePath = String(input?.planFilePath || '').trim();
     if (!planText) {
-      const planFilePath = String(input?.planFilePath || '').trim();
       if (planFilePath) {
         try {
           planText = String(fs.readFileSync(planFilePath, 'utf8') || '').trim();
@@ -812,7 +814,7 @@ export function createClaudeTurnPublisher({
         }
       }
     }
-    const boardPayload = buildClaudePlanReadyBoardPayload({ message, planText });
+    const boardPayload = buildClaudePlanReadyBoardPayload({ message, planText, planFilePath });
     if (!boardPayload) return false;
     await api('POST', '/api/relay-board', boardPayload).catch((error) => {
       dbg('plan board publish failed', error?.message || String(error));
@@ -822,18 +824,30 @@ export function createClaudeTurnPublisher({
 
   /**
    * Publish a completed (non-error, non-aborted) turn: final stream snapshot,
-   * plan-mode fallback board, response row. Mirrors the pre-session-process
-   * turn runner's tail exactly.
+   * plan-mode fallback board, response row.
+   *
+   * A plan card comes only from a plan hand-over (ExitPlanMode). The fallback
+   * covers one case: the CLI handed a plan over but its text could not be
+   * read, so the reply is the plan. A reply is never taken for a plan because
+   * of its list lines: in a session that stays in plan mode, every status
+   * reply with two bullets got a card.
    */
-  async function publishCompletedTurn({ message, state, responseModel, planBoardPosted, consumedSteerIds = null }) {
+  async function publishCompletedTurn({
+    message,
+    state,
+    responseModel,
+    planBoardPosted,
+    planBoardRequested = false,
+    consumedSteerIds = null,
+  }) {
     const result = state.result;
     const finalText = String(
       state.resultTexts.join('\n\n') || result?.text || state.lastStreamedText || '',
     ).trim();
     if (
-      !planBoardPosted
+      planBoardRequested
+      && !planBoardPosted
       && String(message.relayMode || '').trim().toLowerCase() === 'plan'
-      && countPlanLikeLines(finalText) >= 2
     ) {
       const boardPayload = buildClaudePlanReadyBoardPayload({ message, planText: finalText });
       if (boardPayload) {