@dragon708/docmind-browser 1.7.0 → 1.8.0

package/dist/index.d.ts

@@ -44,7 +44,10 @@ type BrowserExtractStructuredDataOptions = BrowserAnalyzeOptions & {
     readonly normalize?: NormalizeStructuredOptions;
 };
 /**
- * {@link extractMarkdown}: structured options plus `markdown` for `renderMarkdown` (`@dragon708/docmind-markdown`).
+ * {@link extractMarkdown}: structured options plus `markdown` (passed through to `extractMarkdown` in
+ * `@dragon708/docmind-markdown`, including structured-serializer knobs). Binary PDF/DOCX converters inside that
+ * package are not used for PDF in-browser and the DOCX bytes→Turndown path is Node-only; the browser still
+ * gets correct DOCX Markdown via structured fallback from {@link extractStructuredData}.
  */
 interface BrowserExtractMarkdownOptions extends BrowserExtractStructuredDataOptions {
     readonly markdown?: RenderMarkdownOptions;
@@ -120,24 +123,33 @@ declare function runOcr(input: BrowserAnalyzeInput, options?: BrowserAnalyzeOpti
 declare function extractStructuredData(input: BrowserAnalyzeInput, options?: BrowserExtractStructuredDataOptions): Promise<StructuredDocumentResult>;
 
 /**
- * {@link extractStructuredData} → `renderMarkdown` (browser-safe; no Node-only APIs).
- * PDF: empty structured stub → usually empty output; see {@link getCapabilities} capability `markdown`.
+ * {@link extractStructuredData} for a full structured envelope, then `extractMarkdown` from
+ * `@dragon708/docmind-markdown` on `{ data, filename?, mimeType? }` with that result as `structuredFallback`.
  *
- * @param input - `File` / `Blob` / bytes as accepted by the browser facade.
- * @param options - `markdown` render options plus structured/`ocr`/`docx` routing (same as {@link extractStructuredData}).
+ * - **PDF:** the markdown package does not load `@opendataloader/pdf` here; output comes from the structured
+ *   fallback (empty in-browser stub — see {@link getCapabilities}).
+ * - **DOCX:** the package’s direct bytes → Mammoth → Turndown path is **Node-only**; in-browser, Markdown is
+ *   produced via `convertStructuredToMarkdown` on the structured envelope (still Mammoth/OOXML-backed via
+ *   `@dragon708/docmind-docx`), with an explanatory warning from the markdown package.
+ * - **Text / image:** unidentified or non-binary bytes use the same structured serializer.
+ *
+ * @param options - `markdown` options plus the same routing as {@link extractStructuredData} (`ocr`, `docx`, `normalize`).
  */
 declare function extractMarkdown(input: BrowserAnalyzeInput, options?: BrowserExtractMarkdownOptions): Promise<string>;
 /**
- * Same as {@link extractMarkdown} but `renderLlmText` (compact plain text for LLM prompts).
+ * {@link extractStructuredData} then `renderLlmText` (`@dragon708/docmind-markdown`). For a structured value you already have, that package's `extractLlmContent` matches `renderLlmText` (no file I/O).
  */
 declare function extractLlmContent(input: BrowserAnalyzeInput, options?: BrowserExtractLlmContentOptions): Promise<string>;
 /**
- * Structured extract → `renderMarkdownSections` for sectioned Markdown in the browser.
+ * Structured extract → `renderMarkdownSections` (`splitStructuredIntoChunks` with Markdown; same as
+ * `extractStructuredChunks` alias in `@dragon708/docmind-markdown`).
  */
 declare function extractStructuredChunks(input: BrowserAnalyzeInput, options?: BrowserExtractStructuredChunksOptions): Promise<readonly MarkdownSection[]>;
 
 /** High-level features the user can ask DocMind for (per input kind and runtime). */
-type PublicCapabilityId = "text" | "metadata" | "html" | "ocr" | "pages" | "structured-output" | "markdown" | "llm-text" | "structured-chunks" | "image-normalization" | "gif-first-frame" | "bmp" | "heic" | "tiff";
+type PublicCapabilityId = "text" | "metadata" | "html" | "ocr" | "pages" | "structured-output"
+/** Browser: {@link extractMarkdown} via `@dragon708/docmind-markdown` `extractMarkdown` + structured fallback (PDF empty; DOCX structured path when binary converter is Node-only). */
+ | "markdown" | "llm-text" | "structured-chunks" | "image-normalization" | "gif-first-frame" | "bmp" | "heic" | "tiff";
 declare function docxIncludeRequested(flags?: AnalyzeDocxIncludeFlags): boolean;
 /** DOCX `word/media` en runtime browser (JSZip; sin pipeline Node). */
 interface DocxEmbeddedImageCapabilities {
@@ -218,7 +230,7 @@ type BrowserExplainAnalysisPlanOptions = Omit<ExplainAnalysisPlanOptions, "inten
 
 /**
  * Epic 1 — **Capabilities:** detects kind from the same hints as `analyzeFile`, then lists which of
- * `text` | `metadata` | `html` | `ocr` | `pages` | `structured-output` | `markdown` | `llm-text` | `structured-chunks`
+ * `text` | `metadata` | `html` | `ocr` | `pages` | `structured-output` | `markdown` (package `extractMarkdown` + structured fallback; PDF empty here) | `llm-text` | `structured-chunks` (split + Markdown sections)
  * and image-specific ids (`image-normalization`, `bmp`, `gif-first-frame`, `heic`, `tiff`) apply in the browser (PDF always unsupported for meaningful content).
  * No Mammoth/Tesseract/PDF parsing. For DOCX, {@link GetCapabilitiesReport.docxStructure} / `docxEmbeddedImages` describe v2 opt-in features.
  */

package/dist/index.js

@@ -4,7 +4,7 @@ import { extractStructuredDataFromDocx, analyzeDocx } from '@dragon708/docmind-d
 export { extractStructuredDataFromDocx } from '@dragon708/docmind-docx';
 import { extractStructuredDataFromImage, preprocessHasEffect, resolveImageFormat, normalizeImageForOcr, ocrTiff, ocrImageDetailed } from '@dragon708/docmind-ocr';
 export { extractStructuredDataFromImage } from '@dragon708/docmind-ocr';
-import { renderMarkdown, renderLlmText, renderMarkdownSections } from '@dragon708/docmind-markdown';
+import { extractMarkdown as extractMarkdown$1, renderLlmText, renderMarkdownSections } from '@dragon708/docmind-markdown';
 
 // src/analyzeFile.ts
 function assertBrowserInput(input) {
@@ -520,11 +520,26 @@ async function runOcr(input, options) {
       return notImplementedResult(kind, "none", [UNKNOWN_FORMAT_WARNING]);
   }
 }
+function browserFileHints(input) {
+  if (input instanceof File) {
+    return {
+      filename: input.name,
+      mimeType: input.type ? input.type : void 0
+    };
+  }
+  return {};
+}
 async function extractMarkdown(input, options) {
   throwIfAborted(options?.signal);
   const { markdown: markdownOpts, ...structuredOpts } = options ?? {};
   const structured = await extractStructuredData(input, structuredOpts);
-  return renderMarkdown(structured, markdownOpts);
+  const data = await toUint8Array(input);
+  const hints = browserFileHints(input);
+  const r = await extractMarkdown$1(
+    { data, filename: hints.filename, mimeType: hints.mimeType },
+    { ...markdownOpts ?? {}, structuredFallback: structured }
+  );
+  return r.markdown;
 }
 async function extractLlmContent(input, options) {
   throwIfAborted(options?.signal);
@@ -576,7 +591,7 @@ var TEXT_META_NOTE = "Plain text has no structured document metadata; extractMet
 var OCR_OFF_NOTE = 'Image OCR may be skipped when `ocr.mode` is "off" in analyze options.';
 var STRUCTURED_OCR_OFF = 'Structured image output uses OCR; when `ocr.mode` is "off", `extractStructuredData` returns an empty envelope with a warning.';
 var UNKNOWN_KIND = "Could not determine file kind from name, MIME, or bytes; all features are reported as unsupported until the kind is known.";
-var MARKDOWN_PDF_BROWSER = "PDF: no parser in-browser \u2014 extractMarkdown / extractLlmContent / extractStructuredChunks only render an empty structured stub; use @dragon708/docmind-node for real PDF \u2192 Markdown / LLM text / chunks.";
+var MARKDOWN_PDF_BROWSER = "PDF: no parser in-browser \u2014 `@opendataloader/pdf` is not loaded here. extractMarkdown still calls `extractMarkdown` in `@dragon708/docmind-markdown`, which falls back to the empty structured stub (same empty Markdown). extractLlmContent / extractStructuredChunks use the structured envelope only. Use @dragon708/docmind-node for real PDF \u2192 Markdown / LLM text / chunks.";
 var MARKDOWN_IMAGE_OCR_OFF = 'Same as structured-output: when ocr.mode is "off", structured (and thus Markdown/LLM/chunk exports) are empty aside from warnings.';
 function slot(id, supported, warnings) {
   return warnings?.length ? { id, supported, warnings } : { id, supported };
@@ -619,13 +634,13 @@ function buildBrowserCapabilityReport(kind) {
           "`extractStructuredData` uses `@dragon708/docmind-docx` (Mammoth + OOXML) and returns `StructuredDocumentResult`; optional `options.docx` slices are forwarded."
         ]),
         slot("markdown", true, [
-          "extractMarkdown: structured DOCX \u2192 Markdown via `@dragon708/docmind-markdown` (browser-safe, no Node APIs)."
+          "extractMarkdown: `@dragon708/docmind-markdown` `extractMarkdown` on bytes + structured fallback. The package\u2019s DOCX-bytes Mammoth\u2192Turndown path is Node-only; in-browser, Markdown is produced from `extractStructuredData` (Mammoth/OOXML in `@dragon708/docmind-docx`) via structured serialization, with a clear package warning that the binary shortcut is skipped."
         ]),
         slot("llm-text", true, [
-          "extractLlmContent: structured \u2192 compact plain text for prompts (same pipeline as Markdown export)."
+          "extractLlmContent: structured envelope \u2192 `renderLlmText` (LLM-ready plain text; no binary PDF/DOCX Markdown routes)."
         ]),
         slot("structured-chunks", true, [
-          "extractStructuredChunks: structured \u2192 Markdown sections (`splitStructuredIntoChunks` in markdown package)."
+          "extractStructuredChunks: structured \u2192 `renderMarkdownSections` / `splitStructuredIntoChunks` (heading-aware chunking + optional parallel `text`)."
         ])
       ];
       break;
@@ -664,7 +679,7 @@ function buildBrowserCapabilityReport(kind) {
           "HEIC/HEIF and TIFF limitations match `getCapabilities` (`heic`, `tiff`) and OCR warnings."
         ]),
         slot("markdown", true, [
-          "extractMarkdown: OCR structured layout \u2192 Markdown when OCR runs; HEIC unsupported; TIFF best-effort.",
+          "extractMarkdown: same bytes + structured fallback through package `extractMarkdown` when applicable; OCR structured layout \u2192 Markdown when OCR runs. HEIC unsupported; TIFF best-effort.",
           MARKDOWN_IMAGE_OCR_OFF
         ]),
         slot("llm-text", true, [
@@ -688,11 +703,13 @@ function buildBrowserCapabilityReport(kind) {
           "`extractStructuredData` decodes UTF-8 (via `analyzeText`) and normalizes to `StructuredDocumentResult` (paragraph block rollup)."
         ]),
         slot("markdown", true, [
-          "extractMarkdown: UTF-8 structured rollup \u2192 Markdown (`@dragon708/docmind-markdown`)."
+          "extractMarkdown: bytes + structured fallback through `@dragon708/docmind-markdown` `extractMarkdown` (typically structured serializer for UTF-8 text)."
+        ]),
+        slot("llm-text", true, [
+          "extractLlmContent: UTF-8 structured rollup \u2192 `renderLlmText` in `@dragon708/docmind-markdown`."
         ]),
-        slot("llm-text", true, ["extractLlmContent: UTF-8 structured \u2192 LLM-oriented plain text."]),
         slot("structured-chunks", true, [
-          "extractStructuredChunks: typically a single Markdown section when only paragraph rollup exists."
+          "extractStructuredChunks: typically one Markdown section when only paragraph rollup exists."
         ])
       ];
       break;
@@ -897,10 +914,22 @@ function buildBrowserExplainReport(kind, intent, ocrMode, plan, docxInclude, ocr
     case "extractLlmContent":
     case "extractStructuredChunks":
       if (kind === "docx") {
-        nativeExtraction = {
-          willAttempt: true,
-          description: intent === "extractStructuredData" ? "`extractStructuredDataFromDocx`: Mammoth plus required OOXML extractors (structure, headings, tables, blocks, pagesApprox, embeddedImages unless disabled), then `normalizeToStructuredResult`. Optional `options.docx` is forwarded." : `${String(intent)}: same structured DOCX pipeline as extractStructuredData, then \`@dragon708/docmind-markdown\` (browser-safe).`
-        };
+        if (intent === "extractStructuredData") {
+          nativeExtraction = {
+            willAttempt: true,
+            description: "`extractStructuredDataFromDocx`: Mammoth plus required OOXML extractors (structure, headings, tables, blocks, pagesApprox, embeddedImages unless disabled), then `normalizeToStructuredResult`. Optional `options.docx` is forwarded."
+          };
+        } else if (intent === "extractMarkdown") {
+          nativeExtraction = {
+            willAttempt: true,
+            description: "extractMarkdown: `extractStructuredData` (Mammoth/OOXML) for a full structured envelope, then `extractMarkdown` in `@dragon708/docmind-markdown`. The package\u2019s DOCX-bytes Mammoth\u2192Turndown shortcut is Node-only; in-browser Markdown uses structured serialization on that envelope (with a package warning)."
+          };
+        } else {
+          nativeExtraction = {
+            willAttempt: true,
+            description: `${String(intent)}: same structured DOCX pipeline as extractStructuredData, then \`@dragon708/docmind-markdown\` (\`renderLlmText\` or \`renderMarkdownSections\`).`
+          };
+        }
         ocr = { mayUse: false, description: "DOCX structured path does not use OCR." };
         limitations = lim(DOCX_ZIP_NOTE_BROWSER);
       } else if (kind === "image") {
@@ -923,17 +952,21 @@ function buildBrowserExplainReport(kind, intent, ocrMode, plan, docxInclude, ocr
       if (intent === "extractMarkdown") {
         limitations = [
           ...limitations,
-          ...lim("Output: Markdown string via renderMarkdown.")
+          ...lim(
+            "Output: Markdown string from `extractMarkdown` in `@dragon708/docmind-markdown`. PDF in-browser: empty (no `@opendataloader/pdf`). DOCX: structured Markdown path when the binary converter is Node-only."
+          )
         ];
       } else if (intent === "extractLlmContent") {
         limitations = [
           ...limitations,
-          ...lim("Output: compact plain text via renderLlmText.")
+          ...lim("Output: compact plain text via `renderLlmText` (structured input only in this runtime).")
         ];
       } else if (intent === "extractStructuredChunks") {
         limitations = [
           ...limitations,
-          ...lim("Output: MarkdownSection[] via renderMarkdownSections.")
+          ...lim(
+            "Output: MarkdownSection[] via `renderMarkdownSections` (`splitStructuredIntoChunks` / `extractStructuredChunks` alias)."
+          )
         ];
       }
       break;
@@ -1088,7 +1121,21 @@ function planForIntent(intentOpt, kind, ocrMode, docxInclude, ocr, analyzeFileOu
         };
     }
   }
-  if (intent === "extractMarkdown" || intent === "extractLlmContent" || intent === "extractStructuredChunks") {
+  if (intent === "extractMarkdown") {
+    const sub = planForIntent(
+      "extractStructuredData",
+      kind,
+      ocrMode,
+      docxInclude,
+      ocr,
+      analyzeFileOutput
+    );
+    return {
+      intent,
+      steps: [...sub.steps ?? [], { id: "markdown_hybrid_package", status: "planned" }]
+    };
+  }
+  if (intent === "extractLlmContent" || intent === "extractStructuredChunks") {
     const sub = planForIntent(
       "extractStructuredData",
       kind,
@@ -1101,7 +1148,10 @@ function planForIntent(intentOpt, kind, ocrMode, docxInclude, ocr, analyzeFileOu
       intent,
       steps: [
         ...sub.steps ?? [],
-        { id: "docmind_markdown_render", status: "planned" }
+        {
+          id: intent === "extractLlmContent" ? "docmind_markdown_llm_text" : "docmind_markdown_chunk_sections",
+          status: "planned"
+        }
       ]
     };
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dragon708/docmind-browser",
-  "version": "1.7.0",
+  "version": "1.8.0",
   "description": "Official DocMind browser facade: analyzeFile and intent APIs (DOCX, image OCR, text). PDF and fs paths use @dragon708/docmind-node.",
   "type": "module",
   "sideEffects": false,
@@ -34,7 +34,7 @@
   "license": "MIT",
   "dependencies": {
     "@dragon708/docmind-docx": "^1.8.0",
-    "@dragon708/docmind-markdown": "^1.0.0",
+    "@dragon708/docmind-markdown": "^1.1.0",
     "@dragon708/docmind-ocr": "^1.1.4",
     "@dragon708/docmind-shared": "^1.2.0"
   },