pypi / firecrawl-anydoc
firecrawl-anydoc
What changed in firecrawl-anydoc 0.2, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-27.
6 breaking7 new APIs and options1 behavior and defaults1 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli initfirecrawl-anydoc 0.215 changes
Breaking changes 6
The CLI reports which pages need OCR and exits with code 3.
Rust conversion now reports scanned or image-only pages with their page numbers instead of silently dropping them or failing as unsupported.
Checkboxes now use one Rust node, Inline::Checkbox(bool), with corresponding checkbox kind and checked field in wasm and Python documents.
ListItem.checked is gone; a task list item is now identified by content that starts with a checkbox.
The Markdown renderer now owns the [x] / [ ] checkbox token for both table cells and lists.
Ordinary-text dollar signs are escaped only when they could open a math span, so prices such as $20.00 are no longer escaped.
New APIs and options 7
The CLI can opt in to hosted OCR for PDFs that fail with needsOcr; only those PDFs are sent to Firecrawl Parse.
The Node package entry point is now anydoc.js, a wrapper over the generated binding; everything index.js exported remains exported.
Setting this environment variable raises hosted OCR limits.
Hosted OCR failures are returned as hosted or HostedError.
Python conversion errors expose scanned page numbers and the total page count.
Python can opt in to hosted OCR for PDFs that fail with needsOcr; only those PDFs are sent to Firecrawl Parse.
Node can opt in to hosted OCR for PDFs that fail with needsOcr; only those PDFs are sent to Firecrawl Parse.
Behavior and default changes 1
Form control checkboxes in spreadsheet formats now appear in their cell as [x] or [ ], followed by a caption when present; hidden controls remain omitted.
Fixes and other changes 1
Node and wasm errors expose scanned page numbers and the total page count.