​
Overview

Allternit integrates anydoc (MIT, @firecrawl/anydoc pinned at 0.1.6) — Firecrawl’s Rust document-to-Markdown converter with prebuilt Node bindings and zero external converter dependencies. It powers the “Open as Markdown” capability across the platform: any supported document becomes clean, LLM-ready GFM Markdown. The converter lives in the office engine (services/office-engine) as POST /markdown, proxied by the gateway at /api/office/markdown and /api/v1/office/engines/markdown. The original file is never modified; conversion is read-only.

​
Supported formats

Known limits (accepted by design): no OCR (scanned PDFs are rejected as unsupported), no HTML or image input, encrypted files are rejected, and decompressed input is capped at 512 MiB.

​
API

Send the raw document bytes as the body with an x-office-filename header — the extension drives format detection (with content-based detection as fallback).
Success returns an artifact-shaped JSON document with the converted markdown, detected format, and a title (first heading when present). Errors are mapped to stable codes:

​
URL → Markdown

anydoc itself is documents-only, so the office engine also exposes POST /markdown-url for web pages. It fetches the page server-side with SSRF guards (http/https only; private, loopback, and link-local hosts blocked; ≤5 redirects; 15s timeout; ~10 MiB body cap), extracts the main content with @mozilla/readability, and converts it to GFM with turndown. A URL that serves a document content-type (PDF, OOXML, …) is passed through the anydoc byte path — one endpoint handles anything.
The response matches the /markdown shape with format: 'html' (or the detected document format), sourceUrl (after redirects), title, and markdown. Error mapping: 400 for invalid/blocked URLs (invalidUrl / blockedUrl), 415 for unsupported content-types, 422 for fetch/parse failures. The Markdown Preview view exposes this as the Open URL as Markdown action on the office launcher and as a URL input in the preview’s empty state.

​
Per-surface behavior

  • Web platform — formats with no native editor (legacy Office, ODF, .rtf, .epub, .csv) open in the Markdown Preview view from the office launcher. Editor-owned formats (.docx .xlsx .pptx .pdf) still open in their editors. The preview offers Copy, Download .md, and Save as artifact (the markdown persists as a markdown-preview/markdown section, so agents and mobile clients see it through the existing artifact channel).
  • Desktop — file-association opens of anydoc-only formats land in a Markdown Preview window; in the shell they open as a preview view.
  • Office add-in — the task pane’s connected-document panel has a View as Markdown action: the current document’s bytes (Document.getFileAsync) are converted via the v1 endpoint and shown in a read-only panel with an Open in platform link.
  • Mobile (iOS) — no dedicated UI in v1: converted documents saved as artifacts already render in OfficeDocumentView via plaintext sections.

​
Source

  • Engine endpoint: services/office-engine/src/markdown.ts (+ POST /markdown in src/index.ts)
  • URL → Markdown: services/office-engine/src/markdown-url.ts (+ POST /markdown-url in src/index.ts)
  • Gateway proxy: cmd/allternit-api/src/office_engine_routes.rs
  • Web view: surfaces/ai.allternit.com/src/views/office/MarkdownPreviewView.tsx
  • Desktop routing: surfaces/allternit-desktop/src/main/office-programs.ts
  • Add-in plumbing: surfaces/allternit-extensions/allternit-office-addin/src/lib/markdown-conversion.ts