AnyDoc
Convert 14 document formats to clean Markdown in milliseconds
About AnyDoc
AnyDoc converts documents into clean GitHub-Flavored Markdown. Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF all go in, and consistent Markdown comes out, which is the format models actually read well. It is written in Rust by the Firecrawl team and released under MIT.
Two design decisions make it worth knowing about. First, there is no machine learning in it at all: it is a parser, so median conversion is under five milliseconds and the output is deterministic rather than a different interpretation each run. That is a sharp contrast with document handling that routes everything through a vision model, which is slower, costs money per page and produces something slightly different every time. Second, every format parses into the same document model and renders through the same serializer, so headings, nested lists, merged table cells and footnotes come out identically whether the source was a .doc from 2003 or yesterday's .pptx. The project claims that of seven converters benchmarked across 100 documents it was the only one to handle all fourteen formats, which is their own benchmark and should be read as such. It also ships as an agent skill, so a coding agent can read any document it encounters. PDF support covers text-based PDFs through a companion tool, which means scanned images are still not solved here.
You install it wherever you work: a Rust crate, an npm package, a Python package, or a one-line CLI invocation with npx. Point it at a document and it returns Markdown. Because it compiles to WebAssembly it also runs entirely in a browser, which is how the demo on the project page works and means files never leave the machine, an unusually strong privacy property for document processing. Installing it as an agent skill teaches an agent to reach for the CLI when it meets a document it cannot otherwise read, and that works across Claude Code, Codex, Cursor and OpenCode.
- •Fourteen Formats, One Output - Office, OpenDocument, RTF, EPUB, CSV and PDF all render through the same serializer
- •No Model, No Service - Pure Rust parsing, median under five milliseconds, deterministic output every run
- •Runs Locally or In-Browser - WebAssembly build means documents never have to be uploaded anywhere
- •Bindings Everywhere - crates.io, npm, PyPI, CLI and WASM
- •Ships as an Agent Skill - One command teaches a coding agent to read documents it otherwise cannot
- •MIT Licensed - Usable inside commercial products without conditions
Anyone building a RAG pipeline, a document agent or an ingestion step who is currently paying a vision model to read a spreadsheet. The speed and determinism matter most at volume, where per-page model costs and inconsistent output both compound. It is also a straightforward upgrade for coding agents that regularly meet documents they cannot open. If your documents are scanned images or handwriting, this is the wrong tool and you still need OCR.
Pricing
No pricing page found
Free and MIT licensed. Available on crates.io, npm and PyPI, as a CLI, as WebAssembly, and as an agent skill.
Checked 2026-09-18
