anydoc turns Word, PowerPoint, Excel and PDF files into Markdown in Rust: Reveneau

Why it mattersDocument parsing is the dull but necessary part of every retrieval and agent pipeline, and most teams solve it twice, with two libraries that handle tables differently.

Firecrawl published anydoc, a Rust library that converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF files into GitHub-flavoured Markdown. The repository was created on 3 August 2026. Measured on 31 August it had 19,631 stars and 1,171 forks, under an MIT licence.

What you get

There are four ways to use it. A Rust crate, a Node package, a Python package, and a WebAssembly build that runs in a browser. The demo page runs the WebAssembly version, so files converted there stay on the machine.

The command line covers the common cases:

npx @firecrawl/anydoc report.docx               # Markdown to stdout
npx @firecrawl/anydoc slides.pptx -o slides.md  # or to a file
npx @firecrawl/anydoc - --format csv < data.csv # read stdin

There is also an Agent Skill in the repository, so a coding agent can convert a document it finds without being told how.

The project says it produces one consistent output no matter which format goes in, and that it was built to turn office documents into Markdown a model can read. Scanned pages need optical character recognition, which anydoc does not do on its own; the repository points those at Firecrawl's hosted service.

Why this keeps being a problem

Every team that builds retrieval over internal documents meets the same problem. The documents are .docx and .pptx and .xlsx, the model wants text, and the libraries in between each handle tables in their own way.

The usual outcome is two or three parsers, one per format, each producing different Markdown, and a set of later processing rules that were written against whichever one was tried first. Then someone uploads a spreadsheet and the whole thing produces a large block of pipe characters.

One library with one output format across all of those inputs is a real simplification. Whether it parses your specific documents correctly is a question only your documents can answer, which is what the browser demo is for.

The honest caveats

This is a company's open-source project, and the company sells a hosted version of the same conversion with OCR added. That is a normal arrangement, and the MIT licence means the self-hosted path is genuinely yours. It is still worth knowing which parts of the product description are free and which are paid.

The repository is 28 days old and had 74 open issues when this was written. Format conversion is the kind of work where the many rare kinds of broken file are the main difficulty, so expect to find files it handles badly and expect that to improve.

Try it on the most badly formatted document you own before you build anything on it.

Source

firecrawl/anydoc on GitHub, MIT licence. Star, fork, issue and date figures read from the GitHub API on 31 August 2026
Browser demo

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.