anydoc converts documents in 5ms

Firecrawl released anydoc, an open-source Rust library that converts various files, including office documents, PDFs, and EPUBs, into GitHub Flavored Markdown for LLM input.

Format-specific parsers generate a common Document model, which then passes through a single Markdown serializer. This allows fixes for table escaping, heading anchors, and footnote handling to be applied consistently across multiple file formats.

• Supported formats: 14
• Median conversion time: under 4.4–5ms
• Implementation: Pure Rust, no ML models or external service calls
• Distribution: Rust crate, Node.js package, Python wheel, WebAssembly
• Features provided: CLI and Agent Skill for coding agents

In the repository benchmarks, 100 documents across 14 formats were compared, with anydoc achieving an overall quality score of 81 and a median conversion time of 4.4ms. However, the comparison data and evaluation methodology are from the project's own published benchmark, and the document corpus is not redistributed.