markitdown-rs

Microsoft's markitdown as one Rust binary: convert PDF, Word, PowerPoint, Excel, HTML, EPUB, Outlook and web pages to Markdown for LLMs and RAG pipelines. Same output, byte for byte. No Python. 23–112× faster.

Download a binary GitHub Comparison

$ cargo install markitdown-rs
$ markitdown order.pdf
Purchase order summary
| Item     | Qty | Unit price | Total |
| -------- | --- | ---------- | ----- |
| Widget 1 | 1   | 1.50       | 1.50  |
| Widget 2 | 2   | 3.00       | 6.00  |
$ markitdown https://en.wikipedia.org/wiki/Markdown | head -3
# Markdown

Plain text markup language

Why markitdown-rs

Drop-in

Same command, same flags, same Markdown. Every test file is checked byte for byte against Python markitdown 0.1.9.

One small binary

A 4.5 MB binary instead of a 328 MB Python environment with pdfminer, pandas, mammoth, magika and onnxruntime.

Fast

23–112× faster end to end; a 100-page PDF in 0.18 s instead of 5.9 s.

Every local format

PDF, DOCX, PPTX, XLSX, XLS, HTML, EPUB, Outlook MSG, CSV, ZIP, RSS, Jupyter, text, images and audio metadata, plus http(s) URLs, Wikipedia and YouTube pages.

Benchmarks

File	Python markitdown	markitdown-rs	Speedup
markitdown `test.docx`	572.9 ms	5.1 ms	112×
`large.docx` (1000 sections)	11,588.9 ms	374.2 ms	31×
markitdown `test.pptx`	599.6 ms	5.7 ms	105×
`deck.pptx` (300 slides)	1,573.5 ms	69.9 ms	23×
markitdown `test.pdf`	649.4 ms	7.0 ms	93×
`report.pdf` (100 pages)	5,864.0 ms	180.3 ms	33×

CLI wall time, median of repeated runs on an Apple M2 Max. Method and raw numbers: README.

Compared with

	markitdown-rs	microsoft/markitdown	markitdown (crates.io)	xberg (kreuzberg)
Output identical to markitdown	yes, tested	reference	no	no (own format)
Install	one binary	pip + Python	cargo	pip / cargo / bindings
Python needed	no	yes	no	no
URL input	yes	yes	no	yes (feature flag)

Full table: docs/comparison.md.