Mistral OCR

A command-line tool to convert PDF, image, and document files into Markdown using Mistral AI's OCR API. Written in Rust.

Features

• Supports PDF, images (jpg, png, gif, bmp, tiff, webp), and office documents (docx, odt, pptx, xlsx, etc.)
• Office documents are automatically converted to PDF via LibreOffice
• Multiple image handling modes:
  • separate — save images as files in a _images/ directory
  • inline — embed images as base64 data URIs (single self-contained .md file)
  • zip — bundle markdown + images into a single .zip archive
• Single static binary, no runtime dependencies

Installation

Prerequisites

• A Mistral API key
• LibreOffice (only needed for office document conversion)

Pre-built binary (recommended)

cargo binstall mistral_ocr

From source

cargo install mistral_ocr

Configuration

Set your Mistral API key as an environment variable:

export MISTRAL_API_KEY='your-api-key-here'

Usage

Basic usage (text only)

mistral_ocr document.pdf

Process an image

mistral_ocr photo.png

Process a Word document

mistral_ocr report.docx --images inline

Extract images as separate files

mistral_ocr document.pdf --images separate

Single self-contained markdown file

mistral_ocr document.pdf --images inline

Bundle everything into a zip

mistral_ocr document.pdf --images zip

All options

<INPUT>               Path to the input file (required)
--model <MODEL>       Mistral OCR model name (default: mistral-ocr-latest)
--images <MODE>       How to handle images: none, separate, inline, zip (default: none)
--output <PATH>       Where to write the output (default: ocr_output.md)

Supported file types

License

MIT — see LICENSE.