ocr-cli

A local-first OCR pipeline that replaces a chain of fish scripts and Python tools with a single Rust binary. Sends PDFs inline (base64) to Mistral OCR, eliminating the need for S3/Lambda/AWS infrastructure.

How it works

ocr-cli paper.pdf --lead 2 --trail 3
         │
         ├─ 1. Truncate pages ─────────── lmpdf (Pdfium bindings)
         ├─ 2. Extract title ──────────── lmpdf + llm-rs (OpenAI)
         ├─ 3. Mistral OCR ───────────── reqwest + base64 (FileChunk POST)
         ├─ 4. Post-process ──────────── Rust port of mistral-postproc
         └─ 5. Move outputs ──────────── vault + archive

Usage

ocr-cli paper.pdf
ocr-cli paper.pdf --lead 2 --trail 3
ocr-cli *.pdf --dry-run

Options

Flag	Default	Description
<files...>	required	PDF files to process
--lead N	0	Pages to remove from the front
--trail N	0	Pages to remove from the end
--vault PATH	~/Documents/Ekuro/	Output directory for markdown notes
--papers PATH	~/Documents/Papers/	Archive directory for source PDFs
--model MODEL	gpt-4o-mini	LLM model for title extraction
--dry-run	off	Stop after title extraction, print proposed filename

Requirements

• Pdfium shared library (searched at /opt/homebrew/lib/libpdfium.dylib or via PDFIUM_PATH)
• MISTRAL_API_KEY environment variable
• OPENAI_API_KEY environment variable

Local dependencies

This project uses two sibling Rust crates via path dependencies:

• lmpdf — Pdfium bindings for PDF manipulation (truncation, text extraction)
• llm-rs — LLM provider abstraction for title extraction

Building

cargo build --release

License

MIT