// the find
Nutlope/llama-ocr
Document to Markdown OCR library with Llama 3.2 vision
A thin TypeScript wrapper around Together AI's hosted Llama 3.2 Vision endpoint that converts images to markdown. Basically one function, one dependency. For developers who want OCR without standing up their own infrastructure and don't mind routing their documents through Together AI's servers.
The API surface is about as minimal as it gets — one function, one required parameter, done. Together AI's free tier makes zero-cost experimentation real, not just a marketing claim. The 90B model default is a reasonable choice; it's the one that actually produces usable output on messy documents.
This is a wrapper around a third-party API, not an OCR library — every document you process leaves your environment. No PDF support yet despite being on the roadmap since launch; the workaround (screenshot each page) is left as an exercise. Last commit was January 2025 and the roadmap items are still unchecked, which suggests the project has stalled. At 26 lines of actual source code, you're really just installing a dependency to avoid writing a fetch call.