quill-of-hermes: LAN OCR that eats PDFs and multi-page TIFFs
OCR as a service usually means cloud APIs, per-page billing, and your documents on someone else’s disk. This is the local counterargument: a FastAPI wrapper around OvisOCR2 running GGUF via llama.cpp’s llama-mtmd-cli, serving your LAN. POST an image, a PDF, or a multi-page TIFF; get back clean Markdown.
The design decisions worth noting: each page is sent to the model independently and results are joined with a horizontal rule, so a 50-page PDF degrades gracefully instead of failing whole. Jobs are serialized to keep the GPU sane, there’s a five-minute timeout per page so one bad scan can’t wedge the queue, and temp files are cleaned unconditionally. The service deliberately stays separate from llama-swap because multimodal llama-server support for this model wasn’t proven — llama-mtmd-cli was.
Features:
- Accepts PNG/JPEG/WebP, PDF (150 DPI render, up to 50 pages), and multi-page TIFF via ImageMagick
- Returns structured Markdown plus
page_countand totalelapsed_seconds - OpenAI-shape simplicity: single
POST /api/ocrendpoint, 100 MB upload limit - Serialized job queue, per-page timeouts, automatic temp-file cleanup
- systemd user service with restart-on-failure; health check at
/health - Companion Hermes skill calls the API and visually verifies uncertain handwriting, equations, names, and numbers rather than trusting raw output
- LAN-only by design — no auth, so keep it behind your firewall
Source: github.com/rahlquist/quill-of-hermes · MIT licensed.