pdfmux

Our take

A self-auditing PDF extraction pipeline: it routes each page to the best of several rule-based backends — PyMuPDF, Docling, RapidOCR, Marker, and others — scores per-page confidence, and re-extracts pages that fail its own audit, with optional BYOK LLM fallback for the hardest ones. That confidence-scored, fail-loudly design is genuinely distinctive for RAG ingestion, and it runs locally with no required API keys unless you enable LLM backends. It is a young single-maintainer project with marketing-forward benchmark claims that deserve independent verification. For anyone whose pipeline silently ingests garbled PDF text today, worth a trial.

reviewed by hand · 2026-07-05

Something wrong or dead here? Report it

Verification record

last verified

25 days ago

github stars

82

last commit

28 days ago

archived

no

license

MIT

in registry since

2026-06-15

github.com/NameetP/pdfmux