20 open-source OCR & PDF extraction tools worth knowing for RAG and agent pipelines in 2026
MaryamMiradi · x · 2026-11-23
A curated map of the 2026 open-source OCR + PDF extraction ecosystem for anyone building RAG/agent systems over real documents, organized into five categories:
- Native PDF extraction: PyMuPDF4LLM (Markdown/JSON), pdfplumber (text/coordinates/tables), pypdf — best when digital text is already clean.
- OCR engines: Tesseract, PaddleOCR, Surya (OCR + layout + reading order), EasyOCR, docTR.
- Document parsers: Docling, MinerU (formulas/tables/reading order), Marker, Unstructured (extraction + chunking).
- VLM-based parsing: olmOCR, OCRFlux (complex layouts, cross-page tables), MonkeyOCR, dots.mocr, Zerox — the category the author finds most interesting, as models now interpret document structure, not just characters.
- Specialists: GROBID (scientific papers), Camelot (PDF tables), Nougat (equations/academic PDFs).
Key point: the best OCR model isn't the best document pipeline — for production agents the goal is preserving enough structure and meaning for correct reasoning. A Docling vs MinerU vs Marker vs olmOCR benchmark may follow.
More from coding & agent
- Developer wires a receipt printer to a Grok bot that prints whenever he's summoned on X — mattyp · 2026-11-23
- Blogger woken at 3am by Neon-driven Gemini task results, gets hooked — TheTuringPost · 2026-11-23
- A personal agent is a system keeping an eventually consistent view of what matters to you — andreisavu · 2026-11-23
- VS Code Marketplace rejected 4 extension versions in 23 hours; install script rewritten to pass scanners — PawelHuryn · 2026-11-23
- Microsoft publishes Azure AI architecture design guide with Foundry Agent reference architectures — adnan_hashmi · 2026-11-23
- Perplexity Launches 'Teammates' AI Agents, Exclusive to $200/Month Max Subscribers — dnlkwk · 2026-11-23