Docling
Docling is an open-source document conversion toolkit whose tagline is “Get your documents ready for gen AI”. It parses documents into a unified DoclingDocument representation and exports Markdown, HTML, WebVTT, DocLang, DocTags or lossless JSON.
Repository snapshot (GitHub API, 2026-11-19)
| Item | Detail |
|---|---|
| Repo | docling-project/docling, MIT (formerly under the DS4SD organisation) |
| Created | 2024-07-09 |
| Activity | last push 2026-11-19; latest release v2.151.1 (2026-11-19) |
| Size | 70,789 stars, 5,385 forks |
| Governance | The README shows an LF AI & Data badge and an OpenSSF Best Practices badge; no maintainer is named in the part read |
| Paper | arXiv 2408.09869 |
Capabilities (README)
- Advanced PDF understanding: layout, reading order, table structure, code, formulas and image classification.
- Many inputs: PDF, DOCX, PPTX, XLSX, HTML, EPUB, Apple Pages, Numbers and Keynote, audio (WAV, MP3), WebVTT, email, images, LaTeX and plain text, plus XML schemas such as USPTO patents, JATS articles and XBRL.
Where it fits
Docling sits at the front of a RAG pipeline: turn PDFs and office files into clean structured text before chunking and embedding (vector-embeddings). Compare unstructured. Output feeds stores such as faiss, opensearch and azure-ai-search.
Sources
- Docling README (first 3,500 characters read) and GitHub API (2026-11-19)