Docling overview and key facts (October 2026)

Docling is an open-source library that parses PDFs, Office files, images, audio and more into a structured format for generative AI. Teams use it as the first step of a RAG pipeline, especially when documents must stay on their own machines.

Docling was initiated by the AI for knowledge team at IBM Research Zurich and is now a project of the LF AI & Data Foundation, published under the MIT license. It reads PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, email formats, LaTeX, ODF files and several XML schemas such as XBRL financial reports. Output formats include Markdown, HTML, DocTags and lossless JSON. Docling runs on macOS, Linux and Windows, offers a command line tool, an API server and an MCP server, and integrates with LangChain, LlamaIndex, Crew AI and Haystack. As of September 28, 2026 the repository had about 68,100 GitHub stars.

• Understands PDF layoutLayout models detect reading order, headings, tables, formulas and code blocks on each page. Downstream search and chat get cleanly structured text instead of a jumble of lines.
• Reads scanned documentsOCR turns scanned PDFs and images into text, and the GraniteDocling visual language model can process pages end to end. Paper archives become usable for RAG.
• Turns charts into dataBar, pie and line charts can be converted into tables or code with a description. Numbers that only existed in images become searchable.
• Transcribes audio and videoAutomatic speech recognition creates transcripts of audio and video files, with keyframes for video. Recorded meetings or trainings can join the same knowledge base as documents.

• Runs fully locally, including in air-gapped environments, which suits confidential documents.
• Governed by a neutral foundation and licensed under MIT.
• Broad format coverage including XBRL financial reports, email files and Apple Pages.

Started by IBM Research Zurich and now hosted by the LF AI & Data Foundation, without a country focus.

Docling is a Python library that your team installs, operates and updates; Python 3.10 or newer is required. OCR and visual models need more compute, so test throughput on your document volume. Some announced features such as metadata extraction are listed as coming soon and are not available yet.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews