Docling
An open-source toolkit, started at IBM Research Zurich, that converts PDFs, Word, PowerPoint, Excel, HTML, images and more into clean Markdown or lossless JSON for AI apps. It understands page layout and tables, runs OCR on scans, works fully on your own machine and plugs into LangChain, LlamaIndex and Haystack.
68.5k (docling-project/docling)
GitHub · Oct 2026
MIT
GitHub · Oct 2026
Tool
AI & Machine Learning
Docling project (LF AI & Data Foundation)
An afternoon to try on your documents
Intermediate
Free
What Docling is
Docling is an open-source document conversion toolkit for generative AI. It was started by the AI for Knowledge team at IBM Research Zurich and is now hosted by the LF AI & Data Foundation. Its job is the unglamorous step that decides the quality of most RAG and extraction systems: getting correct, well-structured text out of messy documents.
What it handles
Docling reads a wide range of formats, including PDF, DOCX, PPTX, XLSX, HTML, EPUB, images such as PNG, TIFF and JPEG, LaTeX, emails and even audio files. Its PDF understanding includes page layout, reading order, tables, code and formulas, and it offers OCR (optical character recognition) for scanned pages and images. It can also use vision language models for hard pages.
Output can be Markdown, HTML, DocTags or lossless JSON, all from one unified DoclingDocument format, so you can feed clean Markdown to a model or keep full structure for your own processing.
Why it matters when choosing a stack
- Privacy: it runs locally, including in air-gapped environments, so sensitive documents never leave your servers. That makes it a strong default when a hosted parser is not allowed.
- Integrations: ready-made connectors for LangChain, LlamaIndex, Crew AI and Haystack.
- Quality first: tables split across pages and two-column layouts are where cheap PDF-to-text fails. Test Docling on your ugliest documents and compare.
Getting started
Install with pip install docling, then convert a file or URL from the command line with docling <path or url>, which writes a Markdown file. A few lines of Python do the same inside your app. Look at the output yourself before building on it.
Prerequisites and time
Basic Python and a command line. A first conversion takes minutes; testing it properly on a set of your real documents takes an afternoon. Large batches and OCR run faster with a GPU but work on a CPU.
Free or paid?
Free and open source under the MIT licence.
Best for: RAG and extraction projects over PDFs and Office files, especially when data must stay on your own machines.
Ready to start?
Opens on Docling project (LF AI & Data Foundation) in a new tab.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.