66.1k stars · MIT · v2.126.0 (2026-09-04)
IBM's document parser — PDF, DOCX, PPTX, XLSX, EPUB and more into structured Markdown, locally on CPU.
▶Repo detailsthe review · specs · pros & cons · install
What it is
A document parser from IBM Research, now under the LF AI & Data Foundation, that converts PDF, DOCX, PPTX, XLSX, HTML and images into clean structured Markdown. It preserves reading order, headings and table structure, and runs locally on CPU.
Why it matters
Grasppy's pitch already says "chat, discussion or document", but the document half only becomes real when someone can drop a 90-page PDF on the page and get a map back. That is the moment the addressable audience changes shape: not just people with long AI conversations, but anyone holding a research report, a contract, a set of meeting minutes or a textbook chapter. Docling is the shortest credible path there, and because it runs on CPU locally, no document ever leaves your VPS — a real selling point for anyone considering pasting something confidential. Its Markdown output feeds straight into Chonkie (Edition 2).
- MIT and local — no per-page API fee and no document leaving your server
- Keeps tables and headings intact, which is exactly what a subtopic map depends on
- Enormous, very active project with institutional backing behind it
- First run downloads several hundred MB of layout and table models — plan the disk and the cold start
- A long scanned PDF takes minutes on CPU. This must be a background job with a progress indicator, never a request
- Fast-moving version numbers; pin the version or an upgrade will change your output shape
Its Markdown output feeds straight into Chonkie (Edition 2).
source venv/bin/activate pip install docling docling my-document.pdf --to md # first run downloads the models
