What it does
Docling processes PDFs, office files, web documents, images, audio, and other formats. It provides a CLI and exports structured representations such as Markdown, HTML, and JSON.
Convert a local path or URL through the CLI or Python DocumentConverter, then export the unified document as Markdown, HTML, JSON or another supported representation. The CLI's basic example writes Markdown in the current directory; optional OCR, table/layout and VLM pipelines change extraction behavior. Local model execution can support offline use, but weights are fetched on first use when no artifacts path is supplied. Prefetched local artifacts avoid that step. Processing user data through remote APIs requires an explicit remote-services opt-in. Code is MIT-licensed, while each model keeps its own licence; Python 3.10+ is required.
Strengths
- Unifies several input families with layout-aware document processing.
Watch out for
- Individual model licenses are separate from the codebase's MIT license.
Best for
- Preparing heterogeneous documents for search and analysis pipelines
Sources
- GitHub - docling-project/docling: Get your documents ready for gen AI · GitHub (Oct 26, 2026)
- Pipeline options - Docling (Oct 26, 2026)