Trial
Docling is an open-source toolkit (MIT, LF AI & Data project) for document parsing in AI pipelines: PDF, DOCX, PPTX and images are converted into structured output with tables, layout, reading order and provenance - the typical ingestion step before chunking and indexing in RAG systems.
- One of the most widely used options for document ingestion (>40k GitHub stars, millions of monthly downloads); integrations exist for LangChain, LlamaIndex and Haystack. Comparable production-grade alternatives are Marker and MinerU.
- Since January 2026 the compact Granite-Docling-258M VLM (Apache 2.0) handles complex layouts locally - relevant for data-sovereign pipelines without external OCR services.