Developers preparing PDFs, Office files, and other documents for LLM and RAG pipelines.

Docling is an IBM Research open-source toolkit that converts PDF, DOCX, PPTX, HTML, and other formats into structured Markdown or JSON using layout-aware vision models rather than pure OCR.

Self-hosted as a Python library, optional Docker server, with Granite Docling VLM for enhanced parsing.

Open-sourced by IBM Research and hosted under the LF AI & Data Foundation, active through 2026.

It exists to give AI pipelines fast, accurate document-to-text conversion without proprietary OCR services.

Integrates with LlamaIndex, LangChain, and RAG chunking pipelines like Chonkie.

Open source, MIT license, self-hosted; no API key required; documents processed entirely on local or self-managed infrastructure.