Docling

Docling converts PDFs, Office documents, and images into clean structured text, Markdown, or JSON for AI pipelines. It handles tables and layout well, making it ideal for preparing documents for retrieval and RAG. It is an MIT-licensed project under LF AI and Data.

Docling is an open-source tool that turns messy documents into clean, structured text that AI systems can actually use. Point it at a PDF, Word, PowerPoint, Excel, HTML, image or many other formats and it parses the layout, reading order, tables, headings, code and formulas, then exports the result to Markdown, HTML, lossless JSON or its own unified document format. This makes it a common first step for retrieval-augmented generation (RAG) and other AI pipelines, where quality depends on how faithfully source files are converted. It offers advanced PDF understanding, extensive OCR for scanned documents, and optional vision-language and speech models, and it can run locally for sensitive or air-gapped data. Docling is distributed as a Python library with a simple CLI and plugs into frameworks like LangChain and LlamaIndex. Started by IBM Research and now an LF AI & Data project, it is used by developers and data teams building document-driven AI.