Docling

Docling parses PDFs, Word files, slides and more, producing clean markdown or JSON with tables, reading order and figures preserved. It was released by IBM Research and is now maintained under the Linux Foundation.

It runs locally, which matters when documents are confidential. Integrations with LangChain, LlamaIndex and others make it easy to drop into a RAG pipeline.

Accuracy is competitive with paid parsers on many layouts. If you want a free, self-run solution for document ingestion, start here.