Docling

An open source document parser that runs locally on macOS, Linux and Windows, with OCR, structured exports and integrations for AI applications.

Docling turns documents into structured content for developers building search, retrieval and AI applications. It handles page structure and scanned text so applications can work with more than a file's plain text.

PDF structure and OCR: Extract reading order, tables, code and formulas, classify images, and read scanned PDFs and images.
Document formats: Process Word documents, spreadsheets, presentations, HTML, EPUB, Apple office files and email, plus specialized XML for patents, journal articles and XBRL financial reports.
Audio, video and charts: Transcribe audio, extract video transcripts and representative keyframes, and convert bar, pie and line charts into tables or code with descriptions.
Structured exports: Bring different file types into a shared DoclingDocument representation and export Markdown, HTML, WebVTT or lossless JSON.
AI integrations: Connect document processing to LangChain, LlamaIndex, Crew AI and Haystack, or use the MCP server with agents; visual language model support includes GraniteDocling.

Docling runs locally on macOS, Linux and Windows, including in air-gapped environments for sensitive documents. It's an open source Python library under the MIT license, with a command-line interface and a separate docling-serve API server for running it as a service.