Docling

Docling is an open-source document processing library that parses diverse document formats and provides integrations with the generative AI ecosystem. It is built for developers who need to turn PDFs, Office files, and other documents into structured data for use in AI applications. The project is hosted by the LF AI & Data Foundation and originated from the AI for knowledge team at IBM Research Zurich.

A developer installs Docling with pip install docling, which requires Python 3.10 or higher and runs on macOS, Linux, and Windows for x86_64 and arm64 architectures. The command-line tool converts a document, such as docling followed by a URL or file path, into a Markdown file containing structured content, and the DocumentConverter class in Python offers the same conversion with more configuration options, exporting results such as Markdown directly in code.

Docling parses more than a dozen formats including PDF, DOCX, PPTX, XLSX, HTML, EPUB, email formats, images, video files, and audio, with advanced PDF understanding covering page layout, reading order, table structure, code, formulas, and image classification. It exports to Markdown, HTML, DocTags, and lossless JSON, supports extensive OCR for scanned documents, and plugs into frameworks such as LangChain, LlamaIndex, Crew AI, and Haystack for agentic AI.

Docling's codebase is released under the MIT license, while individual model licenses apply separately for model usage. It supports local execution for sensitive data and air-gapped environments, and Python 3.9 support was dropped as of Docling version 2.70.0. The project can be run as a service with the docling-serve API server, connected to through an MCP server, or used from a simple CLI, and parsing of video files, ODF documents, and XBRL financial reports was added in recent updates.

Use cases

• Preparing scanned PDFs for a generative AI pipeline
• Extracting tables and charts from financial reports