Docling is a Python-based document processing framework designed to prepare diverse document formats for generative AI applications. The project parses and converts multiple file types including PDF, DOCX, PPTX, XLSX, HTML, EPUB, WAV, MP3, WebVTT, email formats (EML, MSG), images (PNG, TIFF, JPEG), LaTeX, plain text, and ODF files (ODT, ODS, ODP). It also supports XBRL documents for financial reporting and provides extensive OCR capabilities for scanned PDFs and images.
The core strength of Docling lies in its advanced PDF understanding capabilities, which extract page layout, reading order, table structure, code blocks, formulas, and image classifications. The framework produces a unified DoclingDocument representation format that can be exported to multiple formats including Markdown, HTML, WebVTT, DocLang, DocTags, and lossless JSON. It supports application-specific XML schemas such as DocLang, USPTO patents, JATS articles, and XBRL financial reports.
Docling integrates seamlessly with the generative AI ecosystem through plug-and-play connectors for LangChain, LlamaIndex, Crew AI, and Haystack. The project includes support for Visual Language Models like GraniteDocling, automatic speech recognition for audio files, and a Model Context Protocol (MCP) server for agent connectivity. Users can run Docling locally for sensitive data and air-gapped environments, or deploy it as a service using the API server (docling-serve). A command-line interface provides simple document conversion capabilities.
Recent additions to the project include parsing of ODF files, XBRL documents, email files, EPUB e-books, plain-text and Markdown supersets, and chart understanding that converts charts into tables or code with detailed descriptions. Upcoming features include metadata extraction for titles, authors, references, and language detection, as well as complex chemistry understanding for molecular structures.
The repository shows active development and community engagement.
Docling is hosted as a project within the LF AI & Data Foundation and was initiated by the AI for knowledge team at IBM Research Zurich. The codebase is released under the MIT license, with individual model licenses referenced separately. The project maintains comprehensive documentation covering installation, usage, configuration, recipes, and extensions, along with practical examples demonstrating various application use cases.
