Convert complex document structures into data that generative AI can understand
If you have PDF, Word, or Excel files that are hard to feed directly into AI, this tool bridges the gap. It analyzes the reading order and structures like tables and formulas, converting them into a format suitable for AI processing. This allows developers to avoid writing separate preprocessing code.
Recently supported formats and features
Recent updates added the ability to process video files (MP4, AVI, etc.) with subtitles. ODF documents, XBRL financial reports, emails, EPUB e-books, and Apple Pages, Numbers, and Keynote files are now supported. Chart understanding has also been enhanced, allowing bar and pie charts to be converted into tables or code with added descriptions. Metadata extraction and complex chemical structure understanding features are scheduled for release soon.
Installation and basic usage
Install it in a Python 3.10+ environment using the pip install docling command. It supports x86_64 and arm64 architectures on macOS, Linux, and Windows. Using the command-line interface (CLI), you can generate a Markdown file in the current directory by entering a document URL or path after the docling command. For Python code, the recommended approach is to instantiate the DocumentConverter class and call the convert method to export the results as Markdown. To use a VLM model, add the --pipeline vlm option to the CLI.
Pre-usage considerations
This tool is released under the MIT license, but you must check the license of individual models when using them. Python 3.9 support was discontinued starting from version 2.70.0, so you must use version 3.10 or higher. The project is hosted by the LF AI & Data Foundation and was initiated by the AI for knowledge team at IBM Research Zurich. It is advisable to use local execution features when handling sensitive data, and to verify XML schema support in advance when dealing with specific industry standards (such as patents or financial reports).
