Learn how IBM Docling (docling ibm) evolved from groundbreaking document layout research at IBM Zurich into the open-source standard for AI document processing.
Document parsing for Generative AI should be fast, transparent, local-first, and accurate across complex scientific papers, financial spreadsheets, and multi-column reports. IBM Docling was built to eliminate proprietary cloud API lock-in by providing a 100% free, MIT-licensed parsing engine that runs natively on standard CPU and GPU hardware.
Researchers at IBM Research Zurich's Deep Search for Scientific Data (DS4SD) group created TableFormer and advanced neural layout models to solve complex PDF table extraction and multi-column reading order recovery.
IBM open-sourced Docling on GitHub under the permissive MIT License, publishing the landmark research paper arXiv:2408.09869. The library rapidly gained thousands of GitHub stars and PyPI downloads.
Docling joined the Linux Foundation's LF AI & Data Foundation as a hosted open project. IBM integrated Granite Docling vision-language models, launched docling-serve REST API, and built native Model Context Protocol (docling-mcp) support for AI desktop agents.