Docling

by IBM Research

Open SourceSelf-HostedMIT

Document conversion toolkit that transforms PDFs into structured, AI-ready formats. Features custom layout analysis and TableFormer for complex table extraction.

Layout AnalysisTable ExtractionDocument Conversion
GitHubđź“„ arXiv PaperDocumentation

Overview

Docling is IBM Research's document conversion toolkit designed for AI pipelines. Unlike pure OCR tools, Docling focuses on preserving document structure and hierarchy—making outputs suitable for RAG, extraction, and indexing workflows.

Key components include a custom RT-DETR-based layout model trained on IBM's DocLayNet dataset, TableFormer for complex table structure recognition, and a custom PDF parser (docling-parse) built for production performance.

Docling produces a structured DoclingDocument data model that maintains provenance for each extracted element, enabling reliable export to Markdown, JSON, or HTML.

Strengths

  • Excellent table structure recognition (TableFormer)
  • Preserves document hierarchy and provenance
  • Purpose-built for RAG and AI pipelines
  • Handles PDFs, DOCX, PPTX, HTML, and images
  • Active IBM Research backing

Limitations

  • Not a standalone OCR engine (uses external OCR)
  • Heavier resource requirements than simple OCR
  • Relatively new with evolving API

Best Use Cases

  • RAG document preparation
  • Complex PDF to structured data conversion
  • Enterprise document processing pipelines
  • Research paper and technical document extraction