Docling: RAG Frameworks

Founded 2024|Zurich, Switzerland (IBM Research)|Part of IBM Research people|Free open-source (Apache 2.0)

What is Docling?

Docling is IBM Research's open-source document conversion toolkit, designed for AI-driven workflows that need clean, structured data from messy documents. It converts PDFs, DOCX, PPTX, HTML, images, and more into JSON or markdown while preserving layout, tables, equations, code blocks, and lists.

In 2026, IBM released Granite-Docling-258M — an ultra-compact open-source vision-language model purpose-built for document conversion under Apache 2.0. Granite-Docling delivers significantly better recognition accuracy than traditional OCR by retaining the original layout structure and identifying complex elements like tables, math, and code blocks. The output uses DocTags, a universal markup format developed by IBM Research that captures every page element and its contextual relationships.

Strategically, IBM has positioned Docling for production use: launched the Docling OpenShift Operator with Red Hat (targeting banks), donated the project to the Linux Foundation's Agentic AI Foundation alongside BeeAI and Data Prep Kit, and is integrating it across Red Hat and IBM Cloud document workflows. Free, fully open-source, and self-hostable.

Key features

Core capabilities this platform advertises.

Converts PDFs, DOCX, PPTX, HTML, images to structured JSON/markdown
Granite-Docling-258M VLM model purpose-built for document understanding
DocTags markup preserves layout, tables, equations, code blocks
Apache 2.0 — fully open-source and self-hostable
Production deployment via Docling OpenShift Operator (Red Hat)

Strengths and tradeoffs

What this tool does well, and the limitations to keep in mind.

Pros

Purpose-built VLM beats general-purpose OCR on complex layouts
Apache 2.0 license — fully open and self-hostable
IBM-grade engineering with Linux Foundation governance
DocTags standardized markup makes output portable across tools
Production deployment story via Red Hat / OpenShift

Cons

Setup complexity higher than hosted document APIs
Granite-Docling-258M still requires GPU for fast inference
Less polished UX than cloud DocAI services from Google/AWS
Smaller ecosystem than Unstructured.io for non-IBM stacks

Plans & pricing

What's included in each plan, and how the tiers compare.

Open Source (Apache 2.0)

Forever

Full toolkit + Granite-Docling-258M weights
Self-hostable on any infrastructure
DocTags universal markup output
All conversion features (PDF, DOCX, PPTX, etc.)

Red Hat OpenShift Operator

Bundled with OpenShift

Production deployment via OpenShift
Targeted at banking and regulated industries
Red Hat enterprise support

View official pricing page

Common use cases

RAG and AI engineering teams that need accurate, structured ingest of PDFs, DOCX, and complex documents into LLM pipelines

RAG ingest pipelines that need clean structured text
Financial and legal document parsing (banking)
Scientific paper ingestion preserving equations and tables
Enterprise knowledge base ingestion at scale
On-prem document conversion for regulated environments

Run Docling in production with Respan

10KFree traces/mo

1,000+Models

5 minSetup

Respan lets you trace LLM and agent calls across any model or framework, A/B test prompts on production traffic, and route requests across 1,000+ models through one gateway.

Try Respan free