What is Docling?
Docling is an open-source document parsing and conversion library developed by IBM. It transforms complex documents (PDFs, DOCX, PPTX, images) into machine-readable formats like JSON and Markdown, making unstructured data ready for Retrieval-Augmented Generation (RAG), question-answering systems, and other GenAI applications.
"Docling enables AI-powered document understanding with advanced layout analysis, table extraction, and seamless integration with LangChain and LlamaIndex."
Multi-Format
PDF, DOCX, PPTX, HTML
AI-Powered
DocLayNet & TableFormer
Local Execution
Runs on laptop
MIT License
Open Source
Core Features
Multi-Format Document Parsing
Docling processes a wide variety of document formats and converts them into structured, machine-readable JSON and Markdown. Supported formats include PDFs (native and scanned), Word documents, PowerPoint presentations, images, and HTML.
AI-Powered Layout Analysis (DocLayNet)
Uses IBM's DocLayNet model for sophisticated page layout analysis. This computer vision approach identifies document structure, reading order, headers, paragraphs, and other elements—often bypassing traditional OCR for improved speed and accuracy.
Table Recognition (TableFormer)
TableFormer provides high-accuracy extraction and structuring of tables from documents. It preserves cell relationships, headers, and data integrity—critical for enterprise documents with complex tabular data.
Unified Docling Document Format
All processed documents are transformed into a consistent "Docling Document" structure. This uniform representation includes text, tables, images, document hierarchy, and layout metadata—making downstream processing predictable and reliable.
Granite-Docling Model
Ultra-Compact Vision-Language Model (VLM)
Granite-Docling is a 258M parameter open-source VLM designed to enhance the Docling pipeline. It enables end-to-end document understanding in a single pass, handling:
- Inline and floating math equations
- Code blocks and technical content
- Multilingual support (Arabic, Chinese, Japanese)
- Complex document layouts
Quick Start
# Install Docling
pip install docling
# Basic usage - convert PDF to Markdown
from docling.document_converter import DocumentConverter
# Initialize converter
converter = DocumentConverter()
# Convert a document
result = converter.convert("path/to/document.pdf")
# Export to Markdown
markdown = result.document.export_to_markdown()
print(markdown)
# Export to JSON
json_output = result.document.export_to_json()
print(json_output)Tip: Docling runs locally on standard hardware. No cloud API required—ensuring data privacy for sensitive enterprise documents.
Integration with LLM Frameworks
Docling is designed for seamless integration with popular LLM frameworks, making it ideal for building RAG pipelines and document Q&A systems.
LangChain Integration
from langchain_community.document_loaders import DoclingLoader
loader = DoclingLoader(file_path="report.pdf")
docs = loader.load()
# Use with your RAG pipeline
for doc in docs:
print(doc.page_content[:200])LlamaIndex Integration
from llama_index.readers.docling import DoclingReader
reader = DoclingReader()
documents = reader.load_data(
file_path="report.pdf"
)
# Index and query
index = VectorStoreIndex.from_documents(documents)Use Cases
RAG Pipelines
Prepare documents for retrieval-augmented generation
Document Q&A
Extract answers from enterprise documents
Data Extraction
Extract tables and structured data from PDFs
Contract Analysis
Parse legal documents for clause extraction
Knowledge Base
Build searchable archives from document libraries
Report Processing
Automate analysis of financial reports