Skip to main content
Docling is available as an official LlamaIndex extension, providing two powerful components: the Docling Reader for document loading and the Docling Node Parser for intelligent chunking.

​
Overview

The LlamaIndex Docling integration provides:
  • Docling Reader - Load documents with high-fidelity structural preservation
  • Docling Node Parser - Parse documents into LlamaIndex nodes with structure awareness
  • Lossless Serialization - Preserve complete document structure as JSON
  • Flexible Export - Export to simplified formats like Markdown when needed

​
Installation

pip install llama-index-readers-docling llama-index-node-parser-docling

​
Components

​
Docling Reader

The Docling Reader loads document files and populates LlamaIndex Document objects with Docling’s rich data model.

​
Basic Usage

from llama_index.readers.docling import DoclingReader

# Create reader
reader = DoclingReader()

# Load documents
documents = reader.load_data(file_path="document.pdf")

# Access document content
for doc in documents:
    print(doc.text)
    print(doc.metadata)

​
Export Formats

from llama_index.readers.docling import DoclingReader
from docling.datamodel.base_models import FormatOptions

# Export as Markdown
reader = DoclingReader(
    export_type="markdown"
)
docs = reader.load_data(file_path="document.pdf")

# Export as JSON (lossless)
reader = DoclingReader(
    export_type="json"
)
docs = reader.load_data(file_path="document.pdf")

​
Docling Node Parser

The Docling Node Parser uses knowledge of Docling’s format to intelligently parse documents into LlamaIndex Node objects for downstream usage.

​
Basic Usage

from llama_index.readers.docling import DoclingReader
from llama_index.node_parser.docling import DoclingNodeParser

# Load documents with Docling Reader
reader = DoclingReader(export_type="json")
documents = reader.load_data(file_path="document.pdf")

# Parse into nodes
parser = DoclingNodeParser()
nodes = parser.get_nodes_from_documents(documents)

# Nodes are ready for embedding and retrieval
for node in nodes:
    print(node.text)
    print(node.metadata)

​
Advanced Parsing Options

from llama_index.node_parser.docling import DoclingNodeParser

# Configure parser
parser = DoclingNodeParser(
    include_metadata=True,
    chunk_size=1024,
    chunk_overlap=128
)

nodes = parser.get_nodes_from_documents(documents)

​
Complete RAG Pipeline

Here’s a full example combining both components: 
from llama_index.readers.docling import DoclingReader
from llama_index.node_parser.docling import DoclingNodeParser
from llama_index.core import VectorStoreIndex
from llama_index.embeddings.openai import OpenAIEmbedding

# 1. Load documents
reader = DoclingReader(export_type="json")
documents = reader.load_data(file_path="document.pdf")

# 2. Parse into nodes
parser = DoclingNodeParser()
nodes = parser.get_nodes_from_documents(documents)

# 3. Create embeddings and index
embed_model = OpenAIEmbedding()
index = VectorStoreIndex(nodes, embed_model=embed_model)

# 4. Query the index
query_engine = index.as_query_engine()
response = query_engine.query("What is the main topic of this document?")
print(response)

​
Features

Structure-Aware

Preserves document hierarchy and relationships

Lossless Export

JSON export maintains complete document structure

Smart Chunking

Node parser respects document structure when chunking

Rich Metadata

Includes page numbers, headings, and structural information

​
Use Cases

​
Knowledge Base RAG

# Process multiple documents
reader = DoclingReader(export_type="json")

documents = []
for file in ["doc1.pdf", "doc2.docx", "doc3.pptx"]:
    docs = reader.load_data(file_path=file)
    documents.extend(docs)

parser = DoclingNodeParser()
nodes = parser.get_nodes_from_documents(documents)

index = VectorStoreIndex(nodes)

​
Table-Aware Retrieval

# Docling preserves table structure
reader = DoclingReader(
    export_type="markdown",
    pipeline_options={"do_table_structure": True}
)

documents = reader.load_data(file_path="report.pdf")
parser = DoclingNodeParser()
nodes = parser.get_nodes_from_documents(documents)

# Tables are preserved in node content
for node in nodes:
    if "table" in node.metadata.get("type", "").lower():
        print("Found table:", node.text)

​
Integration Benefits

1

Official Components

Maintained as official LlamaIndex integrations
2

Two-Component System

Reader and Parser work together seamlessly
3

Format Flexibility

Choose between lossless JSON or simplified Markdown
4

Production Ready

Used in real-world RAG applications

​
Resources

Tutorial

Step-by-step guide

Reader Docs

API reference for Docling Reader

Parser Docs

API reference for Node Parser

GitHub

Source code

​
PyPI Packages

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us