Turn documents into validated knowledge graphs with Docling Graph

Convert documents into schema-enforced entities and graph relationships when the job is exact knowledge extraction rather than generic document parsing.

Prerequisites

Python 3.10+, docling-graph package, source documents supported by Docling, optional model provider credentials for remote LLM extraction

Installation

Use the upstream install or setup path that matches your environment:

• pip install docling-graph
• git clone https://github.com/docling-project/docling-graph
• uv sync --extra dev
• uv run pre-commit run --all-files

Requirements and caveats from upstream:

• 
• Python 3.10 or higher

• Convert document from URL (each line except the last must end with )

Basic usage or getting-started notes:

• ✍🏻 Input formats: Docling’s supported inputs: PDF, images, markdown, Office, HTML, and more.

• 📐 Structured extraction: LLM output is schema-enforced by default; see CLI and API to disable.

• 🐛 Trace capture: Debug exports for extraction and fallback diagnostics.

• Source: https://github.com/docling-project/docling-graph

• Extracted from upstream docs: https://raw.githubusercontent.com/docling-project/docling-graph/HEAD/README.md

Documentation

• https://docling-project.github.io/docling-graph/

Source

• Agent Skill Exchange