Neo4j LLM Graph Builder converts unstructured data into a Neo4j knowledge graph using LLMs and LangChain, with multiple models and embedding options. It supports local, cloud, and separate backend/frontend deployment, plus token usage tracking and schema-driven graph creation.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Neo4j Labs/llm-graph-builder is a tooling project that constructs knowledge graphs in Neo4j from unstructured data (PDFs, DOCs, TXTs, YouTube videos, web pages, etc.) using Large Language Models and the LangChain framework. It provides a frontend and a FastAPI backend to upload sources, select LLMs, and generate graphs with nodes, relationships, and properties.
How it works
The application accepts files from various sources (local, GCS, S3, web), uses a chosen LLM to extract graph components, and stores the results in Neo4j. It supports schema configurations, graph visualization via Neo4j Bloom, and a chat interface for querying the data. It includes token usage tracking and an embedding model selection mechanism for vector representations. It supports multiple embedding model providers and can operate with or without per-user usage tracking. Deployment options include local Docker-Compose, backend/frontend separation, and cloud deployment via Google Cloud Run.
Getting started
Prerequisites include Python 3.12+ and Neo4j 5.23+ with APOC. Backend setup involves creating a .env in backend, configuring NEO4J_URI, NEO4J_USERNAME, NEO4J_PASSWORD, NEO4J_DATABASE, and running the following commands:
cd backend
python3.12 -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn score:app --reload
The README also documents frontend/backend separate runs and cloud deployment steps, along with source access and model configuration details.
Recent releases
- v0.8.7 (2026-07-23): Features include updated support for latest models across OpenAI, Gemini, Anthropic Claude, Fireworks DeepSeek, and others; added a link to Neo4j Document Intelligence.
- v0.8.6 (2026-06-11): Updated packages/dependencies; resolved Wikipedia extraction error.
- v0.8.5 (2026-02-11): Release focuses on embedding model support and token usage tracking; embedding model options described.
- v0.8.4 (2026-01-15): Dependency upgrades, model updates, cleaner architecture, and usage tracking improvements.
- v0.8.3 (2025-06-24): Added Data Importer support to import graph models into LLM Graph Builder; Claude 4 Sonnet support.
Traction
Stars: 4981 Forks: 849 Open issues: 53
Behind the repo
Linked data mentions Neo4j, LangChain, and support for Neo4j Aura databases; deployment examples include Google Cloud Platform and docker-compose workflows. The repository targets integration with Neo4j databases and various LLM providers, with environment variable configurations for backend and frontend.
Caveats
License: Apache-2.0 Created: 2024-01-11 Last push: 2026-08-03 Languages: Jupyter Notebook (used as primary language indicator) Prerequisites specify Neo4j 5.23+ with APOC and Python 3.12+. The README emphasizes backend/frontend separation for development and deployment, and requires explicit configuration of environment variables for authentication, embedding models, and LLM providers.