graphrag: Build Graph-Powered AI Search and Chat

Overview

TigerGraph GraphRAG is a Python application for building knowledge graphs from documents and using them to answer natural-language questions. It combines TigerGraph graph and vector capabilities with an LLM to retrieve relevant context and produce responses, and also supports questions against existing graph data through curated database queries.

It is aimed at teams that want an integrated chat interface and APIs for document ingestion, graph-backed retrieval, and question answering. It makes most sense when TigerGraph is an acceptable database dependency and the team is prepared to configure and operate its own LLM services.

Key Features

• Builds a knowledge graph from documents, with configurable chunking, entity extraction, and community detection.
• Combines vector search and graph traversal for document retrieval.
• Supports natural-language questions answered through existing TigerGraph queries and graph data.
• Offers Classic retrieval and Agentic chat modes, including Planned and Reactive agent styles.
• Ingests local files and cloud-stored documents, with an optional Amazon Bedrock Data Automation workflow.
• Provides a web chat interface, backend APIs, and an admin UI for graph and ingestion workflows.
• Connects to external MCP tools in the agentic engine.
• Supports multiple LLM providers, including OpenAI, Azure, Google, AWS Bedrock, Ollama, Hugging Face, and Groq.

Use Cases

• Knowledge teams can index internal documents and answer questions with graph-linked context instead of relying only on general model knowledge.
• TigerGraph developers can let users ask natural-language questions about data already stored in their graphs.
• Platform administrators can deploy a document Q&A service with Docker Compose or Kubernetes and manage ingestion through its UI.
• Teams exploring agentic retrieval can compare planned and reactive approaches and connect additional tools through MCP.

Project Facts

• Language: Python
• License: AGPL-3.0
• Stars: 23
• Forks: 23
• Topics: none listed
• Archived: no

Getting Started

For a quick Docker-based setup, set an LLM API key and run:

curl -k https://raw.githubusercontent.com/tigergraph/graphrag/refs/heads/main/docs/tutorials/setup_graphrag.sh | bash

This deploys GraphRAG with TigerGraph Community Edition. See the README for prerequisites, setup with an existing TigerGraph instance, Kubernetes deployment, and configuration details.

Alternatives

• graphrag: Microsoft GraphRAG builds graph-based context from text with its own indexing pipeline, rather than relying on TigerGraph for graph storage and queries.
• KAG: KAG combines knowledge graphs, source text, and LLMs for domain-specific and multi-hop QA, rather than centering on TigerGraph-backed graph and vector retrieval.
• RAG-Anything: RAG-Anything focuses on parsing and querying multimodal documents through LightRAG, rather than using TigerGraph as its graph backend.
• kotaemon: kotaemon is a customizable document-QA app and toolkit built around RAG pipelines, without graphrag's TigerGraph-centered knowledge graph.

Considerations

• TigerGraph is the only supported graph and vector database. The README identifies hybrid search as the officially supported retrieval method; other retrieval methods and the agentic engine are provided as-is for self-service use.
• The deployment requires TigerGraph 4.2 or later, Docker Compose or Kubernetes, and credentials for an LLM provider. LLM usage may incur costs, especially during graph construction and embedding generation.
• The project is provided as-is, and official support is limited to work delivered through a Statement of Work. Customizations are the user's responsibility.
• The AGPL-3.0 license should be reviewed against the intended deployment and distribution model.