Skip to main content

Graph RAG (Nexigraph)

NexiGraph - Knowledge graph generation and RAG system for strucutred data​

NexiGraph is a knowledge graph system, that can be used to store and retrieve structured documents (e.g. JSON). It can automatically create relations between documents, and allow for GraphRAG. The Question & Answer agent is an example of GraphRAG enabled agent.

It has the following components:

  • server: The server that exposes endpoints to submit strucutred document. It exposes the API through REST and MCP.
  • agent_fkey: An agent that analyses foreign key relationships between the documents and automatically creates relations.
  • agen_qa: An agent that can answer questions based on the knowledge graph. It exposes the chat interface through A2A.
  • client: Client uses the server API to submit documents. Examples of clients can be found in the nexigraph/clients directory.

✨ Features​

  • 📂 Ready-made clients to import data from AWS, Backstage, K8s.
  • 🔗 Auto creation of relations
  • 💬 A2A enabled query agent to query the graph database

🚀 How to run​

Pre-requisites:

  • docker
  • docker-compose
  • make
  • uv

Run the following command to start the nexigraph service locally:

export AZURE_OPENAI_ENDPOINT=<azure-openai-endpoint> OPENAI_API_VERSION=<azure-openai-version> AZURE_OPENAI_API_KEY=<azure-openai-token>
docker-compose up

This will start the components mentioned above. The different components will come up at different times and retry until they connect with each other. A dummy_client will populate dummy data into the graph database

The following should be accessible:

How to query:

  • Once the relations are created you can ask question on the graph database by:
  • docker run -it -e A2A_HOST=host.docker.internal -e A2A_PORT=8096 ghcr.io/cnoe-io/agent-chat-cli:latest