A GraphRAG Agent - Know-Your-Customer

A basic GraphRAG agent built with OpenAI Agent SDK, Neo4j and Neo4j MCP server

Before You Start

Before running the KYC Agent, ensure you have the following prerequisites installed and configured:

1. Python 3.13+

   • Download and install Python 3.13
   • Verify installation with:

   python --version

2. uv Package Manager

   • Install uv
   • Verify installation with:

   uv --version

3. Ollama

   • Ideally, you need a device with 6GB GPU Memory (the model weights take about 4.1GB)

   • Install Ollama

   • Start the Ollama server:

     ollama serve

   • Pull the Text-to-Cypher model:

     ollama pull ed-neo4j/t2c-gemma3-4b-it-q8_0-35k

   • Verify the model is available and being served locally

     ollama ls

   You should see pull ed-neo4j/t2c-gemma3-4b-it-q8_0-35k listed (available locally)

4. OpenAI Key You will be using the OpenAI Agent SDK with an OpenAI model, so you need an OPENAI_KEY

Neo4j

You have two options to create a Free Neo4j database

Option 1: Local Neo4j Docker Instance

Start a Neo4j docker container using the provided docker-compose.yml:

docker compose up -d

Option 2: Neo4j AuraDB Free (Managed Instance)

1. Head over to Neo4j AuraDB Console

2. Create a new database instance, Choose AuraDB Free. Make sure to download your credentials.

Prepare to Run the Agent

1. Clone the repository:

   git clone https://github.com/neo4j-product-examples/graphrag-kyc-agent.git
   cd graphrag-kyc-agent

2. Create a virtual environment and install dependencies:

   uv venv
   source .venv/bin/activate
   uv sync

3. Set up environment variables:

   Create a .env file with your Neo4j and OpenAI credentials:

   OPENAI_API_KEY=sk-... - Your OpenAI key
   

   If you are using an AuraDB Free tier instance, find the downloaded credentials. Look at the NEO4J_URI and locate your instance id.

   For example, If your instance id is NEO4J_URI=neo4j+s://4469a679.databases.neo4j.io

   Your instance id is 4469a679.

   Add the following to your .env file:

   NEO4J_URI=neo4j+s://<YOUR_INSTANCE_ID>.databases.neo4j.io
   NEO4J_USERNAME=<YOUR_INSTANCE_ID>
   NEO4J_PASSWORD=<YOUR_PASSWORD>
   NEO4J_DATABASE=<YOUR_INSTANCE_ID>
   

Load the dataset

python generate_kyc_dataset.py

Run the Agent

Start the agent:

python kyc_agent.py

Run the following sequence of questions

Test the agent tools with these example questions:

1. Get the database schema:

   What is the schema of the database?
   

   This test the MCP Neo4j server get schema to fetch the schema.

2. Find suspicious rings:

   Show me 3 watchlisted customers involved in rings
   

   This tests the find_customer_rings tool.

3. Check shared addresses:

   For each of these customers, find their addresses and find out if they are shared with other customers
   

   This tests both the Text-to-Cypher Generation tool and the MCP Neo4j server exec read query tool.

4. Recent transactions:

   Find more details about the customer with the shared adddress. List accounts and recent transactions
   

   This tests the get_customer_and_accounts tool.

5. Store conversation summary:

   Write a 300-word summary of this investigation into this customer. Store it as a memory, make sure to link it to accounts and transasction mentioned in the conversation
   

   This tests the create_memory tool.