README.md

RAG agent with Neo4j

This is a repo to showcase how to quickly setup a RAG agent with Neo4j graph database. It include vector search and graph RAG using Neo4j's Cypher queries.

I just followed a tutorial to create a RAG agent working with a neo4j DB.

The added value of this repo is:

• I updated the code (imports/API usage were outdated)
• They created an agent with the deprecated langchain's agent object, whereas I created the agent with Huggingface's smolagents.
• Its simplicity.

Reference: Neo4j tutorial

Steps

0. install everything for desktop
1. create db
2. run the connection script to ensure that it works
3. try the search script

1. Install neo4j desktop and create a db

1. download and install free neo4j desktop for windows
2. create a new dbms and create a DB with the test data
   • test data: https://gist.githubusercontent.com/tomasonjo/08dc8ba0e19d592c4c3cde40dd6abcc3/raw/da8882249af3e819a80debf3160ebbb3513ee962/microservices.json
3. select it and go to plugins on the right panel and install APOC extension

2. Create the .env file

Template:

OPENAI_API_KEY=
NEO4J_DB_URL="bolt://localhost:PORT"
NEO4J_DB_USERNAME=
NEO4J_DB_PWD=
HUGGINGFACEHUB_API_TOKEN=
MISTRAL_API_KEY=

Run the _1_connect.py script to ensure connection to db works.

About the models:

• I use mistral AI API to perform inference (it can easily be embedded as a langchain inference object)
• I use HF API to run the agent
• I use sentence-transformers' "all-MiniLM-L6-v2" model to perform text embedding locally

3. Create a new Python env

4. Try the different arguments with _2_vector_search.py

usage: _2_vector_search.py [-h] option

A script to interact with a Neo4j graph database using various methods like
vector search, QA chain, Cypher QA chain, and tool agent.

positional arguments:
  option      Choose one of the following options: ['vector', 'qa',
              'cypher_qa', 'code_agent']

uv run _2_vector_search.py vector

uv run _2_vector_search.py qa

uv run _2_vector_search.py cypher_qa

uv run _2_vector_search.py code_agent