blar-graph 0.0.6

Llm agent to search within a graph

code-base-agent

Introduction

This repo introduces a method to represent a local code repository as a graph structure. The objective is to allow an LLM to traverse this graph to understand the code logic and flow. Providing the LLM with the power to debug, refactor, and optimize queries. However, several tasks are yet unexplored

Technology Stack

We used a combination of llama-index, CodeHierarchy module, and tree-sitter-languages for parsing code into a graph structure, Neo4j for storing and querying the graph data, and langchain to create the agents.

Installation

Install the package:

pip install blar-graph

Set the env variables

NEO4J_URI=neo4j+s://YOUR_NEO4J.databases.neo4j.io
NEO4J_USERNAME=neo4j
NEO4J_PASSWORD=YOUR_NEO4J_PASSWORD
OPENAI_API_KEY=YOUR_OPEN_AI_KEY

If you are new to Neo4j you can deploy a free instance of neo4j with Aura. Also you can host your own version in AWS or GCP

Quick start guide

To build the graph, you have to instantiate the graph manager and constructor. The graph manager handles the connection with Neo4j, and the graph constructor processes the directory input to create the graph.

from blar_graph.graph_construction.graph_builder import GraphConstructor
from blar_graph.graph_construction.neo4j_manager import Neo4jManager

graph_manager = Neo4jManager()
graph_constructor = GraphConstructor(graph_manager)
graph_constructor.build_graph("YOUR_LOCAL_DIRECTORY", "python")
graph_manager.close()

Note: The supported language for now is python, we are going to include Typescript (or other language) if you ask for it enough. So don't hesitate to reach out through the issues or directly to [e-mail] or [e-mail]