CodeSearch

Introduction

CodeSearch is natural language query tool for codebases. It uses the OpenAI API for getting the embedding vectors and indexes them using the FAISS library. When you query the system, it matches the query vector to the top 5 results and display in a web-based UI.

When you enter a project path, the tool searches for all the code files inside the project (including subdirectories), generates its Abstract Syntax Tree (using libCST library), indexes each of the function, along with its filepath, class name, function name, line number and embedding vector. It then stores all of this in a local PostgreSQL database.

• Fast Setup: We have Docker support to quickly setup a working environment
• Interactive UI: We have created a interactive web-based UI for searching
• Language Support: Currently we support Python only. Adding new languages shortly
• GPU and CPU Support: The tool has support for GPU as well as CPU

Setup

• Get your OpenAI API key from link
• Rename the /backend/codesearch/.env.sample to /backend/codesearch/.env and add your API key in it

CPU

• Rename /backend/Dockerfile_cpu to /backend/Dockerfile and ./docker-compose_cpu.yml to ./docker-compose.yml
• Run the Docker container:

$ docker compose up > logs.txt

GPU (NVIDIA)

• Install the NVIDIA container toolkit from their official website link
• Use the default /backend/Dockerfile and ./docker-compose.yml files
• Run the Docker container:

$ docker compose up > logs.txt

Usage

• Once you run the docker compose up command. The application should be up and running on http://localhost:4000/
• Open http://localhost:4000/ on the browser
• Click on Index Project button and enter the absolute path to your project to be indexed
• It should take about 30 secs to index the project. You can see the progress on logs.txt which is created
• Once completed, enter your query in the search box and it will return the top 5 results for your search