Skip to content

Latest commit

 

History

History
126 lines (78 loc) · 3.85 KB

README.md

File metadata and controls

126 lines (78 loc) · 3.85 KB

Update 14th November 2024: Check the branch feature/optimization which runs 2.5x faster than this branch's code (what you saw in the demo video). The optimized branch achieves 2.5x faster performance (10-20 seconds range) through two key improvements:

  1. Optimized Token Management - Reduced HYDE max tokens to 400 and HYDE-v2 to 768. Keeps hallucinated output concise while maintaining relevance. Result: ~5-10 seconds saved

  2. Enhanced Context Processing - Implemented SambaNova Llama 3.1 8b (1000 tok/s) for context filtering and changed gpt4o -> SambaNova Llama 3.1-70b (400 tok/s) for chat responses. Improved context relevance and processing speed. Result: ~20 seconds saved

Worst case scenario codeQA works under 20 seconds as compared to the previous 40 seconds.

What is CodeQA?

A powerful code search and query system that lets you explore codebases using natural language. Ask questions about your code and get contextual answers powered by LanceDB, OpenAI gpt4o-mini/gpt4o and Answerdotai's colbert-small-v1 reranker. Supports Python, Rust, JavaScript and Java with a clean, minimal UI.

Blog Links:

An attempt to build cursor's @codebase feature - RAG on codebases - part 1

An attempt to build cursor's @codebase feature - RAG on codebases - part 2

CodeQA helps you understand codebases by:

  • Extracting code structure and metadata using tree-sitter AST parsing
  • Indexing the code chunks using OpenAI/Jina embeddings and storing them in LanceDB
  • Enabling natural language searches across the codebase by using @codebase in queries
  • Providing context-aware answers with references
  • Supporting interactive chat-based code exploration

Prerequisites

  • Python 3.6 or higher
  • Redis server running on localhost:6379

Setup

  1. Clone the repository:

    git clone https://github.com/sankalp1999/code_qa.git

  2. Navigate to the project directory:

    cd code_qa

  3. Set up a Python virtual environment:

Treesitter is supported >=3.8 to 3.11

python3.11 -m venv venv
source venv/bin/activate

  1. Install the required packages:

    pip install -r requirements.txt

  2. Run the redis server

redis-server

Configuration

You only need to set the OpenAI API key. Jina API key is optional, if you want to use Jina embeddings instead of OpenAI.

Create a .env file and add the following:

OPENAI_API_KEY="your-openai-api-key"
JINA_API_KEY="your-jina-api-key"

Building the Codebase Index

To build the index for the codebase, run the following script:

chmod +x index_codebase.sh

./index_codebase.sh <absolute_path_to_codebase>

This will parse the codebase to get the code chunks, generate embeddings, references and store them in LanceDB.

Usage

To start the server

python app.py <folder_path>

For example, to analyze a JavaScript project located in /Users/sankalp/Documents/code2prompt/twitter-circle, run:

python app.py /Users/sankalp/Documents/code2prompt/twitter-circle

Once the server is running, open a web browser and navigate to http://localhost:5001 to access the code search and query interface.

Use @codebase keyword in queries to fetch context via embeddings Enable re-ranking option to get more relevant results

Technologies Used

  • Flask: server and UI
  • Treesitter: parsing methods, classes, constructor declarations in a language agnostic way using the abstract syntax tree
  • LanceDB: vector db for storing and searching code embeddings
  • Redis: in-memory data store for caching and session management
  • OpenAI, Jina for chat functionalities and colbert-small-v1 for reranker

License

This project is licensed under the MIT License.