SarvX Code AI Assistant

A powerful AI-powered code assistant that uses Retrieval Augmented Generation (RAG) with local LLM processing for privacy and enhanced context-awareness.

Features

• Local Code Understanding: Uses Ollama with the deepseek-r1:8b model for code explanations
• Context-Aware Responses: Leverages RAG with CodeT5 embeddings and Pinecone vector database
• Codebase Indexing: Intelligent chunking and embedding of your codebase
• VS Code Extension: Seamlessly integrated into your development workflow
• CLI Application: Also available as a command-line tool

Components

This project consists of two main components:

1. VS Code Extension: Provides code explanation directly in your editor
2. CLI Application: Command-line interface for code explanations

Setup and Installation

Prerequisites

• Node.js (v14 or higher)
• Ollama with the deepseek-r1:8b model installed
• Pinecone account with an index created (dimension: 768)

VS Code Extension

1. Install the extension from the VSIX file:

   code --install-extension fix-extension/sarvx-code-ai-assistant-0.1.0.vsix
   

2. The extension provides the following commands:

   • Code AI Assistant: Test Connections - Test connections to Ollama and Pinecone
   • Code AI Assistant: Explain Selected Code (Simple) - Explain the selected code

CLI Application

1. Run the CLI application:

   node cli-app-fixed.js explain "your code here"
   

Indexing Your Codebase

Before using the assistant, you need to index your codebase:

1. Configure your Pinecone API key in index-repo.cjs
2. Run the indexing script:

   node index-repo.cjs /path/to/your/codebase
   

Implementation Details

• Embeddings: Uses CodeT5 model to create embeddings of code chunks
• Vector Database: Pinecone stores code embeddings for retrieval
• Local LLM: Ollama runs the deepseek-r1:8b model locally for privacy
• Retrieval: Similar code context is retrieved to enhance explanations

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Pinecone for vector database
• Ollama for local LLM hosting
• CodeT5 for code embeddings