CodeRAG (NVIDIA Powered)
Repository-Aware Intelligence for GitHub Copilot
CodeRAG integrates repository-level context into the GitHub Copilot experience. By indexing source code into a high-dimensional vector space using NVIDIA's advanced embedding models, it enables Copilot to provide responses grounded in the specific implementation patterns and architectural decisions of your repository.
Install Extension · View Backend · Quick Start · Technical Overview
Core Capabilities
| Capability |
Description |
| Semantic Code Search |
Provides contextually relevant code snippets from the repository using vector similarity. |
| #coderag Context Injection |
Supports the #codeRagSearch reference to automatically inject retrieved context into Copilot Chat. |
| @coderag Chat Participant |
Dedicated chat participant for repository-specific inquiries and architectural analysis. |
| NVIDIA Embedding Engine |
Utilizes nv-embedcode-7b-v1, a 7-billion parameter model optimized for code understanding. |
| FAISS Vector Index |
High-performance nearest-neighbor retrieval powered by Facebook AI Similarity Search. |
| Status Bar Integration |
Real-time indexing and search status directly within the VS Code status bar. |
Technical Architecture
The system consists of a TypeScript-based VS Code extension and a Python-based FastAPI backend.
- Indexing Path: The extension triggers a recursive scan of the local workspace. Files are chunked and transmitted to the backend, where the NVIDIA model generates embeddings. These are stored in a FAISS index.
- Retrieval Path: When a user invokes
#coderag or @coderag, the query is embedded and matched against the FAISS index. The most relevant code blocks are returned as grounded context to the Copilot Language Model.
Requirements
- CodeRAG Backend: A running instance of the FastAPI backend (local or hosted on Azure).
- GitHub Copilot Chat: Active subscription and extension installed in VS Code.
- VS Code v1.95.0+: Required for the latest Language Model Tool APIs.
Installation
Marketplace
Search for CodeRAG in the VS Code Extensions tab and select Install.
Backend Setup
Follow the deployment guide in the CodeRAG Backend repository to configure the NVIDIA API and FAISS environment.
Quick Start Guide
Ensure the BACKEND_URL in the extension settings (or source) points to your active server instance (e.g., your Azure Web App URL).
2. Index Workspace
Open the Command Palette (Ctrl+Shift+P) and execute:
CodeRAG: Index Workspace
3. Invoke Context
Use the following syntax in the Copilot Chat sidebar:
How is the routing handled in this project? #codeRagSearch@coderag Explain the interaction between the indexer and retriever.
Security and Privacy
CodeRAG processes your code through your own infrastructure. Communication occurs between the VS Code client and your designated backend server. No source code is transmitted to external third-party services during the retrieval process.
Developed by
Shyam Sunder
B.Tech Computer Science & Engineering (AI & ML)
Alliance University, Bangalore
"Grounding AI in the reality of your code."
