CodeRAG is a VS Code extension that gives you an AI chat assistant fully grounded in your own codebase — with zero data egress. It indexes your code locally using vector embeddings and retrieves the most relevant context before every answer.

Features

Local RAG — indexes your workspace into a local vector database (LanceDB); no code leaves your machine
Multi-provider LLM — works with Ollama (fully local), OpenAI, or Anthropic
Tree-sitter parsing — understands TypeScript, JavaScript, Python, Rust, Go, Java, C, and C++ at the AST level for precise chunk boundaries
Chat panel — sidebar chat with streaming responses and code highlighting (Cmd+Shift+A)
CodeLens — "Ask AI" button above every function and class
Ask about selection — right-click any selected code → Ask CodeRAG About Selection
Auto-reindex — watches for file changes and keeps the index up to date

Requirements

Choose one embedding + LLM backend:

Option A — Ollama (fully local, no API key needed)

Install Ollama
Pull an embedding model: ollama pull nomic-embed-text
Pull a chat model: ollama pull llama3 (or codellama, deepseek-coder, etc.)

Option B — OpenAI

Set coderag.openaiApiKey to your API key
Set coderag.embeddingProvider and coderag.llmProvider to openai

Option C — Anthropic

Set coderag.anthropicApiKey to your API key
Set coderag.llmProvider to anthropic
Embeddings still require Ollama or OpenAI

Quick Start

Open a workspace folder
Run CodeRAG: Index Workspace from the Command Palette (Cmd+Shift+P)
Wait for indexing to complete (status bar shows progress)
Press Cmd+Shift+A or click the CodeRAG icon in the activity bar to open chat

Configuration

Setting	Default	Description
`coderag.embeddingProvider`	`ollama`	`ollama` or `openai`
`coderag.llmProvider`	`ollama`	`ollama`, `openai`, or `anthropic`
`coderag.ollamaUrl`	`http://localhost:11434`	Ollama server URL
`coderag.ollamaEmbeddingModel`	`nomic-embed-text`	Ollama embedding model
`coderag.ollamaLlmModel`	`llama3`	Ollama chat model
`coderag.openaiApiKey`	(empty)	OpenAI API key
`coderag.anthropicApiKey`	(empty)	Anthropic API key
`coderag.anthropicModel`	`claude-sonnet-4-6`	Anthropic chat model
`coderag.topK`	`8`	Chunks retrieved per query
`coderag.maxContextTokens`	`8000`	Max tokens sent to LLM
`coderag.enableCodeLens`	`true`	Show "Ask AI" above functions
`coderag.ignorePatterns`	`node_modules, dist, .git`	Glob patterns excluded from indexing
`coderag.maxFileSizeKB`	`512`	Skip files larger than this

Commands

Command	Description
`CodeRAG: Open Chat`	Open the chat panel
`CodeRAG: Index Workspace`	Re-index the entire workspace
`CodeRAG: Clear Index`	Delete the vector index
`CodeRAG: Show Index Status`	Show file/chunk count
`CodeRAG: Download Tree-sitter Grammars`	Download WASM grammar files
`Ask CodeRAG About Selection`	Ask about selected code (editor context menu)

Privacy

All indexing and search runs entirely on your machine. Only the LLM call leaves your machine — and only when you choose OpenAI or Anthropic. With Ollama everything is 100% local.