RagStack

Semantic code search powered by your own AI. Index any project, search with natural language. Runs on your infrastructure — your code never leaves your network.

Features

• Semantic Search — Ask questions in plain English, find relevant code across your entire codebase
• One-Click Indexing — Click "Index" to send your project to your RAG server. Files are chunked, embedded, and stored in pgvector
• Server Selector — Switch between Local, LAN, Cloud, or Custom server with one click from the sidebar
• Click to Navigate — Search results link directly to the matching code in your editor
• Right-Click Search — Select any text, right-click, "RagStack: Search Selected Text"
• Multi-Project — Index multiple projects under different names, search across all of them
• Remote Friendly — Works over LAN or internet. Index from any machine, query from anywhere
• Private & Self-Hosted — Your code stays on your server. Uses Ollama for embeddings (no OpenAI API needed)

Quick Start

1. Install the extension
2. Click the RagStack icon in the sidebar (magnifying glass)
3. Click the server button at the top to select your AI server
4. Click Index to index your current workspace (first time, takes a few minutes)
5. Type a question and press Enter — results appear instantly

Codeva — Terminal Mode (like Claude Code)

Just type codeva in any VS Code terminal:

codeva

Or via command palette: Ctrl+Shift+P → "Codeva: AI Terminal"

This opens an interactive AI coding assistant — ask questions, search code, index projects, get AI-powered answers with codebase context. No separate install needed.

  ╔══════════════════════════════════════╗
  ║  Codeva — AI Coding Assistant       ║
  ║  powered by RagStack                ║
  ╚══════════════════════════════════════╝

  codeva> how does the wallet page fetch real balance?
  Searching codebase...
  Found 6 relevant chunks
  Thinking...

  The wallet page fetches the real balance via...

  ── Sources ──
  72%  src/pages/Wallet.tsx
  68%  src/api/agent.ts

Terminal commands:

Server Configuration

Click the server button in the sidebar to choose your AI stack:

Network Examples

Same machine (development):
  ragstack.serverUrl = http://localhost:30810

Office LAN (shared AI server):
  ragstack.serverUrl = http://192.168.1.109:30810

Remote / VPN / Cloud (works from anywhere):
  ragstack.serverUrl = https://ai.mycompany.com/rag

Kubernetes with Traefik:
  ragstack.serverUrl = https://ai.mycompany.com/rag
  (Traefik strips /rag prefix and routes to the RAG server NodePort)

Requirements

RagStack requires a running RAG server with:

• Ollama — Local LLM runtime with nomic-embed-text model
• pgvector — PostgreSQL with vector extension
• RAG Server — FastAPI server (included in the RagStack repo)

Quick Setup (Docker)

# Pull the embedding model
ollama pull nomic-embed-text

# Start pgvector
docker run -d --name pgvector -p 5432:5432 \
  -e POSTGRES_PASSWORD=password \
  pgvector/pgvector:pg17

# Start RAG server
docker run -d --name rag-server -p 30810:8100 \
  -e DB_HOST=host.docker.internal \
  -e DB_PASSWORD=password \
  -e OLLAMA_HOST=http://host.docker.internal:11434 \
  your-rag-server-image

Kubernetes Setup

Deploy Ollama, pgvector, and the RAG server as K8s services. Expose the RAG server via NodePort (30810) or through an Ingress/Traefik route. See the RagStack repo for full K8s manifests.

Settings

Commands

How It Works

Your Code (VS Code)              Your AI Server
┌──────────────┐     HTTPS/HTTP     ┌─────────────────┐
│  Index:      │ ──────────────────>│  RAG Server     │
│  file chunks │   POST /index      │  (FastAPI)      │
│              │                    │       │         │
│  Search:     │ ──────────────────>│       v         │
│  "how does   │  POST /retrieve    │  Ollama         │
│   auth work" │                    │  nomic-embed    │
│              │ <──────────────────│       │         │
│  Results:    │   JSON response    │       v         │
│  score + code│                    │  pgvector       │
└──────────────┘                    │  (cosine search)│
                                    └─────────────────┘

1. Index: Each file is chunked (~800 chars), embedded into 768-dim vectors via Ollama's nomic-embed-text, and stored in pgvector. Unchanged chunks are skipped automatically on re-index.
2. Search: Your query is embedded the same way, then pgvector finds the most similar code chunks using cosine similarity.
3. Navigate: Click a result header to open the file and jump to the matching code in your editor.

Supported Languages

TypeScript, JavaScript, Python, Go, Rust, Java, C#, Ruby, PHP, Shell, SQL, SCSS, CSS, HTML, YAML, JSON, TOML, Markdown — and any other text-based file via custom ragstack.indexExtensions.

Support

• Website: multitrader.ai
• Email: osvaldo.martini@gmail.com
• Issues: GitHub Issues

License

MIT