MiniRAG - Minimalist, Self-Hosted, Single-Binary RAG

Killer Feature:
Start chatting with your data in seconds:

minirag -docs=mydata

That's it. No setup, no config, just point to your folder and go!

MiniRAG is a single-binary, self-hosted chat webapp with Retrieval-Augmented Generation (RAG).
It runs as a single static binary on your machine—no cloud, no dependencies, no external services.

• Self-hosted: All data and LLM inference stay on your device.
• Single binary: Download, run, and you're ready—no Docker, no Python, no Node.js required.
• Private & secure: Your documents and chat never leave your computer.

MiniRAG is a lightweight, production-ready implementation of Retrieval-Augmented Generation (RAG) that combines the power of large language models with efficient document retrieval. Built with Go and React, it provides a modern, responsive interface for interacting with your documents.

🍺 Install MiniRAG via Homebrew

brew tap komly/minirag
brew install minirag

Then run:

minirag -docs=mydata

• By default, MiniRAG will store its data in ~/.minirag.
• You can specify a different data directory with the -data flag if needed.
• The server will listen on :7492 by default. You can change this with the -http flag.

---

🚀 Features

• Document Indexing: Automatically indexes text documents with vector embeddings
• Smart Search: Uses semantic search to find relevant document chunks
• Chat Interface: Modern React-based chat UI with source attribution
• Ollama Integration: Works with any Ollama-compatible model
• Production Ready: Single binary deployment with embedded frontend

📋 Prerequisites

• Go 1.21 or later
• Node.js 18+ and pnpm
• Ollama running locally
• Text documents to index

🛡️ 100% Local LLM & Data Safety

minirag runs entirely locally on your machine:

• All LLM inference and document indexing happen on your computer.
• No data ever leaves your device.
• Your documents and chat history are never sent to any external server or cloud.

Your data stays private and secure.

🚀 Usage

Development Mode

1. Start the frontend development server:

   cd backend/cmd/minirag/frontend
   pnpm dev

2. In another terminal, start the backend:

   cd backend
   go run cmd/minirag/main.go --dev

Production Mode

1. Build the frontend:

   cd backend/cmd/minirag/frontend
   pnpm build

2. Run the server:

   cd backend
   go run cmd/minirag/main.go

Command Line Options

• --docs: Directory containing documents to index (default: "./docs")
• --data: Directory for storing index and metadata (default: "~/.minirag")
• --ollama-url: Ollama API URL (default: "http://127.0.0.1:11434")
• --ollama-model: Chat model name (default: "gemma3:12b")
• --ollama-embed-model: Embedding model name (default: "nomic-embed-text:latest")
• --http: HTTP listen address (default: ":7492", e.g. ":7492" or "0.0.0.0:7492")
• --dev: Run in development mode
• --force-reindex: Force reindexing of all documents

🔍 API Endpoints

• POST /chat: Chat with the system

  {
    "query": "your question",
    "temperature": 0.7,
    "max_tokens": 1000
  }

• POST /query: Search documents

  {
    "query": "search term"
  }

• GET /debug/db: View database state

🏗️ Architecture

Frontend

• React Vite TypeScript SPA
• Modern UI with responsive design
• Real-time chat interface
• Source attribution display

Backend

• Go application serving the frontend
• Vector database using chromem-go
• Integration with Ollama API
• Document processing and indexing

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📝 License

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

🙏 Acknowledgments

• chromem-go for vector database
• Ollama for LLM integration
• React and Vite for frontend

⚙️ Command-line Flags

The minirag binary supports the following flags:

Flag	Default	Description
-docs	docs/	Path to the directory with documents to index and search.
-data	~/.minirag	Path to the directory for storing vector DB and metadata.
-http	:7492	HTTP listen address (e.g. ':7492' or '0.0.0.0:7492').
-dev	(off)	Enable development mode (useful for hot-reload and debugging).
-force-reindex	(off)	Force reindexing of all documents on startup, ignoring previous metadata.

Example usage

minirag -docs=app/ -data=~/.minirag -http=:8080

• -docs=app/ — use the app/ directory for your documents.
• -data=~/.minirag — store all vector DB and metadata in the ~/.minirag directory.
• -http=:8080 — run the web server on port 8080.

Tip: You can combine flags as needed. If you omit a flag, the default value will be used.

---