Capabilities
No subscriptions. No data harvesting. Just intelligence, locally.
Wake-word detection and push-to-talk. Luna listens, understands, and speaks back with local TTS.
Persistent facts, personality state, and conversation summaries stored in local SQLite and ChromaDB.
Record a meeting and get a timestamped transcript plus minutes: decisions, action items and open questions.
Launch apps, control the browser, manage Spotify, execute calendar tasks and desktop actions.
Real headlines with images by category, and one product compared across Walmart, Costco, Target, Sam's Club and Safeway.
Runs on your machine with Ollama or the cloud model you pick. Reach it from your phone over Tailscale, without opening it to the internet.
See it
Real screens from Luna: meeting minutes, price comparison, your files, and the phone app.




Any model
Run open models on your own hardware, or plug in your own API key. Switch any time in Settings.
Workflow
Four stages, entirely on your machine.
Chat through the UI or trigger Luna by voice. Accepts text, voice and attached files.
Memory facts, personality, calendar state, and activity history are injected into the prompt.
Your local Ollama model or any OpenAI-compatible API handles inference with a full context window.
Luna streams the reply, executes tools, opens widgets, updates memory, and delivers voice output.
Download
Free downloads from the latest release. Run the server once, then use it anywhere.
New to self-hosting? Follow the step-by-step setup guide.
FAQ
Luna is a free, open-source AI assistant and agent engine that you run yourself. It chats, listens, remembers, writes meeting minutes, works with your files and connects to apps like GitHub, Google Workspace and Microsoft 365. Your conversations and files stay on your own hardware.
Yes. Luna is MIT-licensed and free to use, modify and self-host. Local models through Ollama cost nothing to run. If you choose a cloud model such as Claude or Gemini, that provider bills you for your own API key.
Yes. With a local model through Ollama, chat, memory, voice transcription and meeting minutes run entirely on your computer. Features that look things up online, such as web search, news and store prices, need an internet connection.
Local models through Ollama, LM Studio, llama.cpp or any OpenAI-compatible server, and cloud models from Anthropic (Claude), Google (Gemini), OpenAI, Groq, Mistral, Cohere, NVIDIA NIM and OpenRouter. You can switch models at any time in Settings.
Yes. Open your Luna in any phone browser, or install the Android app, and connect to the computer running it privately over Tailscale. Nothing is exposed to the public internet.
Luna stores chats, memory and files in a database on your own machine or server, and nothing is sent to the Luna project. If you pick a cloud model, your messages go only to the provider you chose.
You host Luna yourself, choose any model including fully local ones, and keep your data. It is open source, so you can add skills, connect tools over MCP and change how it works.
A Windows, macOS or Linux computer. The desktop app needs Python 3.10 or newer for its built-in server, or you can run the server with Docker. For local models, install Ollama; a GPU speeds things up, but small models also run on a CPU.
Free, open-source, and always will be. Inspect every line, fork it, or contribute.