We don't collect your data. There is no backend, no accounts, no analytics, no telemetry. Error reports aren't sent either.


Everything stays in your browser. Conversations are kept in IndexedDB, display preferences in localStorage, and the app's files plus any downloaded model weights in Cache Storage. You can export or delete all of it from Settings → Privacy & data. API keys can be encrypted under a passphrase you set. None of this is sent to us.


When you send a message, your browser connects directly to the endpoint you configured: Ollama on your own machine, or OpenAI, Anthropic, OpenRouter, or whatever else you've pointed it at. We're not in that path. What the provider does with what you send is governed by their privacy policy, not ours.

If you want nothing to leave your machine, run a local model: Ollama, LM Studio, llama.cpp, or the in-browser option, which downloads a model once and then runs offline.


In-browser models: weights are downloaded from HuggingFace once and cached. No network needed after that.

Optional features: anything that contacts a third party (the MCP registry, a remote MCP server) tells you who it's contacting when you enable it.

Your provider: as described above.

For more details, see: https://ai.ivx.run/privacy