Questions, bug reports, or feature requests? Email yijun.yu@gmail.com and we'll help.
Where do my prompts go? Nowhere, unless you switch on the one feature that changes that. By default all inference runs on your Mac, and the engine that reads your prompts is code-signed without network access, so it cannot send them anywhere. Nothing is uploaded unless you explicitly opt in to sharing anonymous speed measurements or a reviewed list of your own preferences; both are off until you turn them on, and neither ever includes prompts, conversations or files. See the Privacy Policy.
What is the exception? Hybrid Mind, which puts a cloud API of your choosing beside the local engine. A prompt the cloud side answers is sent to that provider, over HTTPS, with an API key you supply — so it does leave your Mac, and what the provider does with it is governed by their policy rather than ours. It is off until you turn it on, the app states plainly that prompts will leave your Mac the first time you do and lets you cancel, and every reply records which side produced it, so you can tell local answers from cloud ones after the fact. The local engine stays network-locked either way; the cloud request is made by the app, not by the engine holding your conversation.
Do I need an internet connection? Only once, to download a model. After that the app works fully offline.
Why is the model download large? Open-source LLM weights are large data files. The app shows the exact size and asks before downloading.
Can I use it with my coding tools? Yes — it exposes an OpenAI-compatible API on localhost you can point Claude Code, OpenCode, or any OpenAI-style client at.