I shipped a full-screen voice mode that shuts the mic off between turns

Most voice assistants leave the microphone open. I did not want that on a desktop app that already sits on someone's files.

MindBridge is a local AI assistant for Mac and Windows. The model runs on the computer through Ollama. Web search stays off until you turn it on.

Presence is the new mode. Option+Shift+P on a Mac, Alt+Shift+P on Windows, opens a dark screen and a particle sphere. It greets you by the name saved in Settings. You talk. It answers out loud. There is no chat transcript on that screen.

A few rules I kept:

• The microphone is on only while it is listening. While it thinks or speaks, the mic is off.

• Saying "stop" does not cut the voice. A Stop button does, then it listens again.

• Close the window and that session is gone.

• Audio stays on the machine. The words you said do not appear in the interface.

• From this screen it can talk, use your local knowledge base, and search the web only if cloud features are allowed. It does not touch files and it does not generate images. Those stay in the normal chat, where a delete still waits for a yes.

Date, place, and weather come from the computer, including a city you name out loud.

It is in a closed beta. Each spoken turn counts as one chat. On the free plan that is 10 a day, and Presence says so when you hit the limit.

The part I am unsure about: a voice screen with no text feels calm, and you also cannot scroll back. If you run local models, would you keep the transcript hidden, or want a way to pull it up after?