Skip to content

Voice and the island

On Linux and macOS the agent listens for "Hey Jarvis", records what you say next and plays Jarvis's answer, as a voice node does. Voice on Windows is in progress.

Listening

Listening is on by default. Turn it on or off with Listen for "Hey Jarvis" on the setup page or with the toggle in the shield icon's menu. Test my microphone on the setup page checks that the agent hears you. On a Mac, listening needs the Microphone permission.

After a reply, Jarvis listens for a few seconds for a follow-up without the wake word (3 seconds by default). A chime plays when the wake word is heard and when something goes wrong.

The wake word's sensitivity, the follow-up window, the chimes and the microphone and speaker are settings (wake_threshold, follow_up, chimes, voice_input, voice_output).

The island

The island is a small pill at the top centre of the screen (under the notch on a Mac). It drops in when the wake word is heard, shows sound bars that follow your voice and then Jarvis's, pulses while Jarvis thinks, and counts down the follow-up window with a ring. It never takes focus or clicks and has no Dock or taskbar entry.

It shows on macOS and on Wayland desktops that support overlays (layer-shell), such as Hyprland and sway. Where the desktop can't show it (GNOME, X11, Windows for now), the agent runs without it. The island setting turns it off.