What's new in Kiosk Satellite 2026.8.59
The Voice Satellite settings page can now change the integration's settings instead of just reading them, and a new Last interaction sensor lets automations tell an idle kiosk from one in use.
New Feature: Voice Satellite settings are editable from the kiosk
- The Voice Satellite settings on the device and in the remote admin now writes settings, not just reads them. A general group carries the assigned satellite (a picker that rebinds the kiosk and reloads the dashboard, with a Disabled choice that clears the binding), Auto start, Mute, both Assist pipelines and Finished speaking detection; a Wake Word card carries the wake word engine, both wake words, the sensitivity, the noise gate and Stop word interruption; an Appearance card carries the skin, theme mode, the reactive activity bar and its update rate (now in frames per second), and the text scale.
- Changes land exactly where the Voice Satellite panel would put them: the Home Assistant half writes through the satellite's own entities, and the browser half persists to the panel profile and applies live, so a skin change repaints without a reload and survives one. The browser-side rows need a Voice Satellite version that ships the new hook; the entity rows work with any version.
- An Engine row starts or stops the voice assistant in the dashboard page, through the same path as the Voice Satellite panel's own Start and Stop buttons, with the running state shown next to it, and the General card reports the installed Voice Satellite integration version.
New Feature: A Last interaction sensor (#241)
- A Last interaction timestamp sensor on both the ESPHome and MQTT integrations, stamped when someone touches the screen or speaks to the device, so automations pick their own idle threshold instead of the app guessing one, for example only showing a camera view after five quiet minutes:
now() - states('sensor.<device>_last_interaction') > timedelta(minutes=5). - Motion, announcements, ringing timers and media playback deliberately do not count, so someone walking past or a TTS message never reads as the kiosk being used, and a continuous stream of touches collapses to at most one publish a minute with the final touch always landing accurately.