The house answers in three languages, on local hardware, and never claims more than it has verified. These guides cover the ears, the promises and the screens: building a satellite that hears across a room, scheduling things for later by voice, and putting an episode on any TV with one sentence.
A far-field microphone array that hears you over its own audio, as a native Home Assistant voice satellite — with a device file only forty lines long, because a community packages architecture pinned to an exact upstream commit carries the components and DSP firmware. The build we run in three rooms: hardware, the whole config, flash and adopt, and the lessons that only appear once two satellites share a ceiling.
Home Assistant’s timers only chime. Three small scripts on top of the Scheduler component let the voice assistant schedule a device action at a time or after a delay, in three languages — and speak back the verified time it set, list what is pending, and cancel by voice. Full YAML, plus the rules that keep it honest.
How one spoken sentence becomes verified playback on the right TV: the Jellyfin Sessions API, a Moonfin client per room, and a shell script that never claims what it hasn’t seen — it wakes the screen, brings the app to the foreground, resumes the episode you were part-way through, and only then reports success. Includes the full sanitised scripts and the per-site config, ready to adapt to your own house.