5 posts
Episode 5: six shutters deciding on their own, from their room temperature and the sun's track across their facade. The model was built from the land registry and thirty days of correlation. Its most useful output was telling me where not to act.
Episode 4: fifteen spoken announcements on a school day, from wake-up to bedtime. The sound played fine from day one, and nobody moved. Eight writing rules, a wake-up rewritten three times in one day, and what happens when the bedroom speaker is offline.
Episode 3: the house now carries the shape of the day. Gate open at 7:55, television cut at 7:30 p.m., shutters closed at the same moment, and an official open dataset that suspends the whole thing during school holidays. Every piece took an evening, and that is what made changing my mind three times possible.
Over a year of home automation run through an agent driving the Home Assistant API: 87 devices, nine ecosystems, 75 automations. What I got out of it is ideas I would never have tried, and diagnostics that cross five systems in one sitting.
A survey of a ten-year home automation build: from Jeedom and hand-flashed ESP8266 boards to 87 devices across nine ecosystems. What I built alone, what arrived once I started working with an agent, and the two examples that carry the difference.