I’m building an AI layer on top of Homey where all relevant events and sensor data are collected into a shared context. The AI analyzes things like phase load, heat pump operation, EV charging, and historical data, then decides when it’s appropriate to trigger Homey flows. The goal isn’t to replace Homey, but to give it a reasoning layer that can make more context-aware decisions and continuously learn over time.
Hi @MrBrillan
Will you only focus on energy stuff or will you also collect other device and sensor data like presence, lights with e.g. dimming levels… - could be helpful for habit based automations of all kind.
Yes, the idea is to log most aspects of the home’s behavior. At the moment I’m not including the dimming levels of the lights, but that’s definitely a feature that could be added later. Thanks for the suggestion!
Presence detection is already included, and it’s actually a major part of the whole concept. AI Home is designed to learn both our behavior and the home’s behavior, allowing it to adapt based on how it “knows” we prefer things to be.
Update on last night’s EV charging.
Charging started at the planned time (21:45) and completed at 00:46, delivering 32.15 kWh with plenty of margin before the 07:00 deadline. The learning loop detected that the original estimate (38.5 kWh) was about 6.3 kWh too high and automatically adjusted the ID.4 correction factor to −1.59 kWh. The next charging plan will therefore be more accurate. Despite the earlier labeling issue, the complete planning → charging → verification → learning cycle worked exactly as intended.
Update on the EV charging feature.
When I connect the first car of the day (a Nissan Townstar), I simply enter its current battery percentage in a Homey Flow. Once that’s done, the system calculates the optimal charging windows for both vehicles.
The second car, a VW ID.4, is already integrated with Homey, so its current state of charge is retrieved automatically.
The charging schedule is optimized from the moment each vehicle is connected until 07:00 the following morning, using the full electricity price curve.
AI-generated EV charging schedule:
ID.4 (Parking): 02:30–04:00 – Total cost: 0.56 SEK/kWh
Townstar (Garage): 04:15–05:45 – Total cost: 0.48 SEK/kWh
Another update on the charging project.
I’ve now concluded that the VW integration and vehicle data in Homey simply aren’t reliable enough. The delays are too long, and the reported values are sometimes incorrect.
So I took a step back and redesigned part of the charging logic.
Both of my Zaptec chargers now have Homey Flows covering 5% to 95% State of Charge in 5% increments. Whenever either car is plugged in, Homey sends me a notification asking me to enter the current battery percentage for that vehicle.
As soon as that’s done, the optimized charging plan for the evening/night is generated and displayed.
