Every 5 minutes the system checks solar export. When surplus exceeds 460 W it starts the EV charger, scaling amps in tiers (6 A → 18 A) to match available generation. A rolling 3-reading average prevents kettle spikes from interrupting a session.

A RandomForest model trained on historical Agile electricity prices uses wind speed (57% feature importance) and temperature to forecast costs 4 days ahead. It generates WAIT or CHARGE TONIGHT signals. Upcoming solar days are factored into the decision — free solar can substitute for cheap grid power.

The system reads Google Calendar for upcoming trips. It calculates the exact kWh needed to reach target range, then force-charges the night before departure — regardless of price signals — so the car is always ready.

The home battery force-charges to 100% at 3–4 pm, then discharges throughout the expensive 4–7 pm peak window. Grid import during peak is typically zero. The battery schedule runs 365 days a year, automatically.

When the battery reaches 100% and solar is still exporting, the system triggers the heat pump's hot water mode via API. Surplus solar that would otherwise go to the grid instead heats water for free.

Plugging in the EV at an expensive rate is automatically blocked. The system only authorises charging during solar export, cheap overnight slots identified by the AI, or pre-planned overnight sessions. Unsolicited sessions are paused within seconds.