Aura Tesla Opt: A DIN-Rail Residential Energy Hub with a Resident On-Device Language-Model Agent

Aura Tesla Opt is a wall-mounted, DIN-rail home energy hub that turns a European household's largest loads into a scheduleable resource. A certified Shelly Pro 3EM meter measures all three phases through CT clamps; four certified Shelly Pro 1PM switches (16 A each) control the water heater, wall heater, EV charger and an auxiliary circuit; a fanless N100-class mini-PC hosts Aura, a resident agent whose reasoning core is a local 8-billion-parameter model, IBM Granite 4.2-8B (dense, Apache 2.0, native chain-of-thought reasoning, ~5 GB at 4-bit quantization). Aura runs as an Aura Code service on the host: it reads telemetry, proposes schedules, and acts only after a deterministic safety policy approves each switch. The electronics sit on a 24 V DC bus backed by a 240 Wh LiFePO4 pack (~19 h host ride-through), with a DC-DC converter feeding the 12 V host and network switch. Aura runs a closed loop of hourly planning, nightly habit extraction, and a weekly governed fine-tuning cycle behind an evaluation gate with one-command rollback. Model proposes, policy disposes, outbox executes; every decision is journaled. Modelled first-year savings from load shifting: 150-400 EUR. The report also specifies industrial scalability and the path from blueprint to a working, installed unit.