CarWatch turns your car into an offline AI agent on a Raspberry Pi 5

CarWatch is developed on a specific vehicle, a 2021 Mercedes-Benz E 300e plug-in hybrid named "@eclass", and connects to the car's OBD port via a Raspberry Pi to pull real-time vehicle data without going through the cloud. The project is published in Python and requires no subscription or cloud account, which sets it apart from connected-car services that typically depend on a manufacturer's backend. It also offers practical uses beyond diagnostics: managing multiple vehicles as agents in one account, detecting arrival and departure at home via Wi-Fi context rather than GPS for smart-home routines, charging the hybrid at the cheapest electricity price based on rate fluctuations, and comparing range against route plans to warn early about charging or refueling stops. A notable design choice is switching between AI models depending on the question, with larger models suited to explaining warning lights and smaller, faster ones for simple queries like current speed, and the dashboard showing each model's real performance on the Raspberry Pi. The project states four principles of honesty, including claiming only what it can sense and only what is verified, and a rule that onboard models may cite only sensor numbers they can point to in the latest actual reading. It also lists an unsolved mystery: continuous CAN broadcast frames the adapter can receive, which likely contain richer information than polled PIDs, and it invites outside community help to decode them.