OBDient

Most car-diagnostic apps are a cloud LLM behind a chat box. OBDient plugs into any ELM327 OBD-II adapter, reads a car's engine live, and diagnoses faults in plain language through a compact model - CARpsy - that runs entirely on the phone via the QVAC SDK.

Instead of a flat vector lookup, OBDient's retrieval layer - SHIMI - is a confidence-weighted knowledge graph built on a SKOS ontology, so a question about one fault code automatically pulls in the ignition and fuel-system knowledge a real mechanic would also reach for. A node's confidence only rises once a repair's outcome is independently verified against the car's own data, never a same-session thumbs-up, which is how the system keeps validated knowledge separate from unverified guesses.

A deterministic state machine - not a learned router - splits the job across roles: a template interviewer collects the case for free, CARpsy diagnoses from it offline, and a cloud senior advisor is reached only if the owner explicitly opts in, and never receives the VIN, plate, or raw sensor readings.

Watch the OBDient demo →