Open source / Car control

Let your agent use your car.

Let your AI agent read codes, watch live sensors and diagnose your car through a $15 OBD-II adapter. Or a virtual one, when there's no car nearby.

Star on GitHub
No car? car-use --sim p0420
01 / Read

Codes. Signals. Readiness.

Trouble codes, the freeze frame the car saved, emissions monitors and live sensors, all in clean names with units.

02 / Understand

Reason over live data.

Watch sensors over time and compare them. The agent explains what the numbers mean, not just what the code is called.

03 / Connect

Bring your own agent.

Claude Code, Codex, OpenClaw, Hermes, anything with a shell. One prompt sets it up.

Example

A check-engine light, diagnosed.

The agent reads the code, then confirms it against live data before naming a cause, the way a good mechanic does.

you the check-engine light came on this morning
● dtcs() → P0420 catalyst efficiency below threshold, bank 1
● freeze_frame() → set at 2210 rpm, 72 km/h, fuel trims normal
● watch(["o2_b1s1", "o2_b1s2"], seconds=10)
→ rear O2 switches 15×/10s, same as the front one
agent The catalyst isn't storing oxygen. Trims are normal, so it isn't a mixture
or sensor problem upstream. Check for an exhaust leak before the rear sensor,
then the converter. ✓
Virtual cars

No car nearby? Use ours.

Built-in cars with real faults, each behaving the way the fault does on a real car. They're good for learning, for demos, and for testing your agent in CI.

p0420Dead catalytic converterThe rear O2 sensor mirrors the front one.
p0171Running leanTrims high at idle and normal under load: a vacuum leak.
p0300Random misfireUnsteady idle; cylinder codes pending on two cylinders.
cleared-codesA seller cleared the codesNo codes, but monitors incomplete after 11 km.
dying-batteryCharging faultVoltage falling with the engine running.
Fleets & heavy equipment

Cranes, excavators, trucks. One agent.

Rental and construction fleets run a mix of brands, each with its own telematics portal. car-use reads them all into one place, so an agent can answer questions and do the paperwork.

Cranes

Engine hours, location, utilisation by site, overload and safety alarms, inspection due dates.

Excavators & loaders

Working vs idle hours, fuel burn, fault codes, service intervals by engine hour.

Generators & compressors

Runtime, fuel level, alarms, and rental hours that come from the machine, not a timesheet.

Trucks & trailers

Heavy-duty J1939 data, trips, drivers, fuel and toll records, permit and fitness expiries.

Which cranes are idle, and how far are they from the sites that need one?
Which machines hit their 500-hour service this week? Book them in.
Draft this month's rental invoices from the hours the machines actually ran.

Sources: J1939 and OBD on the machine, GPS trackers, and makers' telematics feeds (ISO 15143-3). Read-only. car-use never operates lifting or moving equipment.

car-use cloud · early access

Give your car its own brain.

An always-on agent for every car, with its own LLM endpoint, a sandbox, maps and memory. The same harness and skill, reachable from anywhere.

A brain per car

An always-on agent that knows this car's history, quirks and schedule.

Sandbox

A virtual twin built from your car's own data. Try things there first.

Maps & places

Routes, chargers, fuel, parking and service centres, in the agent's context.

One API

An LLM API and a vehicle API for every car you connect, whatever the brand.

From anywhere

History, relay to the car, approvals on your phone. No laptop in the passenger seat.

Car to car

Let your cars talk to each other.Coming

Every car's agent gets an address. Your cars, your family's and your fleet's can message each other and coordinate, over Agent Intercom, the open protocol for agents talking to agents.

  • Fleets: machines hand jobs to whichever one is closest and free.
  • Families: the car with more range takes the long trip.
  • You: ask any of your cars a question from Claude Code or Codex.
youWho should do the airport run at 6?
@civicI'm at 18% fuel and need 40 km. @model-3, can you take it?
@model-3At 82%. Taking it. Preconditioning at 5:45.
# site-ops
@crane-07Lift finished at Site B. I'm free.
@dispatchSite D needs a 50 t crane tomorrow, 3.2 km from you. Assigning you; inspection is valid until the 28th.
Built for what cars become

Today it reads. Tomorrow it books the ride.

Every backend declares what it can observe and do. When a carmaker or robotaxi platform opens an API, it plugs in as a new backend. No redesign needed.

A0

Observe

Signals, trouble codes, readiness, history.

Today
A1

Act

Climate, locks, charging, clearing codes.

You approve
A2

Setpoints

Destination, charge limit, assist settings.

The car executes
A3

Missions

Drive to, summon, park.

The carmaker drives
A4

Services

Book a ride, dispatch a vehicle.

The operator drives
Safety

A car has brakes. We act like it.

A phone harness can send the wrong message. A car harness must never do the wrong thing.

Read-only by default

Raw requests are limited to read services. Coding, actuation and flashing are refused, whatever the agent sends.

The owner decides

Anything that changes the car has to be enabled per vehicle by a person at a real terminal, then confirmed each time.

A verifiable audit trail

Every decision goes into a hash-chained log. car-use audit --verify detects edits.

Never

Emissions defeats, odometer changes, immobiliser or key work, or control of a moving car.

Give your agent the keys.

Read-only keys, to start.

Star on GitHub
Copied. Paste it into your agent.