How we use AI

Where AI is used

Only in two places, and only when you start them:

• Ask the engineer: diagnoses your session and gives up to three setup changes and up to three driving tips, each with the metrics behind it.
• Ask the coach: writes the coach's notes in the driving report.

Everything else is computed by our own code, with no AI: laps, corners, tires, balance, the trace viewer, the track map, the driving report's numbers and session comparisons. AI output is always shown as its own cards or notes, so you can tell it apart.

When you use the site in Spanish or German, the AI answers in that language. Setup parameter, corner and track names stay as they are in iRacing, so you can find them in the garage.

The model

We use Claude, a model made by Anthropic, through Anthropic's API. We may change to a newer model version as they come out.

What the AI sees

• Sent: per-lap and per-corner summaries computed from your telemetry (times, speeds, braking, throttle, balance, tires, ride height), your setup values, our car and technique notes, and the complaint you typed.
• Not sent: the raw telemetry, your email address or account details, or the names of drivers in the session.

How answers are checked

• Every recommendation must cite the metrics that support it, and must decide whether the problem is the car or the driver before suggesting anything.
• Answers come back in a fixed structure that our code validates; a malformed answer counts as a failed run, and failed runs are refunded.
• Suggested setup values are checked against your current setup and, where we know them, against the car's legal garage range. We don't know every car's ranges yet: values we can't check are marked as not checked, so check those in the garage.
• Each answer carries a confidence level and flags thin data, such as few clean laps.

Limits

AI can be wrong, even when it sounds sure. Treat its answers as a second opinion from an engineer who has only seen your data: try one change at a time in practice, and compare sessions to see whether it worked. The advice is for sim racing in iRacing, not for real vehicles.

Training and retention

Under Anthropic's commercial terms, what we send isn't used to train its models. Anthropic keeps it for a limited period under its own retention policy. We keep the results in your account until you delete them. See the privacy policy.