Skip to contentGarageFix
Plans and pricingFor mechanics
Sign in

Safety

How we handle safety

GarageFix helps you think through a car problem. Some problems are not safe to drive with or to work on yourself. This page explains how we decide when to tell you so.

The safety warning

When your description points to a danger, the answer starts with a safety warning to contact a qualified workshop, shown above any likely causes or checks.

contact a qualified workshop

What triggers it

  • A safety-critical system together with a sign that it is failing: for example brakes with a pedal that sinks, steering that feels loose, or a leak.
  • A danger that is serious on its own: fire, a burning smell, a fuel leak, overheating, the oil-pressure light, exhaust fumes in the cabin, or a flashing engine light.
  • The AI's own judgement that it is not safe to keep driving.

A word like "brakes" on its own is not enough: most questions about brakes are about changing pads.

When the AI is unavailable

If the AI can't be reached, GarageFix answers from its built-in data and uses broader rules: any question about brakes, steering, airbags, high-voltage or LPG systems gets the warning. It errs on the side of caution.

Never behind a paywall

Your plan never changes the safety rules. Even when a free account has used its messages for the month, a question about a safety-critical system still gets the warning, using the broader rules described above.

In every language

The rules work in all ten languages of the site, including when you write in a different language from the one you picked.

What the AI can't do

  • It can't see, hear or test your car. It works only from your description, photos and recordings.
  • Answers can be wrong. Likely causes come with a confidence level; check before you buy parts.
  • Parts links are search links, not checked for fit.
  • The inspection pre-check is an owner pre-check, not the official test.
  • GarageFix is information only, not a substitute for a qualified mechanic.

How we check it

The warning rules are tested against labelled real-world questions in every language. A change is not accepted if it removes the warning from a case marked dangerous.