Built by someone who got ripped off, for people about to get ripped off.
AmpAuto is software that reads check engine codes and tells you what's most likely broken on your specific car, before you spend a dollar at a shop. I'm Nate. I built it.
Why I built this
In 2024, my 2013 Toyota Camry threw a P0420 code. I took it to two shops. Both quoted me a catalytic converter replacement, $1,800. The "diagnostic fee" alone was $150.
That sucked. Not the saving money part. The part where, if I hadn't happened to be technical, I'd have paid $1,800 for a part I didn't need. Most people are not me. Most people pay the $1,800.
So I built AmpAuto. It uses AI trained on real failure data to tell you which of the possible causes is most likely on YOUR exact vehicle. It costs $5.99 instead of $150. It gives you the math before you spend money at the shop.
What AmpAuto is (and isn't)
It IS:
- An AI diagnostic tool. You give it a code or symptoms, it gives you back the 3 most likely causes ranked by probability, plus what to test first.
- Built on real data: OBD2 manufacturer documentation, public TSBs (technical service bulletins), NHTSA recall databases, and aggregated repair outcome data.
- Specific to your year, make, model, mileage, and engine. Generic code lookups are useless. This isn't that.
- A research tool for getting a second opinion before you authorize parts.
It is NOT:
- A replacement for a mechanic. A real wrench on your car will always beat AI.
- Always right. AI gets things wrong. We rank by probability, not certainty.
- Anything to do with selling you parts or repair services. We don't sell repairs. We just tell you what's most likely.
How the AI actually works
No black box. Here's the actual stack:
- Foundation model: Anthropic's Claude (Sonnet 4.5), a large language model with strong reasoning on technical/automotive content.
- Prompt engineering: The model is given a structured task with your exact vehicle data and codes, and asked to return a JSON-structured diagnosis with probabilities, evidence, and confirmation tests.
- Reasoning context: The model has knowledge of OBD2 specifications, common engine-specific failure patterns, manufacturer TSBs, NHTSA recall campaigns, and class-action settlements through its training data.
- Validation: Output is parsed for required structure (3 ranked causes, cost ranges, a verification test per cause, etc.). Malformed responses are rejected.
Data sources we draw on
Wherever a claim on this site references costs, failure patterns, or repair complexity, it's grounded in one of these:
Limitations we're honest about
AI is wrong sometimes. Probability rankings are not certainty. Every diagnosis ends with a confirmation test you (or your mechanic) should run before buying parts.
The AI doesn't physically see, hear, smell, or feel your car. It can't catch a chassis-related fault that doesn't trip a code. It can't replace hands-on diagnostics.
When the diagnosis doesn't help you, email a one-line reply to your receipt and we refund you in full. No questions, no forms. We win when you win.
Who I am, in case you want to verify
I'm Nate. I'm not anonymous. You can find me at the links above on the founder card. If you want to ask me something directly, email me. I read every email and reply within a day.
Talk to me
Bug report, feature request, "your diagnosis was wrong," "your diagnosis saved me $1,400," or anything else.
Email: nate@ampauto.io
Try it for yourself
Enter your code or symptoms, get the most likely cause free. Full report $5.99, refundable.
Run a free diagnosis →