Almost every guide assumes you already know what's wrong. Search for a rattle, a leak, or a warning light and you get forty forum threads, each describing a different vehicle, a different year, and a different root cause. The hard part was never turning the wrench — it was identifying the thing you're looking at and deciding whether it's the actual fault or just the nearest symptom.
That's a perception problem before it's a knowledge problem. And perception is exactly what modern computer vision is good at.
The obvious way to build fi[x]in would have been to send every photo to a server, run a large model, and return an answer. It's cheaper to build and easier to update. It's also the wrong trade.
Photos of a repair are photos of your property — your driveway, your engine bay, your basement, sometimes your license plate or your front door in the background. Shipping all of that to someone else's computer to identify a corroded battery terminal is a bad bargain for the person taking the picture.
So the diagnosis models run on the device. That constraint shaped everything downstream: model architecture, size budget, how aggressively the engine can be retrained, and how each release gets validated. It is harder. It is the reason the app can work in a parking garage with no signal, and the reason identifying a corroded terminal doesn't require uploading a photo of your house.
Most apps change their models quietly. A result you got last month is not the result you get today, and nothing tells you why. fi[x]in names each generation of its on-device engine instead — Sapphire 4.8 and Emerald 5.3 are the newest, and Emerald 5.0, Sapphire 4.3 and Ruby 3.9 are still there to choose.
The two new ones, from fi[x]in 1.5.0, show what the naming is for. Sapphire 4.8 is more accurate on the same categories as 4.3, and is the default on paid plans. Emerald 5.0 recognizes 13 more kinds of problem, at some cost on the ones Sapphire already covers, so it's an option rather than a default. Sapphire 4.3 is now the free plan's engine.
Emerald 5.3, from fi[x]in 1.5.2, shows it again. It is more accurate than 5.0 on the problems both cover, but it files flat tires under tires and wheels, and boat navigation under marine electronics. So it has its own name instead of quietly taking 5.0's place, and 5.0 is still there for anyone who prefers it.
The naming isn't decoration. It means:
fi[x]in gives you a ranked set of likely causes — each labelled a strong, possible or low match rather than a bare percentage — the checks that confirm or rule it out, and the tools and supplies to have on hand if it turns out to be right. It is a very well-read second opinion that has seen a lot of photos.
It is not a substitute for a licensed professional on anything involving structural, electrical, gas, or brake safety. The app says so at the point where it matters, not buried in a policy: a fault that can make a car, boat, or house unsafe comes with a warning to stop and have it checked, rather than a push to do it yourself.
AM Solutions is a small Seattle company, not a venture-funded growth machine. That limits how fast things ship. It also means there's no pressure to monetise attention, sell data, or bolt on features that make a quarterly number look better. The business model is simple and stated plainly on the store page: the app is free, and the paid plans are a monthly subscription.
The near-term roadmap is unglamorous and specific: broaden repair category coverage, get fi[x]in Web out of beta, and get the hi[r]in professional network off the ground. Longer term, the interesting question is whether a genuinely private, on-device diagnosis engine can stay competitive with cloud models as those keep growing. Betting that it can is the whole premise of the company.
Photo-first repair diagnosis for cars, boats, and homes — running on your device, not someone's server.
