Sweep - Your floor turns green where you've vacuumed

Point your iPhone at the floor while you vacuum. Sweep follows the vacuum head and paints every strip you cover, right on your floor in AR. Finish and get a top-down map with the area in m². Works with any stick vacuum. No account, no video saved.

Add a comment

Replies

Best

Hey Product Hunt 👋

I'm Mustafa, an indie iOS developer. Sweep started as a short demo video on X: a floor turning green where the vacuum had been. It got 1M+ views, and 143 people joined the waitlist before the app was even on the App Store.

How it works: point your iPhone at the floor while you vacuum. Sweep follows the vacuum head and paints every strip you cover, right on your floor in AR. When you finish, you get a top-down map with the area in m², saved in History.

My GPT-6 Astra story: Sweep began as one line in Codex: "AR cleaning coverage map, camera-on vacuum companion." That weekend GPT-6 Astra turned it into a first iPhone prototype. The hardest part was following an ordinary vacuum head on camera, with no stickers or markers. I sent it screen recordings of failed cleans, and it analysed the frames and told me plainly when a fix still wasn't good enough (at one point only 52 of 279 frames matched). We kept iterating until the trail stayed on the floor.

In the latest update:

• A saved head for every vacuum

• Pick the room before you start

• Your own coverage colour

Free on the App Store, no account, no video saved. Works with any stick vacuum.

I'd love to hear what feels off when you try it. Honest feedback helps a lot 🙏

How did Astra change the scope or ambition of what you built?
Without Astra I wouldn't have attempted this at all. I'm a solo iOS developer, and real-time AR computer vision — following an ordinary vacuum head on camera, no marker on it, painting coverage onto the floor — was outside what I'd take on alone. It began as one line in Codex: "AR cleaning coverage map, camera-on vacuum companion." Astra had a working iPhone prototype running that weekend, then debugged it from my screen recordings of failed runs — analysing frames and telling me plainly when a fix wasn't good enough. Six days later it was on the App Store.