VOLT A caffeine tracker that worked on me

When I saw the real numbers, I got a bit worried about my health in the future, and I basically stopped drinking them that much. It went in steps:

Tracking alone doesn’t do it. It’s also about discipline, and it’s not for everyone.

Most trackers count cups. VOLT counts what’s still working. Every drink decays on a half-life, and Home adds them all up into one number that ticks down in real time and rolls like an odometer. Under it, the day’s curve runs to your bedtime, and you can scrub it with a finger to read any hour, with soft haptic ticks. If 50 mg or more will still be active when you go to bed, the app warns you.

There are studies behind the numbers: the 5-hour average half-life (people range from about 1.5 to 9.5 hours), the 400 mg a day that the FDA and EFSA give as guidance for healthy adults, and the 50 mg bedtime warning, set cautiously after Drake et al. (2013) found that even moderate caffeine hours before bed disrupts sleep. The fifteen starter drinks use EU caffeine values and serving sizes. It’s an estimate, not medical advice.

Tracking had to be smooth, or I’d stop doing it. Quick Log on Home keeps the three drinks I use most, ranked by use. One tap logs it, a toast confirms it, and fast taps stack up to three toasts. A wrong log is a swipe away, with Undo.

The Drinks tab is the catalog: add your own drinks, edit them, and every row shows its size, so a 250 ml Red Bull and a 473 ml one read apart.

Onboarding takes four short steps, each a question: “When do you power down?” for bedtime, “How hard does it hit?” for sensitivity, “Your body, your curve” for sex, weight and age, which tune the half-life estimate, and “Draw the line” for the daily limit.

I started when Claude Design came out and tried it in my flow. It gave me three directions for Home:

I took VOLT, but the app looks different from that draft now. It looked generic and AI-sloppy to me.

I polished it with agent skills (frontend design, Apple design, Emil Kowalski’s animation skill) and by prompting at whatever felt off, until the classic AI-generated details were gone: the weird color lines, the pills. The motion follows Emil’s principles: ease-out for things that enter and leave, a spring for things that move on screen, and under about 0.3 s for anything you touch. Reduce Motion is respected everywhere.

I didn’t write any of the code. Claude Code built it, Codex reviewed it, and I stayed on the visuals and the UX, as usual. It’s my first iOS app, in SwiftUI.

Two bugs taught me the most. I wanted a colorless blur under the status bar. SwiftUI’s material added a grey tint, and the next try, a Metal shader over the whole app, turned every screen black. We rolled it back, and there’s no top effect now. The other came and went: after a few quick logs, the list stopped scrolling. It wasn’t our code but a known bug in Apple’s iOS 18 and 26 SDKs, with swipe gestures inside scroll views, so the swipe to remove now runs on UIKit underneath. Every failure like that goes into a lessons log in the repo, so the agents don’t try the same thing twice.