AI resource tracker
Building a macOS app without writing a single line of codeI use Claude Code daily and i was looking for something that showed me the resources used within a current session without having to use /usage inside Claude Code or log in the Anthropic website. There were some existing apps but i wanted to see how far i could get with "vibe coding" and without any knowledge of SwiftUI or Xcode (i only use Xcode for it's simulator to test websites i build).
AI Resource Tracker
The result is AI Resource Tracker, a native macOS menu bar app written in Swift/SwiftUI. It reads your existing Claude Code OAuth credentials from Keychain, polls the Anthropic API, and shows your current usage percentage right in the menu bar. It also shows when your 5-hour usage window resets. That's it. No Electron, no web views, just a lightweight native app.
How It Went
I described what I wanted, and Claude Code started building. My job was reviewing, correcting, and steering it in the right direction. For the most part that worked surprisingly well, it scaffolded the entire Xcode project, wrote the SwiftUI views, set up the API integration, and handled Keychain access.
Where it fell apart was authentication. The OAuth flow just wouldn't work and Claude kept going in circles trying to fix it. That's where I had to step in and debug it myself to get it working. Once auth was sorted, the rest came together pretty quickly and it was a matter of connecting the dots to get something working.
So no, I didn't write any Swift. But I also wasn't just sitting back watching. It's more like directing a fast but occasionally confused developer, you need to know what you want and catch issues early.
Going from Idea to Release
The whole thing from first prompt to a working .app with a GitHub release happened in a single session. That's the part that still feels a bit surreal. I have zero Swift experience and yet there's a working native macOS app with my name on it. Free to use and you can download it via the link below.
The project is open source (MIT) and the architecture supports adding other AI providers if anyone wants to extend it.