A few months ago I kept doing the same dumb thing: I’d open my editor, ask an AI to help with one small task, and then keep working like nothing had changed.
No dashboard. No budget check. No feeling for what the model was actually costing me.
That worked fine right up until it didn’t.
The bills were never catastrophic in one shot. That’s the sneaky part. It was death by a thousand tiny prompts. A refactor here, a copy tweak there, a long context dump because I was too lazy to trim the conversation. By the time the invoice showed up, the useful signal was gone. I wasn’t reacting to the spend. I was reacting to a monthly postmortem.
I’m a solo dev. I don’t need another dashboard tab to babysit. I need the number in my face while I’m building.
That’s why I built TokenBar: tokenbar.site
It sits in the menu bar and shows me what the AI is doing right now. Not at the end of the month. Right now. It’s the difference between a smoke alarm and a detective report.
Once I started seeing cost as a live signal, my behavior changed immediately. I cut bloated prompts. I stopped tossing entire files into context just because I could. I noticed when a model switch was quietly tripling my spend. And I stopped pretending that “I’ll look at it later” was a strategy.
The weirdest part is how boring it feels once it’s visible. That’s exactly the point. You don’t want AI costs to be a mystery. You want them to be normal, obvious, and hard to ignore.
That’s the whole idea behind TokenBar. Keep the signal visible. Keep the surprise bills small.
Top comments (0)