I use Claude, Cursor, and other AI tools all day while building macOS apps.
A weird thing started happening: I could feel that I was spending more, but I had no clean sense of where the tokens were going while I was actually working.
By the time I opened dashboards, the useful moment was already gone. The expensive prompt had already happened. The long agent run had already finished. The "why is today so high?" question showed up too late.
So I built TokenBar.
It is a small macOS menu bar app that shows AI token usage live while you work. The goal is not finance-team analytics. It is immediate visibility.
What I wanted was simple:
- glance up and see usage moving in real time
- notice when one workflow is way more expensive than I expected
- catch runaway usage before the bill becomes a surprise
- keep cost awareness present without opening another dashboard tab
A few things surprised me once I started using it myself:
Cost intuition is usually wrong when you are moving fast.
I would assume one-off chats were the problem, then realize the real spend came from repeated background usage and long sessions.Friction matters.
If checking usage takes even 20 seconds, I do it after the fact. A menu bar app changes that because the information is already where my eyes go."Cheap enough" stacks up fast.
A bunch of small requests spread across a workday can still add up to real money, especially if you bounce between tools.
TokenBar is live at tokenbar.site and it is $5 lifetime.
I am curious how other people here keep track of AI usage while coding. Do you actually watch token spend in the moment, or do you only check after the bill lands?
Also curious which tools are the hardest to reason about cost-wise right now. Claude Code, Cursor, API playgrounds, something else?
Top comments (0)