I've been building with LLMs for about two years now. Claude, GPT-4, Gemini — I rotate through them depending on the task. And for the first year and a half, I had basically no visibility into what any of it was costing me in real time.
That's the problem nobody talks about.
Everyone talks about which model to use. Which one is smarter, which one is cheaper per million tokens, which one handles long context better. There are benchmarks and blog posts and Twitter threads comparing every model released.
But the actual day-to-day problem? You're deep in a coding session. You've got Cursor open, you're prompting Claude through the API directly, maybe you've got a small script that calls GPT-4o for some side task. And you have no idea what's accumulating.
The bill comes at the end of the month and it's just... a number. $23. $67. $112. No context. No breakdown of which session blew up. No way to know whether it was that one afternoon where you kept retrying the same prompt or the background agent you forgot was running.
The Cost Visibility Gap
Here's what's weird: we track everything else about our systems.
Memory usage? There's a widget for that. CPU? Menu bar. Network? Sure. Database query times? Logged and graphed.
But AI token spend — arguably the most variable, most unpredictable cost in modern dev work — gets zero real-time visibility. You get a monthly invoice and a vague sense of dread.
I started noticing this more as I leaned harder into AI-assisted development. Some days I'd barely use anything. Other days I'd be in problem-solving mode, having long back-and-forth conversations with Claude to debug something gnarly, and I'd blow through a surprising amount without realizing it.
The usage dashboards on these platforms are... fine. They exist. But you have to open a browser, log in, navigate to billing, and by then you've already context-switched out of whatever you were doing. Nobody checks that in the middle of a flow state.
What I Actually Wanted
I wanted the same thing I have for everything else: ambient awareness.
Not a dashboard. Not a weekly email. Just — always-on, glanceable, real-time numbers. The way my CPU gauge just sits there and I can glance at it without thinking.
So I built TokenBar — a macOS menu bar app that shows your live token usage and spend across OpenAI, Anthropic, and Gemini. It just sits in your menu bar and updates in real time.
Click it and you see the breakdown: today's spend per provider, token counts, which model is eating the most. It's the thing I kept wishing existed.
Why This Problem Is Getting Worse
As AI coding tools get better, the token usage problem compounds. Cursor, GitHub Copilot, Windsurf, Zed — these tools are all making API calls on your behalf, often more than you'd expect. Some are on flat monthly subscriptions so you don't feel it directly. But if you're also making direct API calls alongside them, you're in a situation where your costs are spread across multiple systems with no unified view.
And models are getting cheaper per token, but they're also getting smarter — so we're using them for more things. The net effect is that total spend stays surprisingly high even as per-token rates drop.
The developers who are going to get blindsided are the ones who assume "cheaper per token" means "lower bills." It usually doesn't. It means "I can now afford to use it for things I couldn't justify before."
The Fix Is Boring But Real
The solution isn't a complex observability platform. It doesn't need to be. For most individual developers and small teams, the fix is just: know what you're spending before the invoice arrives.
One number in your menu bar. That's honestly most of what you need. When you see your daily spend creeping up, you adjust. When you realize a particular workflow is expensive, you optimize it. When you've got a hard monthly budget, you can actually stay inside it.
I built TokenBar because I needed this for myself. $5 lifetime, no subscription, just the visibility that should have existed from day one.
If you're spending real money on AI APIs, you probably need this too.
