Quick practical post for anyone building with LLM APIs on Mac.
The problem
You're iterating on prompts, testing agent loops, debugging tool calls. You have no idea how many tokens you're burning until you check the dashboard later. By then, you've already spent $20 on a debug session that should've cost $2.
My solution
I built TokenBar — a native macOS menu bar app that shows token usage in real time.
Here's how I actually use it:
1. Start coding with LLM calls
I open my project, start hitting OpenAI/Claude APIs.
2. Glance at the menu bar
TokenBar shows a live counter. I can see tokens ticking up as requests run.
3. Catch problems early
If usage spikes unexpectedly during a prompt loop or retry chain, I see it immediately — not 6 hours later.
4. Stop wasting money
I've caught runaway agent loops within seconds instead of discovering $30 bills the next morning.
What it supports
- OpenAI, Claude, Gemini, Cursor, OpenRouter, Copilot, Vertex AI, and more
- Runs locally — no cloud data collection
- $5 one-time, no subscription
If you're on Mac and working with LLMs daily, check it out: tokenbar.site
Happy to answer questions about how it works under the hood.
Top comments (0)