I built TokenBar because I wanted one small place to answer a boring but important question:
What is my AI workflow actually using right now?
When you work across Claude Code, Codex, Cursor, OpenAI, Claude, Gemini, OpenRouter, and whatever else gets added to the stack next month, usage visibility gets messy fast. Each provider has its own dashboard, reset window, credits model, and vocabulary. Some tools show tokens. Some show dollars. Some only become clear after you hit a limit.
TokenBar is my attempt to make that visible from the macOS menu bar.
It is a local-first Mac app for tracking AI usage without turning the workflow into a spreadsheet. The menu bar view shows the things I kept checking manually:
- current AI usage
- remaining credits or limits where available
- reset windows
- alerts before limits become a problem
- top model and activity
- provider mix
- supported provider and app activity
The new layer I just added is system visibility beside AI visibility.
That means TokenBar now shows CPU, memory, energy pressure, heat, and active apps alongside AI spend and usage. The goal is not to replace Activity Monitor. The goal is to connect two things that usually live in different places.
If I am running an agent loop, editing in Cursor, testing with multiple providers, and watching my Mac get warm, I want to see the AI side and the Mac side in one glance.
In practice, it feels like Activity Monitor for AI spend, tokens, reset windows, and Mac heat.
Why system load belongs next to AI usage
AI tools are not just a line item on a bill. They affect the whole development environment.
Sometimes the expensive part is obvious because a provider dashboard shows a spike. Sometimes the pain is local. The fan spins up. The machine gets sluggish. Energy pressure climbs. A background app is doing more than expected. A coding agent is busy, but the provider dashboard does not tell you why your Mac feels worse.
Putting both views together makes the workflow easier to reason about.
For example:
- Is my token usage climbing because I am actively prompting, or because an agent loop is running in the background?
- Am I near a reset window, or do I need to slow down for the day?
- Which model or provider has been doing most of the work?
- Is the Mac under real pressure, or does it just feel warm for a moment?
- Which active apps are part of the current workload?
None of these need a giant analytics dashboard. They need a small, persistent readout that is there when you need it and quiet when you do not.
Local-first by design
The app is built as a native macOS utility, not a web dashboard. The data stays local-first, and the interface is meant to be quick enough that you can check it without context switching.
I am building it for people who use mixed AI workflows every day: coding agents, IDE assistants, direct API usage, browser chats, and provider hopping. If your day moves between Claude Code, Codex, Cursor, OpenAI, Claude, Gemini, OpenRouter, and other tools, the value is not another individual provider page. It is a single place to see what is happening.
TokenBar is priced simply: Basic is $5 lifetime, Pro is $15 lifetime.
If this sounds useful, you can check it out here:
I am especially interested in feedback from people using multiple providers or running agent-heavy Mac workflows. The next question I am working on is how much signal I can show in the menu bar before it becomes noise.
Top comments (0)