AI Pulse · Under the Radar

Three separate solo builders have just built tools to answer the same question: where did my AI coding spend actually go?

tare lets users inspect Claude Code session logs with natural-language queries. tokentab breaks costs down by model, project and day, and picked up 211 GitHub stars in a day. cccost takes a different route, intercepting Anthropic API calls through Node’s fetch() hook to record usage directly. These are local-only tools aimed at Claude Code, Codex and Gemini CLI users. You can see the breakdown for tokentab on GitHub.

The interesting part is not which repository wins. The authors do not appear to have coordinated or even known about each other. They arrived at the same problem within days: developers are using AI coding tools heavily enough that a monthly bill is no longer enough. They want to know which project, model, session or behaviour caused the spend.

For a business owner, this is an early operational signal. AI costs become difficult to manage when they sit inside individual tools, team accounts and API invoices. Finance sees the total. The people doing the work cannot always see the cause. That makes it hard to set sensible budgets, spot waste or decide which workflows are earning their keep.

This is the kind of thing we build into an AI command centre: one place to see usage, cost and activity before small experiments turn into an expense line nobody can explain.