AgentCost

Checking my usage limits meant a different dashboard for every provider, and none of them knew about the others. I built AgentCost to watch them in one place — the terminal, where the agents already run.

Live gauges show how close each usage window is to its cap, and the sessions that spent it get priced from the logs already on disk: cost per project, per model, per day, whether Claude Code, Codex, Copilot, or Gemini wrote them. A full-screen dashboard when I want to dig, plain text or JSON when I just want the number. Agents can read it too, so a session can check what it is spending while it works.

The pricing is cache-aware. Most of what an agent reads is cached at a fraction of the normal rate, so counting raw tokens makes the bill look several times worse than it is. Each token gets priced at the rate that actually applies to it, and on a subscription that total is what the same work would have cost in API credits: a running answer to whether the plan pays for itself.

No account, no telemetry, nothing to set up. It reads the logs my tools already write, and it never modifies them.

In daily use on my machine.