Claude Code usage and limit tracker

Each limit shows by name and percent, for example “Claude weekly 33%, Codex weekly 84%”. It changes colour at 80 and at 95 percent. Without a fresh reading it says “unknown” rather than showing zero.

Every limit window gets its own bar, with the percent, the reset time in your local time, where the reading came from and how old it is. It also shows your pace, the hours left before you hit the limit at that pace, and how many tokens one percent is worth. Dismiss a warning and it stays away until that window resets.

Every minute, every five minutes or never, as you choose, AigoPilot asks Claude Code's own /usage command for the numbers. That calls no model and uses none of your limit.

An optional Claude Code status line, such as “5h 24%, 7d 41%”, keeps the numbers fresh while you work in a terminal. A real terminal session calling it has not yet been seen, and the VS Code panel never calls a status line, so it does not help there.

One row for every model call, by project, session, sub-agent and model. Sub-agents count under the session that started them, nothing is counted twice, and the history stays even if you delete the transcripts. Checked against an independent count, it matched to the last digit.

Each project card shows its share of this week's usage, and turns a warning colour at 25 and at 50 percent. Sort the projects by usage to see the hungry ones first.

An estimate of how much of the five-hour and weekly windows a session took, and how much its last turn took.

Give a session a budget as a percent of your week. You are told when it goes past it, and if you like, the session is held until you say continue. The budget editor works on screen. Holding a session at its budget has so far been tested only with a test program, not with the real CLI.

If a session's last hour cost more than ten times its usual hour, and more than a dollar at API prices, you get an alert, and the session can be held.

For a subscription the big number is always the percent of the limit. Any dollar figure carries the label “API value, not charged to you”.

Enter what your plan costs and see each project's share of it in dollars.

One row per session per day. It opens correctly in Excel and other spreadsheets, project names in other alphabets included.

Switch the usage page between computers. A project's share counts every computer that is switched on.

Add an API key and its credit. It is checked at once, and then daily with a free call. Spending is broken down by project, session and task, with budgets per key, project, day or month.

Warnings come at 50 and 80 percent, and at 100 percent calls stop. The worst-case cost is set aside before each call, so a budget cannot be overshot, and you can choose to fall back to a cheaper model instead. The vault, the ledger and the budgets are tested. Adding a key from the page and making a real paid call have not been done on our test machine yet.

When Codex's plan window reaches 100 percent, prompts wait instead of spending your extra credit, until you choose to spend it. The refusal and its button have not yet been seen with the real Codex.

For every account you see whether it is signed in and on which plan, as the CLI reports it. A lost sign-in is a critical alert with a sign-in button next to it.