Watching several AI coding agents at once

Every allowance is counted on its own

Each client's windows stand apart, with their own percentages and their own reset times. Nothing is added together into one number, because a figure combining a weekly Claude window with a five-hour Codex one would not tell you when either of them stops you.

That is also what makes a second agent useful when the first runs out: its allowance was never the same allowance.

When two programs share one plan

A plan can be reached from more than one program — a Kimi or GLM coding plan used through OpenCode and through Pi, for instance. What pays is the plan, not the program, so the same credentials found in either place produce one set of windows with several clients against it, rather than the same allowance counted twice.

Two different keys stay separate, even on the same plan with the same name, the same reset time and the same percentage. Only the provider can say that two credentials are one account, and among these only Kimi has a way to.

When one of them is signed out

A client that is missing, signed out or simply failing never hides the others: the rest keep reporting, and the reason is named beside the account that could not be read.

Local activity and token counts usually survive a sign-out too, since those come from records already on your Mac. It is the quota that needs an account.

Switching accounts

• Each account keeps its own history and its own burn rate; two accounts of one client never share either.
• The account behind the latest successful reading is the current one, and only current accounts drive the glow, the menu-bar figure and the alerts.
• An account you have stopped using keeps its last reading, greyed under its name, until it has not been seen for thirty days.
• Changing accounts is neither a reset nor an exhaustion — an account that becomes current again simply starts a new alert baseline.

What the watching itself costs

Nothing, in allowance terms. Every account request is a metadata read: no model message is sent and no reset credit is spent, whichever client it is for.

The reads do not pile up either — they run one at a time rather than all at once, and a pass starts only once the previous one has finished.

Common questions

I use the same Kimi plan through two programs. Is it counted twice?

No. The plan is what pays, so both programs draw on one set of windows and appear against it together.

I have two keys on the same plan. Why are they shown separately?

Because nothing proves they are the same account. A matching plan name, reset time or percentage is not proof, and only Kimi offers a way to confirm it.

One of my agents is signed out. Does that break the rest?

No. The others keep reporting, and the one that failed says why beside its account. Its local token counts usually keep working regardless.

Why is there no single number for all my agents?

Because there is no allowance it would describe. Windows of different length, different plans and different providers do not add up into something that could tell you when you will be stopped.

Does checking all of them use up quota?

No. Each check is a metadata read — no prompt is sent and no reset credit is consumed — and the checks run one at a time rather than in parallel.

Guides

• How to check your Claude Code usage
• Claude Code rate limits, explained
• Getting told when Claude Code needs you
• How to check your Codex usage
• Codex rate limits, explained
• Getting told when Codex finishes a turn
• Quota windows and resets
• In, Out and Cache: how token usage is counted
• Agent HUD on an external monitor
• Using Agent HUD on more than one Mac
• Checking your agent usage on iPhone and Apple Watch