Agent HUD · Guides

Codex rate limits, explained

Codex reports its limits as windows that carry their own length, so what you see depends on what your plan reports rather than on a fixed shape.

Primary and secondary

A plan usually meters two windows at once. Each arrives with its own duration, so one may be measured in hours and the other across a week, and either can be the one that stops you.

Because the length comes from the service, two accounts on nominally the same plan can show different windows. Agent HUD labels each one by the duration it reported rather than assuming a shape.

Reset credits

Alongside the windows, Codex reports how many resets are still available. That is a count, not a percentage, and it is shown exactly as given — the list of expiry dates beside it is supplementary and never used to work the count out.

Spending one is a way out of a full window that waiting is not, which is why the number is worth seeing before you decide to stop for the day.

What spends them fastest

  • Reasoning effort. Codex itself warns that a higher effort consumes usage limits faster, and on a smaller plan the difference is not subtle.
  • Long unbroken sessions, which keep the shorter window busy from end to end.
  • Work that another agent could have done on its own separate allowance.

The context window is a different number

Codex also reports how much of the conversation's context is used. Compacting or starting a new thread clears that; neither touches your usage limit. Two numbers counting to 100% are easy to confuse, and only one of them ends your afternoon.

One account, two programs

Codex Desktop and the Codex CLI sign in to the same account, so they draw on one set of windows. Agent HUD shows them once rather than twice, and usage recorded by one is never copied into the other.

A ChatGPT account you signed into through Pi counts as the same account too. Seen from more than one place it still produces one set of windows, while a genuinely separate account keeps its own notices and reset credits.

Common questions

Why does my window say 5h when someone else's says weekly?
Both are real. Codex reports each window with its own duration, and plans differ. Agent HUD labels what arrived rather than assuming every plan has the same shape.
Do reset credits refill?
The service decides, and reports how many are currently available. Agent HUD shows that number as given and does not predict the next one.
Does lowering reasoning effort really help?
Codex warns that a higher effort consumes usage limits faster, so yes for routine work. It is the cheapest lever before waiting for a reset.
Where do I see all of this at once?
/status gives a breakdown inside a session. Agent HUD keeps the windows and the reset countdown on screen outside any one session, next to your other agents.

Guides

Last updated: 2026-09-20 · The behaviour described here is documented in full, with the code it comes from, in Agent HUD Open. github.com/jazzenchen/agent-hud-open