Codex usage limits, and how to avoid mid-task cutoffs

The most frustrating way to lose work with Codex is a long agentic run that stops halfway because you've exhausted your usage. Here's how the limits work and how to stay ahead of them.

How Codex meters usage

Codex tracks usage against a rolling short window (around five hours) and a weekly window. Big, multi-step agent runs consume the short window fast; the weekly window quietly governs how much heavy work you can do across the whole week. You can sail through individual sessions and still run out mid-week.

Why long runs hit the wall

Agentic coding is bursty. A single "refactor this module and run the tests" can spend a large slice of your short window in minutes. Because the limit is rolling, the danger isn't a fixed daily quota, it's the pace over the last few hours. Without a forecast you only find out when the task halts.

The fix: forecast, don't guess

Instead of staring at a percentage, project your recent burn rate forward to the reset. If the projection crosses your limit before the window resets, you know now, and can ease off, switch to a lighter model, or finish the critical part first. If it lands comfortably under, you can push.

See it in your menu bar

ReserveGauge reads your live Codex (and Claude Code) limits directly on your Mac, forecasts your week from your real usage history, and warns you the moment you're on track to run out early. No account, no backend, nothing leaves your Mac.