I built AI Meter because I got tired of being surprised by usage limits while working with Claude Code and Codex.
If you use Codex with a ChatGPT plan, your work counts against a short rolling window and a longer weekly window. When you hit one, the only fix is to wait for the reset (or switch to an API key, which bills per token). The trick is knowing before you hit it. Here are the ways to check.
1. /status in the Codex CLI
Type /status in a Codex CLI session. Along with the model and session details, it shows how much of your usage limits you've used and when each one resets. It's the quickest check mid-session.
2. ChatGPT's Codex settings
Signed in to ChatGPT on the web, open Codex and its settings to see your usage. Handy when you're away from your terminal.
3. Keep it on screen
The first two require you to remember to look. I make a free app, AI Meter, that keeps the remaining percentage and reset time in the Mac menu bar or on the Windows taskbar, for Codex and Claude Code. (Full disclosure: it's mine, and it's free and MIT-licensed.)
How it works, so you can judge it:
- It reads the sign-in the Codex CLI already saved (
~/.codex/auth.json) and asks the same usage endpoint the CLI uses, every five minutes. - No account, no analytics, no server of mine in between. Your token only goes to OpenAI.
- That endpoint isn't documented, so a change on OpenAI's side can break it. If that happens it shows "AI –" instead of guessing.
- The Mac build is Apple silicon only and not notarized yet, and the Windows installer isn't code-signed, so expect Gatekeeper or SmartScreen warnings. The Microsoft Store version is signed by the Store.
Install options: brew install --cask getguruhat/tap/ai-meter, Scoop, or the Microsoft Store. Source: github.com/getguruhat/ai-meter.
What about ccusage and similar tools?
Tools like ccusage read the session logs saved on your machine and add up tokens. That's useful for seeing which sessions were big, but they only see sessions on that one computer, and dollar figures are API-price estimates, not your bill. The limit shown by /status, ChatGPT and AI Meter is the one OpenAI actually enforces.
When you hit the limit
- Run
/statusto see when the window resets. You can't reset it early. - Wait for the rolling window, or the weekly reset if that's the one you hit.
- To keep going now, sign in to the Codex CLI with an API key. It bills per token and skips your plan's windows.
OpenAI changes Codex limits by plan, so its help pages have the current numbers. The full guide is kept up to date.
Which do you use? Curious whether people check /status by habit or want it always visible.
