How to check your OpenAI Codex usage and limits

Start by working out who is paying

This sounds like a detour and it is the whole guide. Codex can be authorised two ways, and they have completely different meters.

• Signed in with a ChatGPT account. Your usage is drawn against that plan's allowance, in rolling windows. There is no per-token bill, so what you want is how much of the window is left.
• Running on an API key. Every call is billed. What you want is spend, and the OpenAI platform dashboard is the only number that matches the invoice.

People searching for this are usually in the first case, running into a limit mid-task and wanting to know when it resets.

Ask the CLI first

The Codex CLI has a status command available inside a session. It reports what the current session is - the model, the working directory, the account it is signed in as - and, where the account exposes it, how much of your current rate limit window has been consumed and when it resets.

This is the fastest answer to "why did it just stop", and it is the one worth checking before anything else. Codex is a fast-moving CLI, so if the output looks different from what you expected, run its help to see what your version exposes:

Checking the tool in front of you beats trusting a command from an article, this one included. Where a version has renamed or moved something, its own help is right and the article is stale.

The authoritative numbers

There is also a machine-readable path, and it is the one Noveriq uses. The CLI ships a JSON-RPC server, codex app-server, which answers an account/rateLimits/read call using whatever login the CLI already has. That is the same rate limit figure the CLI reports about itself, without screen-scraping a terminal or parsing a session file, which is why it does not drift when the log format changes.

On a ChatGPT plan, the account settings are where the plan and its allowances live. That is the place to confirm which tier you are on and what it includes; the CLI's window readout tells you where you currently sit inside it.

On an API key, the OpenAI platform dashboard shows usage and spend for the key, broken down by day and by model, along with the rate limits attached to your organisation. This is the number that appears on the bill. A local tool can only ever estimate it.

If you hit limits sooner than expected, the same causes apply as with any agent CLI: an enormous context resent every turn, whole files read where a search would have done, and long sessions that never get compacted.

The local session logs

Codex keeps its configuration and its session logs in a .codex directory in your home folder. That is where third-party usage tools read from, and it is worth looking at yourself once:

On the version we checked, codex-cli 0.147.0, each run is written to ~/.codex/sessions/YYYY/MM/DD/rollout-<timestamp>-<uuid>.jsonl, one file per session, foldered by date. Treat that as a sighting rather than a contract: it is an internal format on a fast-moving CLI, which is exactly why the ls above is the instruction and the path is only the example. The sessions are stored as structured logs, one per run, and they carry the record of what happened in each.

If Claude Code is the other CLI you run, the equivalent guide for it covers ccusage properly, and that side of it is much more settled.

Where Noveriq fits, and where it does not

The part it does answer is the one most people arrive here with: how much of the window is gone and when it rolls over. Noveriq has a usage meter, on by default and switchable off in settings, showing your session window and weekly window as a percentage used with a countdown to the reset, alongside how much context each running session has left. With Codex selected as the source, those readings come from the Codex CLI's own rate limit reporting, asked of a codex app-server process on your own machine over local JSON-RPC, so the figure is the CLI's own rate limit reading rather than anything reconstructed from the session logs, and it stays in front of you without a session open.

The rest of it is the other half of running Codex in a terminal you are not watching. Noveriq is a native macOS app that puts every running agent session in the MacBook notch: what each is doing, how long it has been going, and which ones have stopped waiting for a human. Where Codex exposes a decision channel, you answer its requests from the notch without focusing the terminal; where it does not, Jump takes you to the exact tab or split pane instead of the application.

So it answers "how much window is left" and "does anything need me", not "what did that cost". More on the Codex page.

Questions

Why did Codex stop and tell me to come back later?

You have exhausted the current rate limit window on your plan. The CLI's status output is the place to see where the window stands and when it rolls over.

Does the ChatGPT usage page cover Codex?

Codex usage on a ChatGPT plan is drawn against that plan, so the account settings are the right place to confirm your tier. The per-window detail lives in the CLI.

Can I get a per-project breakdown?

Only from the local session logs, since that is the only record that knows which directory a run happened in. That means a parser, with the caveat about formats above.

Does Noveriq work with Codex at all?

Yes. It shows Codex sessions in the notch, answers what Codex exposes, and its usage meter can take Codex as its source, reporting the rate limit window the CLI itself reports along with the context left in each session. It reports nothing about tokens or cost.