Codex weekly limit reached: your actual options
You are mid-task, the account is out, and the refill is days away. There are four real options and one that only looks like one. None of them adds quota.
Option 1: wait, but see it coming next time
The reason a limit hurts is that it arrives without warning in the middle of something. CodexUse shows the 5-hour, 7-day, and 30-day windows Codex reports, per account, before you start. That does not add capacity; it moves the decision to before the work instead of during it. Headroom and 429s covers what those numbers mean and where they come from.
Option 2: another account you already hold
If you have a personal account and an employer seat, or separate client accounts, the next task can start on whichever one has room. Auto-roll does that without you watching: set a percent-left threshold, rank the accounts you allow, and new work rolls to the next one. A rejoin threshold brings an account back once it refills, and a confirm countdown cancels a stale decision before it fires.
The limit of this option is that it needs a second account, and a running conversation does not move between accounts. Auto-roll routes new work.
Option 3: your ChatGPT plan’s Chat allowance
This is the option most people do not know they have. The same ChatGPT subscription that feeds Codex also has a Chat allowance, and it is usually nowhere near spent when the Codex side runs dry.
A ChatGPT Web account answers Codex tasks from your own ChatGPT session, through your regular signed-in Chrome session. Turns spend the Chat allowance, so no OpenAI account’s Codex allowance is touched and no provider bills you. It gets its own Codex window and can be the worker in a run.
The cost is real and worth stating plainly. CodexUse reads the ChatGPT page, types the turn, and sends it, which is not an official OpenAI integration and may not be permitted under your agreement with OpenAI. Setup needs an OpenAI Tunnel, a secret API key with Tunnels Read and Use, and a connector you create in ChatGPT Developer Mode. And because ChatGPT reports no remaining percentage for this route, you cannot see how much of the Chat allowance is left. Read the mechanics and the terms question before you decide, then the setup guide if you do.
Option 4: a model you pay for by the token
A custom-model account points at any OpenAI-compatible endpoint: OpenRouter, DeepSeek, Groq, or a model on your own Mac. It has no OpenAI quota to run out of, because it does not use OpenAI. You pay the provider directly.
Metered billing is not automatically cheaper than the subscription you already have. It usually wins when a cheaper model is good enough for the work in front of you, and loses when you push a frontier model hard. What it reliably gives you is capacity that is not tied to a weekly window.
The one that is not an option
Buying a second subscription to run the same work twice as fast is account stacking, and it is the thing most likely to cause you a problem. CodexUse is for accounts you already hold for real reasons: your own login, an employer seat, separate client accounts. It signs in with the same flow the Codex CLI uses and never shares one subscription between people. Check your own plan terms before pooling quota for automated work.
Set it up before you need it
Every option above takes longer to arrange than the interruption you are trying to avoid. Auto-roll is a few minutes. A custom-model account is a credential and a model ID. A ChatGPT Web account is the longest, because of the tunnel and the connector. Doing any of them while blocked is the worst time to do them.