Codex with more than one ChatGPT account

Codex all day on a ChatGPT plan ends the same way sooner or later: the limit message, a stopped chat, and the ritual. codex logout, codex login as your other ChatGPT account, codex resume to find the session, then remind it where it was.

Shuriken has done the opposite for Claude Code since day one. This post is about Codex, which gets the same treatment, with one honest difference we explain below.

The short version#

Add every ChatGPT sign-in you own as a source. Shuriken reads how much each one has left. When the one in use reaches your switch point, or OpenAI reports a limit, your Codex chats move to the account with the most room and keep going. You don't log out of anything.

How Codex handles accounts on its own#

The Codex CLI is signed in to one account at a time. It caches that sign-in in ~/.codex/auth.json (the CODEX_HOME variable moves the folder) or in your system's credential store (OpenAI's auth docs). So the manual way to run two accounts is codex logout and codex login again, or a home folder per account and a habit of starting Codex with the right one (a shell function per account, with the sessions folder linked so codex resume sees every chat).

That works. It just isn't automatic: Codex has no idea your second folder exists, and a chat that hits a limit stays stopped until you start over somewhere else.

Add each ChatGPT sign-in as a source#

In Shuriken, open Sources, add a Codex source and sign in with your ChatGPT account. Then do it again for the next account. Each sign-in is kept apart, so their tokens never mix, and Shuriken shows each account's use side by side.

A source can also be an OpenAI API key. Turn on Switch to API keys (it's off by default) and keys take chats only once every subscription is out. A spend cap per fallback run, $25 by default, covers the whole run across every OpenAI key it uses, not each key on its own. There is more on how this works in what a source is.

The Free plan holds two sources. Pro holds as many as you add.

When it switches#

The same rules as everywhere else in source switching:

• At your switch point. 99% of what the app shows by default, and you can set it anywhere from 50% to 100%.
• At a limit. If OpenAI reports a limit before your switch point, the chat moves at that moment.
• Where to. The account with the most room left, on every device at once.
• You see it. A "Switched to …" card in the chat says which account took over and why.

If you would rather spread the work out, set How to Spread the work (our Distribute mode) and Shuriken moves to the account with the most left every 15 to 60 minutes. Switch now does it by hand.

Codex moves when the turn ends#

Here is the difference, said plainly.

A Claude Code chat moves on its next request. Shuriken handles Claude Code's requests one at a time, so when the switch point is reached in the middle of a long task, the very next request goes out on the new account, subagents and workflows included. The task never notices.

Codex chats move when the current turn ends. The turn that crossed your switch point finishes on its old account; the next turn starts on the new one, in the same chat. If OpenAI cuts the turn off with a limit before that, the chat moves right then.

So with the default 99% switch point, a long Codex turn that starts at 98% can hit the limit before it finishes, and the chat moves when OpenAI stops it. If your Codex turns run long and you'd rather they never get cut off, set the switch point lower, say 90%.

Codex limits move around#

We are not going to print Codex limit numbers in a blog post. OpenAI's own help article says limits depend on your plan and on the size of the task, and points to its pricing pages for current figures (OpenAI Help Center). They also change from time to time.

We keep a dated page instead: Codex usage limits explained, with the source and the day we read it. If you are weighing Codex against Claude Code in general, Claude Code vs Codex is the fair comparison, and Shuriken vs the Codex app covers what OpenAI's own app does.

Accounts you own#

Source switching only uses the plans and keys you add. Add accounts you own and are entitled to use, for example a personal plan and the work seat you use for that work.

Don't add someone else's login. OpenAI's terms say you may not share your account credentials or make your account available to anyone else (Terms of Use). Sharing a sign-in can put the account at risk. Where you need more room for a team, an API key is the cleaner route. The rest of what OpenAI says, from phone-number limits to Business workspaces, is quoted with its sources in can you use two ChatGPT accounts with Codex.

Try it#

Add a second ChatGPT sign-in, start a long Codex task, and watch the card appear when it's needed. Try Shuriken on macOS, Windows or Linux.