AutoRouter

AutoRouter puts every leading model inside Codex and routes each turn to the one that does it best. Your work stays in Codex: the same app, the same chats, the same workflow. Keep using the ChatGPT plan you already have. AutoRouter saves it for the work that needs it, so it lasts longer and you stop thinking about which model to pick.

Any model, right in your picker:

Frontier and open-weight models from every major provider appear in Codex's model picker next to AutoRouter, and new models show up as soon as they ship. Pick one directly whenever you want it, without switching tools or leaving your conversation. Ask Codex which models are available, or which one fits a task, even before you set anything up.

Routing that fits the task:

Choose AutoRouter as your model and it weighs every request: how hard it is, what kind of work it is, how much of your plan is left, and your own rules. Every model has to clear a quality bar for the task before cost counts at all, so hard work gets your strongest model and quick work doesn't burn your plan.

Set up in one step:

Install the plugin and it installs the AutoRouter Mac app and opens it. Sign in, turn on Codex, and keep working in Codex the way you already do.

Your rules, in plain English:

Once the Mac app is set up, tell Codex how you want to route: "Use the strongest model for architecture work," "Use a lighter model for small fixes," or "Switch to an open-weight model once my ChatGPT plan runs out." Ask what rules you have, reword one, or remove one without leaving the conversation.

AutoRouter requires macOS and the Codex desktop app or CLI, and works with the ChatGPT plan you already have. The plugin doesn't read your ChatGPT conversations.