OpenAI Codex CLI

With the Codex CLI provider, OpenAI's Codex runs each agent's work in its own engine, with its own sandboxing and sign-in, while Cerevisor supervises: the order agents run in, what each one depends on, and what is handed from one agent to the next all stay with Cerevisor.

The main reason to use it: you have a ChatGPT Plus or Pro subscription and don't want to pay per-token for the same model when you're already paying monthly.

Prerequisites

• Node.js (any recent LTS) on this computer. The setup screen links to it and tells you if it's missing.
• A ChatGPT Plus / Pro subscription (or an OpenAI API key; both sign-in modes are supported, but the subscription path is the typical case).
• Nothing else to install by hand — Cerevisor installs the Codex helper for you.

Setup: two steps, plus Advanced

Open it from the setup pane's ChatGPT Plus / Pro tile (small label: via Codex), or from Settings → Providers → + Add provider.

1. Install Codex

Cerevisor checks whether Codex is on this computer and shows Found or Missing. If it's missing, click Install Codex: Cerevisor runs the install for you, shows exactly what it runs, and streams the output. When it's done you'll see one plain sentence — Codex is installed. Next: sign in. on a first install, or Codex CLI updated — now on … when you used Update to latest, or Installed, but Cerevisor can't find it yet. Click "Check again". if the install worked but the helper still isn't visible.

If the install fails, the message says what to do rather than showing a raw error — for example, that Codex needs Node.js on this computer.

2. Sign in

Click Sign in with ChatGPT. Your browser opens; approve with your ChatGPT account. If the browser doesn't work for you, choose Use a code instead for a link and a code you can enter by hand.

The wait can end: after a minute Cerevisor offers help, after five minutes it stops waiting and says so, and Cancel ends it at any point. Cerevisor also notices when the sign-in stops early — you'll see Sign-in stopped before it finished. Try again, or choose "Use a code instead". rather than an endless wait. Other messages you may see: Codex didn't answer within 30 seconds. Check your internet connection, then try again. and Codex started but didn't answer. Sign in again, then try again.

Advanced

Everything that already has a sensible default lives under a collapsed Advanced: execution permissions, the Windows sandbox setting, storing the sign-in in your operating system's keychain, the default model, the connection check, and the full install log. Defaults are unchanged whether or not you open it.

Save

Save & finish becomes available once the connection check has passed — until then the button is greyed and says why in one line. If the check fails, the button reads Save anyway so you can keep the connection and sort out the problem later. Editing a connection you already saved is unchanged.

What Codex is allowed to do

Cerevisor sets Codex up for each agent from your workflow's permissions. You set the permissions you want on the workflow; Cerevisor works out the matching Codex settings.

Your workflow's permissions	What Codex gets
File writes allowed	Codex may write, but only inside the project folder. If web access is also allowed, Codex may use the network.
Everything read-only	Codex can read and nothing else.

How often Codex stops to ask follows the same permissions: a workflow that allows both file writes and shell commands runs without asking, and a read-only workflow keeps Codex at its most cautious.

Cerevisor never gives Codex unrestricted access to your computer. Writes always stay inside the project folder.

Workflow runs let Codex keep its own record of the session (in Codex's folder on your computer, outside Cerevisor) so you can chat with the agent afterward. Cerevisor's own background calls to Codex leave nothing behind.

Known v1 limitations

These are the trade-offs of using Codex as your runner, known limitations, not bugs:

• No per-tool permission prompts. Cerevisor can't step in on individual actions inside Codex. What Codex may do is set before it starts.
• No mid-run conversation continuation. Each Codex run is one request and one answer: when Cerevisor's output check finds a problem, the retry runs as a fresh Codex attempt (with the earlier output handed over as context) rather than continuing the same conversation. You can chat with a finished Codex agent, though: Cerevisor picks the saved session back up, so the follow-up has the run's full context.
• Background features run as small Codex tasks. Run retrospectives, memory upkeep, and the other background features work on a Codex default: each call becomes one small Codex task on your subscription. If a call fails (not signed in, quota exhausted), Cerevisor automatically retries it on another provider in your library that can answer it, and records any fallback spend in Analytics → Models. The chat builder works the same way when Settings → Chat Builder → "Build with a subscription provider" is on; otherwise it asks you to point it at a provider that can chat (the smoothest experience).
• MCP tools added in Cerevisor don't reach Codex. Codex has its own tools.
• Cost is reported as $0.00. With subscription billing, Cerevisor has no way to work out a per-call cost. Analytics marks these sessions as "subscription" so you can tell them apart from per-token runs.

Switching ChatGPT plans

The Codex CLI is tied to your ChatGPT account. If you upgrade or downgrade your plan, Codex picks up the change automatically, no Cerevisor action needed.

Windows note

A Windows account whose folder name contains a space used to read as signed out forever, and no longer does. A Codex installed after Cerevisor was launched is found without restarting Cerevisor.

macOS note

A Mac app opened from Finder does not always see the same programs your terminal does. Cerevisor corrects for this when it starts, so a Codex that works in your terminal is found by Cerevisor too.