Using Codex with OnDemand Apps

Codex is a terminal-based tool, so from Carina OnDemand you run it through an app’s built-in terminal. Because these sessions already run on a compute node (not a login node), you don’t need a separate srun step first.

These app terminals last as long as your OnDemand session. They aren’t subject to the 30-minute limit of Clusters > Carina Shell Access, which runs on a login node and isn’t suitable for Codex.

Note: Before using Codex here, make sure you have a Stanford ChatGPT Edu account and have logged in to Codex — see Using Codex with Carina for setup instructions.

The Command

Once you have a terminal open in any of the apps below, run:

ml codex
codex

If you haven’t logged in to Codex on Carina yet, run codex login --device-auth first and follow the login steps. You only need to do this once; after that, codex will start without asking you to log in again.

Opening a Terminal

VS Code: Terminal > New Terminal (or Ctrl+`).

RStudio: The Terminal tab next to the Console pane.

JupyterLab: From the Launcher, open a Terminal (or File > New > Terminal).

MATLAB: The MATLAB OnDemand session is a graphical environment and doesn’t include its own terminal panel. Run Codex alongside it from a separate session instead — for example, SSH into Carina and request a compute node with srun — while your MATLAB session keeps running independently.

Note: Codex can read and edit files in your project directory while your OnDemand session is open, but each app’s terminal is tied to that session — closing the session ends your Codex process too.

Using Codex with OnDemand Apps

The Command

Opening a Terminal

On This Page

See Also

Support

Request a Carina Project