How to run Codex on a remote server

What you need

• A Linux machine that stays on: a VPS, a cloud VM or a desktop at home.
• SSH access to it, with a key.
• Node.js on the server, and your repositories.
• Your OpenAI or ChatGPT account for Codex.

1. Install Codex on the server

ssh my-server
npm install -g @openai/codex

Then sign in. On a server with no browser, follow the Codex docs for signing in on a headless machine.

2. Bring your code

Clone your repository and set up Git so Codex can commit:

git clone git@github.com:you/your-repo.git
cd your-repo

You’ll need a deploy key or an SSH key on the server that GitHub accepts.

3. Run Codex inside tmux

tmux new -s codex
codex
# detach with Ctrl-b then d

Codex keeps working after you disconnect. Reattach from any device with tmux attach -t codex.

4. Keep it healthy

• Memory: watch it with free -h. If the agent dies overnight, it’s often the kernel killing a process that ran out of memory. Check dmesg for “Out of memory”.
• Disk: node_modules, build caches and Docker images fill small disks fast. Check with df -h.
• Restarts: after an OS update or a reboot, tmux sessions are gone. Start them again.
• Secrets: keep API keys out of your shell history and out of the repo.

The managed alternative

If you’d rather not look after a server, Cube gives you an always-on Linux computer with a Mac app. Sign in with GitHub, install Codex from the list, and every task gets its own worktree and a place in the sidebar. SSH is set up for you (ssh cube), and you can check in from your phone.