OpenAI Codex CLI adds mobile ChatGPT remote-control for Linux servers

The new OpenAI Codex CLI feature lets a daemon run on a Linux server and be controlled directly from the official ChatGPT mobile app, without routing through a desktop computer. Setup starts with installing `@openai/codex@latest` via npm, then launching the daemon with `codex remote-control start`. A one-time pairing code is generated by `codex remote-control pair` and entered into “pair new device” in the app.

The author tried Telegram-based alternatives and custom-integrated systems first, but found them unsatisfactory because they added an extra mediation layer. OpenAI’s solution, according to the author, provides the intended direct control from the app, avoiding third-party bots.

Stopping the daemon uses `codex remote-control stop`. The author recommends running this command whenever connection errors appear, not only when the service is no longer needed.

Stability is still lacking: the feature is experimental and connections drop. It is therefore a proof-of-concept rather than a production-ready infrastructure, and anyone relying on it for continuous tasks should expect interruptions. The direction—mobile control of a CLI running on a remote server—is what OpenAI is developing, not a community-driven side project.

The author notes a clear gap between what is measured and what is promised: the feature works but is not stable. OpenAI has not published official documentation for `remote-control` as of the post date, and commands or behavior may change. The pairing code from `pair` is short-lived, requiring a new pairing each time the daemon restarts.

`codex remote-control stop` serves both as a clean shutdown and as a reset when the connection starts throwing errors, indicating the mechanism does not recover automatically.