How OpenAI’s Codex on Mobile Turns Phones into Decision‑Making Remotes

OpenAI has quietly integrated Codex into the mobile ChatGPT app for both iOS and Android, currently available as a preview. The phone is not meant for writing code; instead, it serves as a remote control that lets you view terminal output, screenshots, code diffs, and test results from a Codex session that continues to run on your computer, Mac mini, or remote server.

When Codex stalls or requires approval for a command, you can simply tap your phone to respond. This design addresses the final shortcoming of long‑cycle agents: previously, users had to stay at their desks to make breakpoint decisions, but now those decisions can be made on a mobile device, which is more convenient for brief decision windows such as confirming an instruction or choosing between two implementation paths.

Security is handled through a secure relay layer: files, credentials, and permissions remain on the local machine where Codex runs, and only the necessary data is transmitted, avoiding direct exposure of the development environment to the public internet.

The feature solves a practical pain point for developers who run long‑running tasks. Instead of staying at their workstation waiting for results or adjusting parameters, they can follow progress and confirm next steps while commuting or away from the desk.

In addition to the mobile update, the Codex desktop client has introduced Remote SSH, allowing connections to remote development environments that include company policies and dependencies, and a Hooks feature for scanning prompts for secrets, performing validation, logging, or customizing behavior per repository.

To try the new capabilities, update both the mobile app and the desktop Codex client to the latest versions. The Windows mobile client’s desktop‑link feature is still pending.