From desktop agent to pocket control panel
OpenAI has brought Codex preview controls into the ChatGPT mobile app on iOS and Android, turning smartphones into a live command surface for AI-driven coding agents. Instead of running code directly on the handset, Codex continues executing on a host machine such as a laptop, Mac mini, or managed devbox, while the phone mirrors its active state. Developers can open the ChatGPT app and instantly resume Codex sessions, inspect active threads, and see where long-running tasks stand without returning to their desks. Because Codex mobile control is available across Free, Go, and other ChatGPT plans, the feature targets both hobbyists and professional teams who already rely on Codex for everyday work. This shift emphasizes supervision over local execution, letting developers focus on directing and validating outcomes while the heavy lifting stays on trusted machines.
Real-time oversight of remote coding tasks
The new integration is designed for continuous oversight of remote coding tasks. Once the phone is paired with a Codex host, the ChatGPT app loads the live environment state, including active chats and pending jobs. Developers receive notifications when Codex finishes a task or pauses to ask for guidance, making it easier to keep long-running operations moving. From the mobile interface, they can review generated code, examine diffs, and check test results before deciding whether to approve or modify the next step. Screenshots and terminal output stream back to the handset in real time, bringing much of the desktop visibility to a smaller screen. Crucially, files, credentials, permissions, and local setup never leave the host environment, so only the working output traverses the secure relay layer to the phone, maintaining a clear separation between control and execution.
Steering Codex agents on the go
Codex mobile control goes beyond a simple start–stop switch to support rich AI coding agent management while away from a workstation. Through the ChatGPT iOS update and its Android counterpart, developers can switch between AI models mid-session, add new context to clarify requirements, or kick off fresh tasks by sending instructions from their phone. The system is tuned for the short decisions that often block agentic workflows: choosing between implementation paths, clarifying ambiguous specifications, or approving a high-impact command. These micro-interactions let developers manage their backlog of remote coding tasks between meetings or while commuting. OpenAI describes this as a new rhythm of collaboration, where the agent handles extended work while humans intervene at key checkpoints, gradually shifting attention from line-by-line supervision toward validating outcomes and adjusting direction as needed.
Secure relay, enterprise workflows, and platform gaps
Behind the scenes, a secure relay layer keeps Codex host machines reachable from mobile devices without exposing them to the public internet. Existing sandbox controls ensure execution remains tied to approved local or remote environments, preserving sensitive tooling and credentials. For organizations, OpenAI is pairing Codex mobile control with broader enterprise features such as remote SSH, hooks for prompt scanning and validation, and programmatic access tokens for CI and automation in eligible plans. These capabilities extend Codex into managed development environments governed by company security policies. However, there are still platform gaps: the current relay depends on Codex for Mac, and Windows phone-to-desktop support remains listed as “coming soon.” Teams already using the macOS desktop workflow can adopt mobile approvals immediately, while mixed-device shops may need to wait for broader host compatibility to fully standardize on this model.