OpenAI brings Codex to ChatGPT on iOS and Android for remote development control

On May 14, 2026, OpenAI began a preview launch of Codex in the ChatGPT mobile app on iOS and Android. The coding agent no longer has to be started and managed from a desktop alone: you can now follow it from a smartphone, see what it is doing, answer its questions and approve its next steps.

How it works

Mobile Codex connects to an environment that is already running: a laptop, a devbox, a dedicated Mac mini or a remote machine. On the phone, the user sees the live state of that environment almost as if sitting at the workstation: screenshots, terminal output, diffs, test results and requests to approve actions. The files, credentials, permissions and local configuration themselves stay on the machine where the agent actually runs. That matters for remote work and long runs.

"From your phone, you can work with all threads, review results,

approve commands and start new tasks."

OpenAI stresses that this is not just a remote button for a single session. The state of active threads is synchronized across devices through a secure relay layer, and the machine running Codex is not exposed directly to the public internet. In other words, the phone does not replace the work computer. It becomes a control panel for long tasks that the agent runs in the background for hours, keeping the same context, decision history and pace between steps.

What's new

The main idea of the update is to keep the developer in the loop even after stepping away from the laptop for half an hour or leaving for the city. Codex still does the work in the main environment, but the phone becomes a point for quick decisions: you can check progress, clear a blocker, adjust direction and send the task on its way for the rest of the day without waiting to get back to the desk.

• follow active threads and task status
• approve commands when the agent hits a permission prompt
• switch the model and direction of work on the go
• start a new bug report, refactoring or research task

OpenAI describes fairly everyday scenarios: asking Codex to investigate a bug while you wait for coffee; choosing between two refactoring options on the road; quickly assembling an incident summary before a client call; dropping a new idea into a separate thread before it slips away.

The feature is rolling out in preview on iOS and Android to all plans, including Free and Go, in all supported regions. The company promised to add support for connecting to the Codex app on Windows later.

A bet on developers

For OpenAI this is not an isolated feature but part of a broader race for a place in the daily development workflow. The company says more than 4 million people use Codex every week. In April the agent gained the ability to work longer in the background on desktop, and in May OpenAI added a Chrome extension so it can act directly in live browser sessions. The mobile layer logically closes another gap: long tasks no longer stop just because a person has left their computer.

Against this backdrop, competition with Anthropic is especially visible. In February 2026 the company released Remote Control for Claude Code, built on a similar idea of remote monitoring and control. OpenAI is responding not only with a mobile interface but also with infrastructure updates for teams: Remote SSH is now generally available, programmatic access tokens have appeared for CI and internal automation, and hooks can now be used for secret scanning, logging, validators and customizing agent behavior in specific repositories. For enterprise customers this is no longer a toy but groundwork for a permanent development layer.

What this means

The market for AI coding tools is moving quickly from "write me a function" to "run the work yourself, and I'll step in when needed." If this approach takes hold, the smartphone will become for developers not a place to write code but a control point for agents that work continuously in a local or remote environment. For teams, that means less idle time between steps and easier management of long tasks.