OpenAI Brings Its Coding Agent to Smartphones with ChatGPT: How It Works and What It Doesn't Do
Since February 2026, when OpenAI released the desktop client for Codex on macOS after months of only a command-line interface, the coding agent platform has followed a predictable expansion pattern across platforms: CLI, then web, then Mac, then Windows, now mobile.
On May 14, the company announced the integration of Codex into the ChatGPT app for iOS and Android, available in preview on all plans, including Free and Go, in all supported regions.
It is worth noting that this is a remote supervision tool that does not allow for development tasks to be performed directly on the phone. The device does not serve as an interface for an already active instance elsewhere, whether it is a laptop, a dedicated Mac mini, or a remotely managed development environment. Once connected, the ChatGPT app loads the live state of that environment: active threads, pending approvals, plugins, project context. Files, credentials, and permissions remain on the host machine. The phone receives screenshots, terminal output, diffs, and real-time test results. The underlying technical layer, as described by OpenAI, is a secure relay that keeps machines reachable between devices without exposing them to the public network, with persistent session state synchronization.
The use case OpenAI has in mind is that of a coding agent working autonomously on long-duration tasks that require intermittent human input, without the need for continuous oversight. The developer starts a task, exits, and receives a notification when Codex needs a decision, whether it's to approve a command, choose between alternative implementation paths, change models, or launch a new task.
Writing code directly from the phone is not possible, and this limitation is consistent with the architecture: the execution environment remains where it is, mobile serves as the control panel.
Enterprise: Remote SSH, Hooks and Programmatic Tokens in GA
Alongside the mobile launch, OpenAI has made Remote SSH and Hooks generally available, both previously in beta. The remote SSH support allows Codex to connect to development environments managed via encrypted networks, including the dependencies, credentials, and security policies already configured in the enterprise infrastructure. The Hooks enable teams to configure specific behaviors for repositories and directories: scanning prompts to intercept secrets, validators, conversation logging, memory management, personalization for specific codebases. Both features are available on all plans. Programmatic access tokens, designed for CI/CD pipelines and automated release workflows, are limited to Enterprise and Business plans. HIPAA support, aimed at healthcare environments with patient data, is available for ChatGPT Enterprise workspaces that use Codex in local environments, covering CLI, IDE plugins, and the desktop app.
The launch of Codex for iOS and Android should be seen on two levels: on one hand, it is a response to Anthropic, which offers mobile access to Claude Code since autumn 2025, while on the other hand, it is a piece of a broader project: OpenAI confirmed in March that it is working on a super desktop app that integrates ChatGPT, Codex, and the Atlas browser into a single experience. The mobile integration is the first building blocks for the cross-device access necessary to reach that kind of convergence.
At launch, the phone can exclusively connect to the Codex app on macOS. Windows support is indicated as imminent, with no date announced. To activate the integration, both apps must be updated: ChatGPT mobile and Codex on Mac. It is worth remembering that the whole setup is currently in a ‘preview’ state and may experience bugs, ongoing changes, and other issues, so it’s advisable to evaluate carefully in which contexts to test the new features.