Codex Control Comes to the ChatGPT Mobile App
OpenAI has updated the ChatGPT mobile app with deep integration for the Codex coding agent, allowing developers to manage long-running coding tasks directly from their phones. Once paired with a Mac or remote development environment, the app connects to the live state of the host machine, surfacing active threads and ongoing work. Instead of being tied to a desk, developers can now oversee Codex sessions from anywhere, using the same ChatGPT interface they already rely on. This move builds on previous background computer-use features in the Codex Mac app that let agents operate independently on macOS. The new mobile layer turns those autonomous workflows into something much more collaborative and responsive, putting a real-time control panel for AI-assisted coding into a device that is always within reach.
Real-Time Monitoring, Approvals, and Collaborative Task Management
The ChatGPT mobile app is not just a remote start and stop switch for Codex. Developers can review outputs, approve or deny commands, and switch between AI models while threads are running. Terminal output, code diffs, test results, screenshots, and approval prompts stream to the phone in real time, creating a continuous feedback loop between the human and the Codex coding agent. Notifications alert users when tasks finish or when the agent needs guidance, making it easier to keep long-running jobs unblocked. From the same interface, developers can add new prompts, clarify requirements, or kick off fresh work without sitting in front of their Mac. The result is a more fluid rhythm of collaboration, where AI coding workflows can be steered in small, frequent interactions throughout the day.
Security, Remote Code Control, and Developer Workflow Management
Despite the new mobility, the design keeps sensitive assets anchored to trusted machines. Files, credentials, and permissions remain on the Mac or remote devbox, while only working outputs and status updates travel to the phone via a secure relay layer. This architecture gives developers remote code control without exposing their environments directly to the public internet. For teams, the feature can support approval-driven workflows, where a senior engineer reviews and authorizes key actions from their phone before Codex proceeds. By plugging into existing Codex sessions rather than replicating a full IDE on mobile, the ChatGPT app focuses on high-impact oversight tasks: reviewing trade-offs, granting permission for terminal executions, or redirecting AI-generated work when it drifts off-scope, all within a few taps.
From Desktop-First to Mobile-First Developer Tools
Bringing Codex workflow control to the ChatGPT mobile app signals a broader shift toward mobile-first developer tools. Codex only recently expanded from a command-line tool to a Mac desktop app, then gained the ability to use Mac applications without hijacking the cursor. Remote SSH support now lets the desktop client detect hosts from a user’s SSH configuration and run threads directly on remote environments. Layering mobile control on top of this infrastructure means developers can supervise and adjust work happening across laptops, Mac minis, and cloud devboxes from a single phone. With similar remote-control features appearing in rival AI coding tools, the competitive landscape is clearly moving beyond local IDE plugins. Real-time, device-agnostic developer workflow management is becoming a core expectation, not an optional extra.
Availability Across Plans and Platforms
The Codex control features are rolling out as a mobile preview within the ChatGPT app on iOS and Android, and are available across all subscription tiers, including the recently introduced ChatGPT Go plan. To use the new capabilities, developers must update both the ChatGPT mobile app and the Codex desktop client on macOS. Once configured, the phone can instantly load the live state of any machine running Codex, whether it is a laptop on a desk or a remote development box in the cloud. Support for connecting to Codex on Windows is still in development, but Mac users can get started immediately. For organizations, OpenAI has also expanded enterprise options around Codex, including programmatic access tokens for CI pipelines and custom hooks for scanning or logging interactions, further integrating AI agents into modern software delivery pipelines.