OpenAI brings Codex to mobile ChatGPT and puts development controls on the phone

OpenAI has expanded Codex beyond desktop: the programming and task management tool is now available in the mobile ChatGPT application. Users can launch or continue work from their phone while Codex itself executes commands on a connected computer.

How Access Works The new mode is rolling out in preview on iOS and Android.

In essence, ChatGPT on a smartphone becomes a remote interface for Codex, which continues running on a MacBook, dedicated Mac mini, or in a remote development environment. Real-time screenshots, terminal output, test results, diffs, and action confirmation requests appear on the phone. The feature is available across all ChatGPT tiers, including Free and Go, but only in supported regions.

OpenAI specifically emphasizes that files, credentials, and local permissions are not transferred to the smartphone. They remain on the machine where Codex is running, and communication between devices goes through a secure relay layer. To get started, you need to update the ChatGPT app and the Codex app on macOS, then link the devices via QR code.

The host machine must remain powered on, connected to the network, and not close Codex. Currently, a phone can connect to Codex on macOS, and the company promises Windows support will be added later.

What You Can Do From Your Phone This is not just about a "run task" button.

OpenAI describes mobile Codex as a full work surface: you can switch between threads, answer agent clarification questions, change models, approve commands, and launch new tasks without returning to your computer. This is especially useful for long scenarios where an agent hits a decision point and waits for human input. For a mobile interface, this is no longer a helper function but a way to maintain workflow momentum.

• Launch a bug analysis and watch how Codex reproduces the error and runs tests Approve the next command or choose one of the refactoring options while on the go View diffs and test results before returning to your desk * Ask Codex to compile a brief incident summary before a client call > "Now you can stay in the context of active work from anywhere," as OpenAI describes the idea of mobile Codex access. According to the company's official data, Codex is already used by more than 4 million people per week. The mobile version should fill an obvious gap: if an agent is working for tens of minutes or longer, users need more than just a desktop interface—they need a way to quickly intervene in the process without pausing for a laptop. This is especially important for teams where development is already tied to remote devbox environments and long automated pipelines. Otherwise, even a powerful agent gets stuck too easily waiting for a human response.

Why

This Matters for OpenAI The launch looks like a logical continuation of Codex's recent macOS update, where the tool learned to work with applications on a computer. Now OpenAI is adding another layer—remote management of these tasks from a phone. In practical terms, the company is building not a separate "code generator," but a broader work environment around ChatGPT and Codex.

This also strengthens ChatGPT's position as a central interface for working with agents, not just for regular chat. Against this backdrop, the move also looks like a response to intensifying competition in AI development. Following growing interest in Claude Code, OpenAI has notably accelerated Codex development and, judging by recent releases, is focusing on products that can be integrated into developers' and corporate teams' daily workflows.

The company also notes that Remote SSH is now available to all plans, so Codex can run not just on a local Mac but also inside managed corporate environments with the necessary dependencies, security policies, and compute resources.

What

This Means AI tools for development increasingly look less like a chat with suggestions and more like a permanent executor who receives direction from humans periodically. If mobile Codex proves stable in real-world use, the phone will become not a backup screen, but a normal control point for long engineering tasks.