OpenAI has released a significant update to its ChatGPT mobile application, integrating remote access capabilities for its powerful Codex agent system. Available now for iPhone, iPad, and Android devices, this update allows users to interact with Codex from anywhere, approving tasks, monitoring progress, and issuing new commands without being physically present at their computer.
Codex, which originally launched as a command-line interface tool before arriving as a standalone Mac application in February, is OpenAI's specialized AI agent designed to autonomously complete complex software development and system administration tasks. The agent can interact with development environments, run commands, edit code, and manage files across multiple machines. Until now, users needed to be at their Mac or have a terminal open to direct Codex's work. The new mobile integration changes that dynamic entirely.
The feature lives directly inside the existing ChatGPT app, rather than requiring a separate Codex mobile app. Users who have Codex running on their Mac—or on a remote machine like a dedicated Mac mini or managed devbox—can connect their phone by scanning a QR code displayed in the Mac app. Once paired, the ChatGPT app loads the live state from the connected environment, providing access to active threads, approval requests, plugins, and project context.
More Than Basic Remote Control
OpenAI emphasizes that the mobile experience goes far beyond simple remote desktop functionality. "Codex in the ChatGPT mobile app is a fully-featured mobile experience for getting work done with Codex," the company stated in its announcement. Users can work across all their threads, review outputs including screenshots, terminal output, diffs, test results, and approval requests, change models, or start entirely new tasks—all from their phone.
This capability is particularly valuable as AI agents increasingly handle long-running, complex workflows. The ability to approve a command, redirect an agent's approach, or review intermediate results without being tethered to a workstation allows developers and power users to maintain momentum even when away from their primary machine. Files, credentials, permissions, and local setups remain securely on the machine where Codex operates, while updates flow in real time to the mobile device.
Setup and Availability
To get started, users need the latest version of Codex for Mac and the latest version of ChatGPT for iOS or Android. After launching Codex on the Mac, the app displays a QR code. Scanning that code from the ChatGPT mobile app establishes the connection. OpenAI says the setup process is streamlined and takes only seconds.
The feature is rolling out today as a preview in all supported regions for iPhone, iPad, and Android. OpenAI also confirmed that support for connecting a phone to Codex running on Windows will follow, though no specific timeline has been provided.
Recent Codex Enhancements
The mobile remote access update builds on several recent improvements to the Codex platform. Last month, Codex gained the ability to use applications on a Mac without taking over the cursor, allowing users to continue working on their machine while Codex executes tasks in parallel. This parallel operation mode, sometimes called "background mode," significantly boosts productivity by letting the agent work independently without blocking user input.
OpenAI has also introduced a dedicated subscription plan for Codex users, recognizing that heavy usage of the agent system requires more resources than standard ChatGPT subscriptions provide. Additionally, the recent release of GPT-5.5 has enhanced Codex's reasoning and code-generation capabilities, making it more reliable for complex tasks. The Images 2 model upgrade has improved Codex's ability to generate and analyze visual outputs, such as UI mockups and data visualizations.
The Evolution of Codex
Codex began as a command-line interface tool, offering developers a way to interact with AI directly within their terminal environments. Its transition to a Mac app in February marked a significant shift, providing a graphical interface and deeper integration with macOS. The addition of mobile remote access completes the triad—desktop, local terminal, and now mobile—ensuring that users can maintain a constant connection to their AI agent regardless of their physical location.
This development aligns with broader industry trends toward agentic AI systems that operate autonomously but require human oversight for critical decisions. By enabling mobile approval and monitoring, OpenAI addresses a key pain point: the need for timely human input without requiring constant presence at a desk. The approach mirrors similar features seen in enterprise DevOps tools, where engineers approve deployments and review build logs from their phones.
The mobile integration also opens up new use cases. Developers can now start a long-running code refactoring task from their laptop, then step away for a meeting or commute while checking progress via their phone. When the agent encounters a decision point—such as choosing between two implementation approaches—it can pause and wait for mobile approval before proceeding. This asynchronous collaboration model is expected to become standard as AI agents take on more complex, multi-step tasks.
Implications for Productivity
Early adopters report that the mobile remote access feature has transformed their workflow. Being able to review diffs and test results while away from the computer reduces context-switching overhead. Instead of waiting until returning to a desk to re-engage with a task, users can stay in the loop continuously. The ability to change direction mid-task—for example, asking Codex to refocus on a different module or adjust its approach—directly from a phone creates a fluid, responsive collaboration loop.
Security considerations are handled carefully. All sensitive credentials, API keys, and project files remain on the host machine. The mobile app only receives real-time output streams and approval requests, not the underlying code or data. OpenAI uses end-to-end encryption for the connection, ensuring that intermediate outputs like terminal screenshots and diff displays cannot be intercepted.
Competing AI agent systems, such as GitHub Copilot's agent mode and Anthropic's computer-use capabilities, are also exploring mobile interfaces, but OpenAI's approach of embedding the functionality directly within ChatGPT gives it a unique distribution advantage. Millions of users already have ChatGPT installed, making adoption frictionless.
Looking ahead, support for Windows remote control will likely be the next major milestone, broadening Codex's reach beyond the Apple ecosystem. Given that many developers work on Windows machines or use Windows-based devboxes, this addition will be critical for widespread enterprise adoption. OpenAI has not yet announced a timeline, but the company's statement suggests it is actively working on Windows support.
Codex's mobile integration represents another step toward ubiquitous AI assistance—where intelligent agents are always accessible, always aware of user context, and always ready to act on instructions, regardless of the device in hand. As AI models continue to improve in reasoning and reliability, the boundary between human and machine work will blur further, making tools like remote mobile access not just conveniences but necessities.
Source: 9to5Mac News