Codex just moved into your pocket, and that quietly changes a lot about how developers, teams, and even non-technical folks can get work done. OpenAI is rolling out Codex inside the ChatGPT mobile app on iOS and Android, so the same agent that has been happily grinding away on your laptop or remote devbox can now be steered, questioned, and redirected from your phone, wherever you are.

At a basic level, Codex on mobile is about staying connected to active work without being chained to a desk. Once you connect the app to any machine where Codex is running – your main laptop, a dedicated Mac mini in the office, or a locked-down remote environment – the phone app loads the live state from that environment. You are not just sending one-off prompts to the cloud; you are stepping into ongoing threads with their approvals, plugins, and project context preserved. Screenshots, terminal output, diffs, test results, and approval prompts stream back to your phone in real time while all your files, credentials, and local setup stay on the machine that is actually doing the work.

That distinction matters, because this is less like a remote control app and more like a roaming cockpit for an AI coworker. Instead of “queue a job and hope it finishes by the time I’m back,” you get a full view into everything Codex is doing. You can jump between threads, review what it just changed, tell it to switch models, or spin up a new task entirely. Under the hood, OpenAI is using a secure relay layer that keeps trusted machines reachable across devices without exposing them directly to the public internet, and that same relay keeps your active session state synced across every device where you are signed into ChatGPT.

The use cases OpenAI highlights will feel familiar to anyone who has ever lost half a day to context switching. Picture debugging on a sleepy weekday morning: you are in line for coffee, a bug report pings you, and instead of making a note to “look into it later,” you tell Codex from your phone to start investigating. Because it is running inside your actual development environment, it can immediately dig into the right files, reproduce issues in the browser, run tests, and begin iterating on a fix. If it needs your approval to modify a sensitive module or run a risky migration, you tap “yes” or “no” from your phone, then glance through the diff before you get back to your desk.

Commutes become decision windows rather than dead time. Imagine you kicked off a big refactor before leaving home, banking on reviewing the result once you reach the office. Mid-commute, Codex realizes there are two reasonable ways to approach the change and pauses for guidance. From your phone, you skim its summary of trade-offs, choose the path that matches your tech debt tolerance or performance targets, and by the time you walk through the door, the work has continued in the direction you chose. Those small “yes, go that way” nudges are exactly the moments that tend to stall long-running tasks when you are away from your keyboard.

The same pattern applies outside pure engineering work. OpenAI calls out a scenario that will resonate with customer-facing teams: you step out of back-to-back meetings and discover a support issue unfolding across Slack, email, shared docs, and web dashboards, with a customer call scheduled in minutes. From your phone, you ask Codex to synthesize the latest updates, highlight the unresolved questions, and turn all of that into a tight briefing. As new details land, you can ask it to refresh the summary again before you hop on the call.

Codex on mobile also encourages you to treat passing thoughts as input, not just inspiration. When a problem or idea pops up while you are at lunch, on a walk, or half-listening to a podcast, you can drop it into an existing thread or launch a brand-new one from your phone. Codex starts sketching out a plan, exploring the code, or drafting an implementation while you stay in whatever you were doing. By the time you are back at a laptop, the work is already in motion instead of sitting in your notes app under “someday.”

A big part of this story is how well it fits enterprise and heavily regulated environments. Many teams already live inside managed remote setups that bundle approved dependencies, controlled credentials, strict security policies, and shared compute. OpenAI is leaning into that with Remote SSH, now generally available, so Codex can connect directly into those environments. The desktop app detects the hosts from your SSH configuration and lets you create projects and run threads on remote machines just like you would locally, which then also become reachable from your phone through the same secure relay infrastructure.

Once that is wired up, the workflow becomes pretty fluid: you start work on a locked-down desktop in the office, hand off long-running execution to Codex, and then keep shepherding that work from your phone without ever punching additional holes in your network or exposing machines directly to the internet. For big organizations that have spent years tightening perimeter security, that “reach it from anywhere without opening it to everywhere” design is not just convenience, it is a requirement.

OpenAI is layering on a few more pieces aimed at teams that want to automate or deeply customize how Codex behaves at scale. Programmatic access tokens can now be issued directly from ChatGPT workspace settings, with scoped credentials built for CI pipelines, release workflows, and internal automations. Hooks, which are now generally available, let you plug custom logic into Codex’s flow: scanning prompts for secrets, running extra validators, logging conversations in your own systems, creating memories, or tailoring behavior for particular repositories and directories. For healthcare environments, there is explicit support for HIPAA-compliant use of Codex in local environments – CLI, IDE, and app – for eligible ChatGPT Enterprise workspaces, which is a notable signal that OpenAI expects clinical and operational workflows to lean on these agents as well.

In terms of availability, Codex in the ChatGPT mobile app is launching in preview on both iOS and Android, and it spans all plan tiers, including Free and Go, across all supported regions. To actually try it, you update the ChatGPT mobile app and the Codex app on macOS, then link your devices so the relay layer can sync state between phone and desktop. Windows users are not left out, but they do have to wait a bit longer – support for connecting your phone to the Codex app on Windows is listed as “coming soon.” Remote SSH and Hooks land across all plans, while programmatic access tokens are reserved for Enterprise and Business tiers, and HIPAA-compliant setups apply only to qualifying Enterprise workspaces using Codex in local environments.

What this all adds up to is a subtle but important shift in how you think about AI coding tools. The early wave of assistants lived either in the editor or in the browser, and they were essentially stateless helpers: you asked for a snippet or a review, you got an answer, and that was that. Codex, as a cross-device agent, especially with a credible mobile experience, is more like a persistent collaborator that lives inside your infrastructure, moves with your projects, and can be nudged from anywhere. The “long-running thread” becomes the central unit of work, not the single prompt, and your phone becomes a natural place to keep those threads moving.

There are still open questions, of course. Developers will want to know how the relay layer behaves on shaky networks, how well mobile notifications are tuned to avoid becoming noise, and how comfortable security teams feel about the new hooks and access token patterns in practice. But the trajectory is clear: if AI agents are going to own larger chunks of the software lifecycle – and increasingly, of business workflows more broadly – they cannot be tied to one machine or one room. Putting Codex into the ChatGPT mobile app is OpenAI’s answer to that reality, and for anyone who has ever Slacked “I’ll look at it when I get back to my desk,” it might be the update that quietly kills that sentence.