You start a refactor before leaving for a meeting. Halfway through, Codex hits a fork – two ways to solve it, and it needs you to pick one. Old workflow: that thread sits frozen until you’re back at your laptop. New workflow: you pull out your phone, glance at the diff, tap a choice, and the agent keeps going.
That’s the pitch behind Codex Mobile, and it’s a bigger shift than the name suggests.
What is Codex Mobile?
Codex Mobile is OpenAI’s coding agent, Codex, built into the ChatGPT app for iOS and Android. It rolled out as a preview on May 14, 2026, and it doesn’t run the code on your phone – it connects to a machine that’s already running Codex (your laptop, a Mac mini, a devbox, or a managed remote environment) and streams the live session to your screen.
So the work still happens where it always did. Your phone becomes a window into it.
How Codex Mobile actually works
When you open Codex inside the ChatGPT app, it loads the live state of whatever’s running on the connected machine: active threads, pending approvals, plugins, project context, the works. From there you can:
- Review what Codex has done so far
- Approve or reject commands before they run
- Switch between models mid-task
- Redirect a thread that’s headed the wrong way
- Start a brand-new task from scratch
Updates come back in real time – screenshots, terminal output, diffs, test results, approval requests. Files, credentials, and local tools never leave the host machine. OpenAI built a relay layer to connect the phone to the desktop app without exposing anything to the open internet.
It’s worth being specific here, because “AI on your phone” gets thrown around loosely. This isn’t a chatbot answering questions about your code. It’s a control surface for an agent that’s already mid-task somewhere else.
Why OpenAI built this now
Two things pushed this. First, Codex has grown fast – OpenAI says it crossed 4 million weekly users by mid-2026. Second, and more interesting: agent tasks got longer. A Codex run used to be something you kicked off and watched. Now it might run for twenty minutes, an hour, longer, and it hits decision points along the way.
That creates a specific kind of friction. A task stalls not because the agent is stuck on a hard problem, but because it’s waiting on a human to say “yes, do that” or “no, try the other approach.” If that human is in a meeting or on the train, the whole thing just… sits there.
Codex Mobile is OpenAI’s answer to that stall. Whether it actually saves meaningful time or just adds another notification to check depends on how disciplined you are about approvals – more on that below.
What’s new alongside the mobile launch
The mobile release didn’t come alone. A few enterprise-facing pieces shipped in the same announcement:
- Remote SSH, now generally available, so Codex can connect to managed remote machines with your team’s dependencies, credentials, and security policies already in place
- Hooks, also generally available, for wiring custom automation into Codex runs
- Programmatic access tokens for Business and Enterprise plans
- HIPAA-compliant use of Codex in local environments for eligible Enterprise workspaces, aimed at healthcare teams
None of that touches most solo developers directly, but it signals where OpenAI is pointing Codex: past the individual coder and toward teams running agent work as infrastructure.
Who this is actually useful for
Not everyone needs this, and that’s fine. It’s built for a specific rhythm of work.
Worth it if: you run Codex on longer tasks that hit approval points, you’re away from your desk for chunks of the day, or you manage multiple Codex threads and need to triage which ones actually need your attention right now.
Skip it if: you write short, self-contained snippets where you’re watching the output anyway, or you’re not comfortable approving code changes from a phone screen where a diff is harder to read carefully.
That second point matters more than OpenAI’s marketing lets on. Reviewing a five-line change on a 6-inch screen is fine. Reviewing a restructured auth flow is not something you want to rubber-stamp between subway stops.
How to get started
- Update the ChatGPT app on iOS or Android to the latest version.
- Make sure Codex is already running on a connected machine – laptop, Mac mini, or a remote environment you’ve set up.
- Open Codex inside the ChatGPT app and connect to that machine.
- Your active threads and approvals should load automatically from there.
The preview currently supports connecting to Codex running on macOS. Windows support is still coming, so if your Codex setup lives on a Windows box, mobile access isn’t there yet.
The bigger pattern here
Codex Mobile isn’t really about phones. It’s a sign that “using an AI coding tool” is turning into “managing an AI coding agent,” and management doesn’t happen at one desk anymore. Cursor, Claude Code, and others are all circling the same problem – how do you stay in the loop on agent work without babysitting it.
Whether that’s genuinely useful or just adds a second place to get pinged depends entirely on how you use it. Approve things you’ve actually looked at. Don’t approve things just because the app made it one tap easy.
FAQ
Does Codex Mobile run code on my phone?
No. Your phone displays and controls a session that’s running on a connected computer – laptop, Mac mini, or remote machine. Nothing executes locally on the device.
Is Codex Mobile available on all ChatGPT plans?
Yes, the preview is available across plans, including Free and Go, in all supported regions.
Can I connect Codex Mobile to a Windows PC?
Not yet. The initial preview connects to Codex running on macOS. OpenAI has said Windows support is coming.
Is my code or data sent through OpenAI’s servers unnecessarily?
Files, credentials, and local setup stay on the host machine. OpenAI built a relay layer specifically so the phone doesn’t need direct exposure to that machine over the open internet.
