How to use Codex from your phone

By Techguyver28 September 2026 · 7 min read

You start Codex on something and walk away from your desk. Then you want to check on it, or answer its one question, from your phone.

Short answer: You can use Codex from your phone three ways: the ChatGPT mobile app runs cloud tasks on a copy of your repo in OpenAI's sandbox, or pairs by QR code to mirror a live Codex session running on your own Mac or Windows computer. There is no official standalone remote control for the local Codex CLI outside that ChatGPT app pairing, so most SSH-based setups are third-party workarounds, not something OpenAI ships.

That single sentence hides two different products. One runs your code somewhere you do not control. The other runs it on your own machine and just gives your phone a window into it. Which one you want depends on whether you are okay with OpenAI's sandbox touching your repo, or whether the work has to stay on your Mac.

The two Codexes: cloud and local

Cloud Codex lives at chatgpt.com/codex. You pick a repository and branch, give it a prompt, and it works asynchronously on a copy of that repo inside OpenAI's own cloud sandbox, not on your computer. This is what the mobile ChatGPT app has supported since Codex rolled into it on iOS and Android, in preview, on every plan including Free, according to OpenAI's own announcement.

Local Codex is the codex command-line tool, or the Codex desktop app, running on your own Mac with your real files, your real shell, and your real git history. Nothing about it is cloud by default. This is the one that stops mid-task and waits for you to approve a command or answer a question, and it is the one an SSH client or a phone-pairing tool is trying to reach.

Mixing these up is the most common mistake in "run Codex from my phone" advice. A guide that walks you through chatgpt.com/codex is solving cloud tasks. It will not help you check on a Codex CLI session that is running against the repo on your laptop right now.

What the ChatGPT mobile app actually does for Codex

The ChatGPT app on iOS and Android gives you both halves, and they work differently.

For cloud tasks, you can start one, watch it work, review the diff it produces, and merge or discard it, the same as on the web.
For a local Codex worker, you connect the app to your computer by scanning a QR code from the ChatGPT desktop app, and the phone then mirrors the live state of that session, per OpenAI's own page: active threads, pending approvals, and diffs, updated in real time.
From either surface, the app lets you approve or reject a command, change the model, or start a new task.

The catch on the local-pairing side: the computer your phone connects to has to be running the ChatGPT desktop app, which OpenAI's remote connections guide lists for macOS and Windows (checked September 2026). A Codex session on a Linux box has no official phone-pairing path.

Is there an official Codex CLI remote control?

Not as a separate product, no. The QR-code pairing between the ChatGPT desktop app and the ChatGPT phone app is the official route, and it is specific to that pairing flow, not a general "remote control the CLI from anywhere" feature. Search "codex remote control" and you mostly find that same pairing flow described differently, plus a scattering of unofficial SSH setups.

On OpenAI's own developer community forum, a thread asking exactly this question got no official answer. The suggestions that came back were all workarounds: rent a small cloud server, SSH into it from an iPhone or Android terminal app, and run Codex there instead of on a laptop. One reply put it plainly: "it's even possible to use device dictation features, but your mileage may vary." That is a community patch, not something OpenAI built or supports.

The honest read: if your Codex session is running on a machine you own and you want to reach it from your phone, official support means the ChatGPT desktop app on a Mac or Windows computer, plus the QR pairing above. Anything else is you, or a third party, stitching it together.

The SSH route, and what it actually costs you

The workaround people land on most is: spin up a small server (a VPS, a Raspberry Pi, anything reachable), install Codex there, and SSH into it from a terminal app on your phone, such as Blink on iOS. This works, in the sense that a terminal is a terminal, and Codex behaves the same wherever it runs.

You are now running Codex on a machine that is not your laptop, which means your actual project either lives there permanently or you are pushing it back and forth.
Typing shell commands on a phone keyboard, even into a good terminal app, is slow and error-prone for anything past a one-line reply.
You are managing a server, keeping it patched, and paying for it, on top of managing Codex.
None of it pushes you a notification when Codex stops and needs an answer. You have to remember to check.

SSH is the right tool if you already wanted a cloud dev box for other reasons. It is a heavy answer to "I want to approve one command from my phone."

If Codex is running on your own Mac and you want your phone to reach it

This is the case the QR-pairing feature and the SSH workaround both address in different ways, and it is also the case MightyMouse is built for, for the piece that is not specific to Codex at all: getting your phone talking to your Mac in the first place.

MightyMouse adds itself to Codex as an MCP server, the same plug-in slot Codex uses for every other tool, and it can write that setup for you in one click. Codex is also one of the AI accounts the app itself can run, alongside Claude and the others.

With that wired up, two things become true without touching chatgpt.com or any pairing flow:

When a Codex session calls the approval tool because it wants to overwrite a file or run a risky command, MightyMouse pushes a lock-screen notification straight to your iPhone. Tap Approve or Deny, and Codex keeps going, no unlocking, no app to open. The push carries only an ID; your phone fetches the actual question directly from your Mac.
If you need to actually look at what it is doing, or take over, MightyMouse turns your iPhone into a live screen, trackpad, keyboard, and microphone for your Mac. Open Settings on your Mac, scan the code with your iPhone, and Safari opens a control page: drag to move the cursor, tap to click, hold the mic to talk. It is free and unlimited on your own Wi-Fi; away from home it runs over a secure tunnel with a free monthly cellular allowance.

That is the real difference from the SSH route: nothing here is a general-purpose remote desktop bolted onto Codex. The approval push is agent-aware because it hooks into the same MCP tool call Codex already makes; the screen and trackpad are not agent-specific at all, so they work the same regardless of what is running in the terminal. For the mechanics of why that push never carries your code through anyone's server, see how lock-screen approvals actually travel.

When this is not the answer

If your Mac is asleep or the lid is closed, there is nothing running for your phone to reach, whichever method you pick, MightyMouse included. All of this solves being away from your desk while Codex runs, not being away from a machine that is powered off.

If you actually want Codex working while your laptop is off entirely, that is the case for cloud Codex at chatgpt.com/codex, not for any of the local-pairing or remote-access options above. And if a long editing session is what you are after rather than checking in or approving a step, a 6-inch mirrored screen is a poor substitute for sitting at the real keyboard. It is fine for a quick look, not for driving a big refactor.

The tradeoff to be clear-eyed about either way: cloud Codex means a copy of your repository sits in OpenAI's sandbox while the task runs. Local Codex, reached from your phone by whatever method, keeps everything on your own machine, but only as long as that machine is on.

Approve Codex from your lock screen, or take over from your phone.
Your iPhone as a screen, trackpad and mic for your Mac. Free to try.

Common questions

Can I use Codex CLI remotely from my phone?

Officially, only by pairing the ChatGPT mobile app by QR code to a Codex session running in the ChatGPT desktop app on a Mac or Windows computer, which mirrors that session live. Anything past that, like SSH into a remote box, is a workaround, not something OpenAI ships as a CLI remote-control feature.

Does the ChatGPT app let me review diffs from Codex on my phone?

Yes, for both cloud tasks and a paired local session, you can watch diffs and test output update in real time and approve or reject them from the app.

Is cloud Codex the same as running Codex on my Mac?

No. Cloud Codex at chatgpt.com/codex works on a copy of your repository inside OpenAI's own sandbox. Local Codex runs on your actual machine, with your real files and shell, which is also why it is the one that can stop and wait for your approval.

Can I get a notification when my local Codex session needs me?

Codex does not push that on its own. MightyMouse adds it by registering as an MCP server Codex can call: when Codex asks for approval, your phone gets a lock-screen notification with real Approve and Deny buttons.

Does Codex work on Windows or Linux for phone pairing?

Windows, yes: OpenAI's remote connections guide says it supports computers running the ChatGPT desktop app on macOS and Windows (checked September 2026). Linux is not on that list, so a Linux box needs a workaround such as SSH.

Whichever agent stops mid-task and waits on you: MightyMouse turns your iPhone into a lock-screen approval button, a screen, and a trackpad for your Mac, so you do not have to SSH back in to answer it.