Codex remote control: run Codex from your phone

Codex remote control works through the ChatGPT mobile app. You pair it with the ChatGPT desktop app on a Mac or Windows PC, under Settings › Connections › Control this Mac or PC, scan the QR code, and you can start, steer and approve Codex work on that computer from your phone. The computer has to stay awake and online.

OpenAI shipped this in May 2026: in the week of 11 May the mobile app could connect to a Mac, as a preview, and Windows PCs followed in the week of 25 May (What's new). The terminal side is thinner. There is a codex remote-control command, but OpenAI marks it experimental and documents no phone flow for it.

How to set up Codex remote control on your phone#

OpenAI's Remote connections guide lists four things you need: Codex access in your ChatGPT account and workspace, the latest ChatGPT mobile app on iOS or Android, the latest ChatGPT desktop app on a macOS or Windows host signed in to that same account and workspace, and whatever MFA, SSO or passkey your workspace demands. In a ChatGPT workspace, an admin may also have to turn on Remote Control access first.

1. On the computer, open the ChatGPT desktop app and go to Settings › Connections › Control this Mac or PC.
2. Select Set up or Add, approve remote access and finish any verification.
3. Scan the QR code with your phone, confirm the same account and workspace in ChatGPT, and finish any MFA, SSO or passkey step.
4. In the mobile app, open Codex (older versions still call it Remote) and pick the computer.

Each QR code pairs one phone with one host, so a second computer means a second pairing. Once paired, the Codex on mobile page says you can start a task in a project on that computer, follow it, send new instructions, approve commands before Codex runs them, and review diffs and test results. The work happens on the host, with its files, credentials, MCP servers and sandbox settings.

The host is the weak point. If it sleeps, loses the network or closes the app, the phone loses it. OpenAI's advice for a Mac laptop is specific: the lid open and power connected, and with the lid closed an external display as well. Choosing Sleep stops remote access either way.

What does codex remote-control do in the CLI?#

The CLI reference describes codex remote-control as starting "remote control through the managed background server when eligible, with a foreground fallback" (developer commands). Its subcommands:

• codex remote-control start and codex remote-control stop run and stop the local app-server daemon with remote control on.
• codex remote-control --no-daemon runs it in the foreground until Ctrl-C.
• codex remote-control pair prints a short-lived manual pairing code.
• --json gives machine-readable output on any of them.

The command index tags it experimental, and the CLI's own help line reads "[experimental] Manage the app-server daemon with remote control enabled" (source). OpenAI says "managed remote-control clients and SSH remote workflows" use it. What it doesn't say is where you type that pairing code, or whether a machine started this way shows up in the phone app. The remote connections page is blunt about the phone: "Mobile setup starts from the app; you can't set it up from the Codex CLI or IDE extension." Going by the docs, the command is plumbing for those managed clients and SSH workflows, not a way to pair your phone. Run codex remote-control --help on the day, because experimental flags change.

For a terminal on another machine, the documented route is the app server itself: codex app-server --listen ws://IP:PORT on the host and codex --remote ws://host:port in the terminal you sit at. Plain ws:// is for localhost or a host forwarded over SSH only; anything else needs wss:// behind TLS. OpenAI calls the app server "primarily for development and debugging", liable to change without notice, and the remote connections page warns against exposing app-server transports on a shared or public network: reach a machine outside your network through a VPN or a mesh networking tool instead.

Codex remote on Linux and over SSH#

OpenAI's remote access "supports hosts running the ChatGPT desktop app on macOS and Windows". A ChatGPT desktop app for Linux exists in preview (Linux app), but the remote connections page doesn't list Linux as a host.

What the desktop app can do is reach a Linux machine as an SSH host:

1. Add the machine to ~/.ssh/config as a named host (Host devbox). Codex reads concrete aliases and ignores pattern-only hosts.
2. Check that ssh devbox works from the computer running the desktop app.
3. Install and sign in to Codex on the remote machine, with codex on the PATH of the remote user's login shell. The app starts the remote Codex app server through SSH.
4. In the desktop app, open Settings › Connections, enable the SSH host and choose a project folder.

Your phone still pairs with the Mac or PC, and OpenAI says ChatGPT then works in the remote environment. That means two machines stay on: the Linux box and the desktop host in front of it.

Codex cloud tasks: no computer at all#

The mobile app also offers Cloud. You pick an environment instead of a connected computer and describe the task, and it runs in OpenAI's cloud. Nothing at home has to stay awake. The trade is that the task can't reach the files and tools on your own computer.

The do-it-yourself route: SSH, tmux and Tailscale#

This works with the Codex CLI on any system, Linux included, and needs no pairing:

1. Install Tailscale on the computer and the phone, so both sit on one private network (Tailscale docs).
2. Turn on SSH on the computer (on a Mac: System Settings › General › Sharing › Remote Login).
3. Start tmux new -s codex, run codex inside it, and detach with Ctrl-b d. tmux keeps the program running after you detach (tmux wiki).
4. From an SSH app on the phone, connect to the computer's Tailscale address and run tmux attach -t codex.

You get the real CLI, approvals included, and a cramped terminal on a small screen with no alert when Codex stops to ask something. Our Claude Code phone guide walks through the same setup in more detail; only the command you run inside tmux changes.

Happy: happy codex#

Happy is an open-source (MIT) app for iOS, Android and the web. Its README says the original CLI "wraps the Claude Code and Codex harnesses directly in your terminal and lets you continue from your phone" (GitHub). Install it with npm install -g happy and start Codex with happy codex instead of codex. Traffic between your devices is end-to-end encrypted, and the computer still does the work.

Codex remote control not working?#

OpenAI's troubleshooting section covers five failures. If the computer doesn't appear on the phone, check that the desktop app is running, that Allow other devices to connect is on, and that both devices use the same ChatGPT account and workspace. A connection you haven't used since 8 June 2026 has to be paired again, after updating both apps.

Signing out of ChatGPT turns Remote Control off but keeps the pairing, so after you sign back in, turn it on again. If Add then shows an error, restart the desktop app on the host. When an approval request never reaches the phone, open Codex (or Remote) in the mobile app, check the account and scan the QR code again. A blocked setup usually means an unfinished SSO, MFA or passkey prompt, or a workspace where the admin hasn't enabled Remote Control access. And a session that keeps dropping means the host slept, lost the network or closed the app.

Anthropic's Remote Control for Claude Code fails in different ways; those fixes are in Claude Code Remote Control not working.

How Shuriken runs Codex from your phone#

Shuriken is a harness: Codex, Claude Code, Gemini CLI and the other agents run in chats on your own computers and servers, and you follow every chat from the web app at app.shuriken.sh. There's no per-host pairing on the phone. Every computer and server you've added is there, Linux included: a server is added from Settings › Devices › Add device › A server, runs headless as a service, and its installer sets up the agent CLIs. You don't need a Mac or PC in front of it.

From the phone you start a Codex chat in any project, answer its questions, read the changes in the chat's branch and reply by voice. While Codex works you can type the next message into the queue, where it waits its turn, or press Steer to send it into the running turn. The queue row with Steer is new in 5.1. What travels between devices is end-to-end encrypted, and our cloud relays it without being able to read it (what the phone app can do).

The honest limits: the web app gets account alerts by push, not "a chat needs you" alerts. Those come with the native iOS and Android apps, planned for November 2026 (until then, see phone notifications for the options). The computer still has to stay awake. A computer signed in to Shuriken starts on Keep awake: Always, and Cocaine mode keeps a Mac running with the lid closed, without the external display OpenAI's setup asks for. The phone feature page shows the rest, and if Codex runs on more than one of your machines, download Shuriken and open app.shuriken.sh on your phone.