Remote Control

Control desktop Codex App from a phone or another device

Remote control does not run Codex directly on the phone. It connects an iPhone, iPad, Android phone, or another Codex App device to a logged-in, online, awake Mac / Windows Codex App host. Your project files, commands, browser, plugins, MCP, Computer Use, and PhotonMark config all come from that host.

Start Setup Download Codex App

Codex mobile setup screen and ChatGPT mobile Codex project list — The phone sends instructions and approvals; the desktop Codex App host executes the task.

Enable it in desktop Codex App first

Mobile setup must start from Codex App on Mac or Windows. Codex CLI or IDE Extension alone is not enough.

Allow this computer to be discovered and controlled

In Settings > Connections and enable the relevant option, such as Allow this device to be discovered and controlled. It means authorized devices in your ChatGPT account can discover and control this computer.

config.toml is only a supporting condition

prevent_idle_sleep can reduce computer sleep during long tasks; remote_connections and remote_control can remain for compatibility, but the Codex App Connections state is the source of truth.

Requirements

Desktop host

Install the latest Codex App. Mobile control hosts are currently mainly macOS and Windows.
Keep the computer powered on, online, awake, and logged in to the same ChatGPT account and workspace.
If Codex needs to operate a browser or desktop app, configure Browser, Chrome Extension, or Computer Use on that host.
If using PhotonMark Codex, PhotonMark TOML also needs to be configured in this host's user-level config.toml .

Phone or another device

Install the latest ChatGPT app on the phone and make sure the Codex entry is visible.
Use the same ChatGPT account and workspace.
Complete any MFA, SSO, or passkey steps required by the account.
A Business / Enterprise workspace may require an admin to allow Remote Control first.

How should Pay users log in?

Codex App can log in with an OpenAI API key, but that is not the same as PhotonMark Pay. API-key login uses your own OpenAI Platform API bill; Codex Pay changes the model request and quota source through PhotonMark TOML.

1. Local use can log in with an API key

For local desktop Codex App, CLI, or IDE Extension use, OpenAI API-key login is officially supported. But it does not use ChatGPT plan quota and does not use PhotonMark Pay balance.

2. Mobile remote control uses ChatGPT login

To control desktop Codex App with the ChatGPT mobile app, the host and phone still need to use the same ChatGPT account and workspace that can log in normally. This authorized-device binding cannot be replaced by an OpenAI API key.

3. Route model requests through Pay TOML

After completing ChatGPT login and phone binding, put the Codex Pay TOML from the PhotonMark dashboard at the top of the user-level config.toml . The Pay config contains requires_openai_auth = false, so model requests do not need the user's own OpenAI Platform API key.

If a free ChatGPT account requires phone verification, MFA, SSO, passkey, or another account check when logging in or enabling Codex, PhotonMark Pay cannot bypass that OpenAI account verification. Use an account that can already log in to ChatGPT mobile and see Codex, or complete OpenAI's verification flow first. API-key login can be used for some local Codex workflows, but do not treat it as a replacement for mobile remote-control account binding.

Setup steps

Start in desktop Codex App

Open the Codex App that will act as the host and click Set up Codex mobile. This flow enables remote access for this computer and shows a QR code.

Allow device discovery and control

Open Settings > Connections and confirm this computer allows authorized devices to discover and control it. If you see Allow this device to be discovered and controlled, turn it on.

This means authorized devices in your ChatGPT account can discover and control this computer. It does not expose the computer to the public internet.

Scan with phone to complete binding

Scan the QR code with the ChatGPT app on iPhone, iPad, or Android, confirm the same ChatGPT account and workspace, and complete verification. After success, this host appears in mobile Codex.

Test a small task

Open this host's project from the phone and send a simple instruction, such as viewing project files or running a read-only command. After confirming output, approval requests, screenshots, or diffs are visible, use it for longer tasks.

Feature switches that can remain in config.toml

If your user-level ~/.codex/config.toml already contains the three lines below, you can keep them. They usually make the host more suitable for remote connections and reduce sleep during long tasks. But do not use these three lines alone to decide remote control is enabled; the real connection state still depends on Codex App Settings > Connections。

[features]
prevent_idle_sleep = true
remote_connections = true
remote_control = true

Where should PhotonMark TOML go?

The dedicated TOML from the PhotonMark dashboard should still be placed at the top of the user-level config.toml at the top so model_provider switches to the PhotonMark service first.

Does the phone also need PhotonMark config?

Usually no. The phone only controls the desktop host; as long as the host has PhotonMark Codex configured, remote sessions use the same config on that host.

What can you do remotely?

Continue existing threads

Send follow-ups from the phone, answer Codex questions, and continue work already started on the desktop.

Approve commands and inspect results

Review terminal output, test results, diffs, screenshots, and approve next actions when needed.

Use the host's local capabilities

Project files, MCP, plugins, browser, Computer Use, and logged-in websites all come from the host.

FAQ

Question	Check first
Codex or this computer is not visible on the phone	Upgrade ChatGPT app; confirm the same account and workspace; confirm desktop Codex App is online and allowed to be discovered and controlled in Connections.
Task suddenly stops after connection	Check whether the host slept, lost network, quit Codex App, or a laptop was closed without an external display.
Computer Use or browser does not work	These capabilities must be enabled on the host. On Windows, Computer Use runs in the foreground and the host desktop needs to remain usable.
PhotonMark service is not active	Check the host's user-level `config.toml` contains the dedicated PhotonMark TOML, then restart Codex App or start a new session.

Official OpenAI references: Remote connections, Codex app settings, Configuration Reference。