Codex Cloud vs Codex remote control

Two ways to work with Codex away from your desk

Codex Cloud and Codex remote control solve different problems.

Codex remote control lets you steer Codex work from another device, such as your phone. The work runs on a connected host (the computer that does the work). The connected host supplies the files, tools, credentials, plugins, browser setup and permissions.

Codex Cloud runs tasks in a hosted cloud environment. Use it to move work off your computer or to run several tasks at the same time. Work in Codex Cloud also continues while your own computer is asleep.

Quick verdict

If the important environment is your computer or a connected SSH host, use Codex remote control.

If the task can run in a published cloud environment built from your GitHub repositories, use Codex Cloud.

After either one, review the output yourself. If other people must open the output, publish it with a static host. Examples are Cloudflare Pages, Netlify and GitHub Pages. The output can be a folder, a report, a prototype or a static site. Neither of the two Codex options does this publishing step for you.

The core difference

Question	Codex remote control	Codex Cloud
Where does the work run?	On the connected host or in its SSH project	In an isolated workspace created from a published cloud environment
How do you start or steer it?	ChatGPT on a phone, or another Mac or Windows computer	ChatGPT on the web, the desktop app (Work in > Cloud), mobile, or codex cloud in the terminal
Best for	Continuing local work from another device	Moving tasks off your computer, or running tasks in parallel
Files come from	The host machine or connected SSH project	The repositories and setup saved in the cloud environment
Does your computer need to stay awake?	Yes. If the host sleeps, loses its network or closes the app, remote access stops	No. Tasks continue while your computer is asleep
Can it use local secrets and tools?	Yes, if they are on the host and the permissions allow them	Only values that you add as environment variables or network secrets

Use remote control when files live locally

Choose remote control when:

• The repository is on your laptop or on an SSH host that you already use.
• The project depends on local tools, local services or desktop apps.
• The task needs your local browser setup or signed-in websites.
• You need to approve actions while you are away from the desk.
• You want to continue an existing local chat.
• You need files that are not in a GitHub repository.

Setup starts in the ChatGPT desktop app on the host, not in the CLI. The Remote connections guide from OpenAI gives these steps. On the host, open Settings > Connections > Control this Mac or PC. Select Set up or Add. Then scan the QR code with your phone.

Both devices must use the same ChatGPT account and workspace. In some workspaces, an admin must turn on Remote Control first.

With remote control, the host stays the source of truth (the one place that has the real files). This fact is both its strength and its limit. If the laptop sleeps, loses its network or closes the app, remote access stops until the host is back. On a Mac laptop, remote access can stay available while the lid is open and the power is connected. If the lid is closed, you also need an external display.

For the full setup and the security details, read Codex CLI remote control: setup, security and access. Also read use Codex CLI remote control from your phone.

Use Codex Cloud when the task can be offloaded

Choose Codex Cloud when:

• The repository is on GitHub, and a cloud environment can check it out.
• The environment can install the dependencies and tools of the project.
• The task does not need your local desktop, files or VPN.
• You want several attempts at the same task in parallel.
• You want a long job to continue while your laptop is closed.
• You want to review the generated changes later and open a pull request.

The Codex Cloud guide from OpenAI gives this setup procedure:

1. Open ChatGPT on the web or in the desktop app. Sign in.
2. In a new task, select Work in > Cloud. Open Select environment. Then select Create environment.
3. Select the GitHub repositories to check out. If you see a prompt to connect GitHub, connect it. Then select Get started. Codex examines the repositories, installs dependencies and tools, and tests the workflow.
4. If Codex asks about missing access or settings, answer the questions.
5. Review the setup report. Save your changes. Then select Publish.
6. After Environment published appears, select Start a new task. Describe the work.

Each task gets its own working files from the published environment. A change in one task does not update the environment for later tasks. The environments page from OpenAI also states the main limit. Codex does not move the local files, running processes, browser sign-ins or VPN access of your computer into a cloud task.

Configure cloud work deliberately

Before you depend on Codex Cloud for real work, review four items in the environment settings.

For dependencies, Codex records the tested setup as an install script and start instructions. If a task fails during setup, edit the environment. Let Codex prepare and test the change. Then publish the environment again. Existing tasks keep their own state. Only new tasks use the update.

For secrets, the cloud environments guide gives two ways to supply credentials:

Setting	Use it for	How the value is delivered
Environment variable	A value that a program must read directly	Codex passes it directly to programs in the environment
Network secret	A credential that goes to a specific HTTPS service	Programs see a placeholder. A proxy (a relay server) puts in the real value for allowed domains

For API tokens that go to HTTPS services, use network secrets. Then the raw credential never gets into a local process or a file inside the task.

For internet access, turn on Allow Codex to access internet only if the work needs it. Then select Package managers or Custom domains only. Add the hosts that you need. Use All (unrestricted) only when the workflow really needs broad access.

For source control, use Git. Saved task state does not replace it. Before you move on, commit important work or open a pull request.

Start cloud tasks from the terminal

If you already work in the Codex CLI, the experimental codex cloud command connects to the same cloud tasks. The CLI reference documents three forms:

codex cloud

To open an interactive picker (a menu in the terminal), run the command with no subcommand.

codex cloud exec --env ENV_ID "Fix the failing date parser tests"

This command submits a task directly. You must give the --env flag. The --attempts flag (1 to 4) asks for several attempts at the same task.

codex cloud list --json

This command lists recent cloud tasks in a format for scripts. To bring a finished cloud diff (the changes from a task) into your local checkout, run codex apply TASK_ID. If git apply fails, for example on a conflict, the command stops with an error.

Do not confuse either with public hosting

Codex remote control and Codex Cloud help you produce work. Neither one turns the output into a website that other people can open.

For example, Codex can create this folder:

codex-cloud-output/
  index.html
  report.md
  data/results.csv
  screenshots/
  docs/

If other people must see that folder, it still needs a publishing step. Use this short procedure:

1. Review the generated files in a browser on your own machine.
2. Remove secrets, .env files, logs and all files that are only for local use.
3. Select a static host. Cloudflare Pages accepts a folder that you drag and drop, or npx wrangler pages deploy <folder>. Netlify accepts a dropped folder or netlify deploy --prod. GitHub Pages publishes from a branch of a repository.
4. Decide who can see the site. You can put Cloudflare Access in front of Cloudflare Pages. Netlify has password protection on paid plans. Vercel has Deployment Protection. GitHub Pages sites are public, unless your organization uses GitHub Enterprise Cloud.
5. Share the stable URL of the project. When the output changes, deploy again to the same project.

Step-by-step guides are available for Cloudflare Pages, Netlify and GitHub Pages.

Example decision

Use remote control in this case:

I started Codex on my laptop.
I need to leave the desk.
I want to approve the next step from my phone.
The repo and local credentials are on the laptop.

Use Codex Cloud in this case:

I want Codex to work on a GitHub-connected project while my laptop is off.
The cloud environment is published and its tests pass.
The task can run without local desktop apps or VPN access.

After either one, use a static host in this case:

Codex created a folder with a report, demo page, CSV and screenshots.
I need to share it with a client, and only the client should open it.

Summary

Codex remote control lets you steer a host that you already trust. Codex Cloud runs a task in a prepared, isolated environment. Publishing is a third, separate step.

Use this workflow:

1. Select the Codex option that matches the location of the files and tools.
2. Review the output of Codex on a real screen.
3. Publish the folder with a static host. If the content is private, use access control.

A docs portal with Ask AI
and support tickets

Connect a repository, or describe your product to start. Each docs site has search, Ask AI, an API console, an MCP server and support tickets.

• Publishes from GitHub on every push
• Works with Mintlify repositories
• API console from your OpenAPI file
• MCP server and llms.txt for AI agents
• Ask AI and support tickets
• Analytics for people and AI agents

Start Free→ Questions? Ask AI in our docs

No card required.