Codex Cloud runs Codex coding tasks on OpenAI's machines instead of your laptop. You prepare a reusable cloud environment with your GitHub repositories, dependencies and tools, start a task from it on the web, desktop app, phone, CLI or a GitHub comment, then review the diff and commit or open a pull request.
This guide follows OpenAI's Codex Cloud overview and Cloud environments pages, read on 24 October 2026. If you mostly run Codex on your own machine, the Codex CLI guide is the better starting point.
What is Codex Cloud and when should you use it?
OpenAI's docs describe the point plainly: Codex Cloud runs coding tasks in the cloud "so work can continue while your computer is asleep." Each task gets its own isolated workspace, created from a published environment, so you can run several tasks against the same setup without them stepping on each other.
Reach for it when:
- The work is long or parallel. Hand off a refactor milestone or a batch of bug fixes and come back to finished diffs.
- Several tasks share one setup. Prepare dependencies once, publish, and every new task starts from that snapshot.
- You want to continue on another device. You can reopen the same task on the web, in the desktop app or on mobile.
Stay local when the task needs files, running services, browser sign-ins or VPN access that only exist on your computer. The Codex environments page is explicit that these "aren't automatically transferred" to a cloud task.
How do you set up a cloud environment?
A cloud environment is the reusable setup tasks start from: repositories, dependencies, tools and access settings. You create one on the web or in the desktop app, signed in with your ChatGPT account:
- In a new task, choose Work in > Cloud, open Select environment, and select Create environment.
- Pick the GitHub repositories to check out. Connect GitHub if prompted.
- Select Get started. Codex inspects the repositories, installs dependencies and tools, and tests the workflow.
- Supply missing access or information when asked. You can also request specific versions, commands or services.
- Review the setup report, configuration and files, save changes, and select Publish.
- When Environment published appears, select Start a new task.
You can also start from Settings > Codex Cloud > Environments > Create environment. You do not have to write an install script yourself. Codex records the tested setup in two fields: an Install script for dependencies and assets, and a Start skill with instructions to start services and check they are ready.
To change the setup later, open the environment's … menu, select Edit, describe the change, let Codex test it, and select Republish. New tasks use the update; existing tasks keep their own state.
How do credentials, internet and private services work?
Codex asks for values it cannot infer. Select Manage beside Environment variables or Network secrets in the environment configuration and pick the right type:
| Type | Use it for | How it is delivered |
|---|---|---|
| Environment variable | A value a program must read directly | Passed directly to programs in the environment |
| Network secret | A credential sent to a specific HTTPS service | Programs get a placeholder; the proxy swaps in the real value for allowed domains |
Network secrets work over HTTPS on port 443 and never put the raw credential in a local process or file. If teammates share an environment, each person can supply their own values from the Personal vault tab in Settings > Codex Cloud.
For internet access, turn on Allow Codex to access internet, then choose Package managers or Custom domains only and add hosts under Additional allowed domains. All (unrestricted) is there when a workflow needs broad access. The Package managers preset covers registries such as registry.npmjs.org, pypi.org, crates.io, proxy.golang.org and Maven Central, plus GitHub source and release hosts.
For internal services, the environment's Advanced settings offer a VPN connection (Tailscale is the supported provider today) and OIDC connections that give tasks short-lived credentials for AWS, Google Cloud and Azure. Support varies by service, so check OpenAI's compatibility tables before you rely on one.
How do you start a Codex Cloud task?
There are several entry points, all landing in the same kind of task:
- Web or desktop app: choose Work in > Cloud, select a published environment, describe the task and send it.
- Mobile: open Codex and select a published environment. New environments are created on the web or desktop.
- IDE extension: the
/cloudslash command runs the chat in the cloud and/cloud-environmentpicks the environment. See the Codex VS Code extension guide. - CLI: the
codex cloudcommand, marked experimental, submits and lists tasks from your terminal. - GitHub: comment
@codexon a pull request (details below). - Slack or Microsoft Teams: in Enterprise workspaces with Cloud delegation enabled, ask
@ChatGPTto work on a repository.
The CLI route is handy for scripting:
# open an interactive picker of environments and tasks
codex cloud
# submit a task to a specific environment, with up to 4 best-of-N attempts
codex cloud exec --env ENV_ID --attempts 2 "fix the flaky date parsing test"
# list recent tasks as JSON for scripts
codex cloud list --json --limit 10
--env is required for codex cloud exec, and running codex cloud lists the options. Authentication uses the same credentials as the main CLI.
How do you review the diff and ship the result?
When a task finishes, inspect the changed files and test results, ask for follow-up changes in the same task, and commit or open a pull request when you are happy. If you would rather finish the work locally, pull the diff into your checkout:
codex apply TASK_ID
Codex prints the patched files and exits non-zero if git apply fails, for example because of conflicts. Run it on a clean branch so a conflict is easy to back out.
A task's saved VM state is recoverable for up to seven days after you last start a turn or resume it. OpenAI's docs add: "Saved state doesn't replace source control." Commit anything you want to keep.
How does Codex review GitHub pull requests?
Code review runs through the GitHub integration rather than a cloud environment you manage. OpenAI's GitHub code review guide lists the steps:
- Connect your GitHub repository to Codex.
- In Codex code review settings, choose the repository and turn on code review.
- Comment
@codex reviewon a pull request. Codex reacts to the comment and posts a review that flags only P0 and P1 issues.
Turn on Automatic review under Personal preferences to review pull requests without a comment. Codex follows any ## Code Review Rules section in your AGENTS.md files, so you can teach it the checks your team cares about.
Other mentions start work too. @codex fix the P1 issue or @codex fix the CI failures starts a cloud task with the pull request as context, and Codex can push a fix to the branch when it has permission. These GitHub-triggered tasks currently run on Codex Cloud (Legacy), which OpenAI keeps for Code Review and the Linear and GitHub integrations and plans to deprecate.
Which plans include Codex Cloud?
Codex Cloud requires signing in with ChatGPT; API key sign-in has no cloud features. The Codex pricing page lists Codex cloud for Plus, Pro, Business and Enterprise or Edu, with these default VM sizes per task:
| ChatGPT plan | vCPUs | Memory | Disk |
|---|---|---|---|
| Plus, Edu Plus | 2 | 8 GiB | 8 GiB |
| Pro, Business, Enterprise | 4 | 16 GiB | 32 GiB |
| Edu, Edu Pro | 4 | 16 GiB | 32 GiB |
As of October 2026, Plus is $20 per month and Pro starts at $100 per month, per the pricing page. Local messages and cloud tasks share your plan's usage allowance, and OpenAI notes cloud tasks may use more of it than local messages.
What are the current limits?
- No computer or browser use in cloud environments yet.
- GitHub only. GitLab and self-hosted GitHub Enterprise Server are not supported for cloud environments; GitLab merge request review is in preview separately.
- Skills: skills stored in your repository work in cloud tasks, but personal skills on your computer are not synced.
- MFA: if you log in with email and password, you must set up multi-factor authentication before using Codex cloud.
Tips for getting good results from Codex Cloud
- Write an AGENTS.md first. Put build, test and lint commands in it so every task knows how to check its work.
- Keep internet access narrow. Start with the Package managers preset and add only the hosts you need. OpenAI's internet access page warns that agent internet access raises prompt injection and exfiltration risk.
- Use network secrets for tokens. A program that only sends a token to an API does not need the raw value.
- Plan locally, build in the cloud. Agree a plan in the CLI or IDE, then delegate each milestone as its own task.
- Republish when dependencies change. Tasks start from the published snapshot, not from your latest lockfile.
Keep your Codex skills where every task can find them
Because cloud tasks read skills from your repository, not from your laptop, it pays to keep skills in Git. Kipt keeps prompts and Agent Skills as Markdown in a private GitHub repo you own and installs the whole library into Codex with one npx skills add command. To turn a prompt you keep reusing into a skill, try the free SKILL.md generator, and see the Codex skills guide for where Codex looks.