Codex Cloud explained: how OpenAI's cloud coding tasks work, who gets them and how to set one up

Codex Cloud is the part of OpenAI’s Codex coding agent that runs on OpenAI’s computers instead of yours. It was one of the developer announcements at DevDay on September 29, 2026, where OpenAI said “now developers can run Codex wherever they need it: on a computer, remotely from a phone, or in the cloud from any device”. This guide explains how it works, how to set up an environment, who can use it and what its limits are, based on OpenAI’s Help Center and developer docs. We haven’t run these tasks ourselves.

What Codex Cloud does

OpenAI says Codex Cloud “runs coding tasks on OpenAI-managed computers”. The flow is:

1. You prepare an environment for your project, which brings together the repositories, dependencies, tools and access settings Codex needs.
2. You publish it, which lets you reuse that setup for new tasks.
3. You start a task, such as investigating a bug, making a code change or running the project’s tests. Each one starts from the prepared setup in its own isolated workspace, with separate working files and changes.
4. The task keeps running while your computer is asleep, and you can pick it up from desktop, web or mobile.
5. You review the changes and test results before using the work, and continue in the same task if more changes are needed.

Environments are created and published on desktop or web. Once one exists you can select it for a mobile task and continue the same cloud task across supported devices.

Who can use it

OpenAI’s DevDay recap lists Codex in the cloud as available on Plus, Pro, Business, Healthcare, Education and Enterprise plans. Plan usage and limits follow your ChatGPT plan, as described in OpenAI’s page on using Codex with your plan, and in a managed workspace your administrator controls Codex cloud access. Being an owner or admin doesn’t by itself make every feature available on your plan. See ChatGPT Pro explained and ChatGPT Go vs Plus for what the plans include, and GPT-6.1 Sol for the newest model available in Codex.

How to create an environment

According to OpenAI’s developer documentation, on the web or in the desktop app, signed in with your ChatGPT account:

1. In a new task, choose Work in, then Cloud, open Select environment and choose Create environment. You can also go to Settings, Codex Cloud, Environments.
2. Select the GitHub repositories to check out, connecting GitHub if prompted.
3. Select Get started. Codex inspects the repositories, installs dependencies and tools, and tests the workflow.
4. Supply anything it asks for and request specific versions, commands or services if you need them.
5. Review the setup report, configuration and files, resolve unfinished work, save, and select Publish.
6. When “Environment published” appears, select Start a new task and describe what you want.

OpenAI says Codex identifies runtimes, package versions, tools and services from your repositories, so you don’t need to write an installation script yourself. It can record the tested setup in an install script and a start skill, which starts services and checks they are ready.

Network access and secrets

Your organisation’s Agent Security settings define the network restrictions for an environment, and the default access shown in OpenAI’s example is package managers only, with further domains added by you. Two kinds of credential are supported:

For network secrets, OpenAI says substitution works for HTTPS on port 443 and doesn’t put the raw credential in a local process or file. Saving an environment-owned network secret also allows access to its domains. A Personal vault lets you keep your own values, and sharing an environment shares the requirements, not your personal credentials.

Keeping work safe

• Commit important work. A new task doesn’t recover another task’s uncommitted changes. Saved virtual-machine state is recoverable for up to 7 days by default after the last start of a turn or resume, but OpenAI says that’s not a conversation-history period and isn’t a substitute for source control.
• Updating an environment. Open Settings, Codex Cloud, Environments, choose Edit, describe the change, let Codex prepare and test it, then save and Republish. New tasks use the update, and existing tasks keep their own state.
• Sharing in Enterprise. Sharing an environment makes its prepared setup available to authorised members but doesn’t give them access to another person’s tasks, and prepared files and environment-owned credentials can give teammates access to shared resources, so review them first.
• Health data. Codex in the cloud isn’t covered by the OpenAI BAA, so don’t use it for protected health information.

What else OpenAI announced for Codex at DevDay

• A refreshed command-line tool that lets you start and steer tasks with your voice, with a new /agents view to delegate and track several tasks, plus better prompt editing, session resume and worktree support. It’s available to all plans.
• Code review in the ChatGPT desktop app, with summaries, diffs and questions to Codex before you give feedback on GitHub pull requests or GitLab merge requests, and automatic first-pass reviews in the cloud.
• Codex Security Cloud, which scans GitHub repositories on demand or on a schedule, checks new commits, removes duplicate findings and prepares fixes, for Pro, Business, Enterprise and Edu users. It includes access to models offered through Daybreak Blue.
• GPT-6.1 Sol and Ultrafast, with Ultrafast offering up to 8x faster generation in Codex on Pro 500 and Enterprise.

OpenAI says the older Codex Cloud (Legacy) still supports Code Review, Security Review and the Linear and GitHub integrations during the transition, and that it plans to deprecate that experience. Environments from it can be migrated when you choose.

Codex from your phone and dots

Codex first came to the ChatGPT mobile app in preview on May 14, 2026, on all plans. OpenAI said more than 4 million people were using Codex every week then. It also lets a ChatGPT dot create tasks in Codex cloud environments, as long as you create the environment in Codex first.

How it compares

Cloud coding agents are now a standard feature of the category. For how Codex sits against Claude Code, Cursor, GitHub Copilot and others, see the best AI coding assistants and how to use Claude Code. For the rest of the announcement, read OpenAI DevDay 2026: everything announced.