Codex Cloud lets you give a coding task its own cloud workspace, then follow the work from desktop, web or mobile. The task can continue while your computer is asleep. At DevDay 2026, OpenAI highlighted reusable environments that hold the project setup for future tasks. Official Codex Cloud overview, DevDay feature guide
The practical appeal is easy to picture: prepare a repository at your desk, ask for a small improvement, and return to the same task from your phone. The quality of the outcome still depends on the brief, the environment and the checks you review.
Last checked September 29, 2026. This is a walkthrough of OpenAI’s documented workflow, with an illustrative task brief. We have not run a separate performance benchmark or measured completion times.
Codex Cloud and Codex Remote run in different places
Kingy.ai previously covered Codex in the ChatGPT mobile app and Codex Remote’s general availability. Today’s reusable-cloud workflow builds on that earlier portability story.
On a small screen, scroll this table sideways.
| Question | Codex Cloud | Codex Remote |
|---|---|---|
| Where does the task run? | Its cloud workspace | Your connected computer |
| What supplies the project setup? | A published cloud environment | The project and tools on that computer |
| Can the laptop sleep? | Cloud work can continue | Keep the connected computer awake and online |
| How do you continue? | Reopen the same cloud task | Reopen the task on the connected computer |
Sources: Cloud overview, Remote guide.
Choose Cloud when the repository and required tools can run in a cloud environment. Choose Remote when the task depends on the setup available on your computer. Decide where the work should execute before starting the task; having a phone interface does not identify the execution environment.
Prepare a cloud environment
Start in ChatGPT on the web or in the desktop app. In a new task, choose Work in > Cloud, open the environment selector, and create an environment if you do not already have one.
Select the GitHub repositories, connect GitHub if requested, and let Codex inspect the project and prepare its dependencies. Supply missing information or access when prompted. Review the setup report and test results, save the configuration, and publish it. Once the environment is ready, start a coding task from it. Official setup walkthrough
For a first attempt, choose a repository with a straightforward install command and an existing test suite. A small, understandable project makes setup failures easier to diagnose. Ask for the runtime and package-manager versions, the installation command, and the checks Codex ran so you can compare them with your normal workflow.
An environment is the reusable setup; each task receives separate working files. Republishing an environment prepares the setup for new tasks, while existing tasks retain their own state. Mobile users should create and publish the environment on web or desktop first, then choose an available environment from Codex on mobile. Environment lifecycle and mobile instructions
Give it a task you can verify
Begin with a change small enough to inspect in one sitting. A focused bug fix or interface adjustment offers a clearer result than an instruction to improve the entire application.
Here is an illustrative brief for a project with a search component:
In the product-search component, make the empty-results message include the search term. Preserve the existing styling. Find and run the relevant tests, and add a regression test if the behavior is not already covered. Show the changed files, test results and any remaining issue. Prepare the change for review without deploying it.
Adapt the component name and expected behavior to the actual repository. The brief supplies a concrete outcome, a scope boundary and evidence to review. If the agent discovers that the request requires a wider change, have it explain that dependency before extending the task.
Keep follow-up instructions in the same task. For example, ask it to correct an escaping problem or adjust the text without restarting the work as a separate assignment. OpenAI documents that reopening a task continues its saved work across devices. Task continuation
Continue from your phone and review the result
After preparing the environment, open Codex in the mobile app and return to the cloud task. You can review progress and supply the next instruction through the supported mobile workflow. Cross-device cloud workflow
Review the code change as you would any other contribution. Check that the behavior matches the request, that unrelated files were not changed, and that the reported tests exercised the affected behavior. If a test did not run, ask for the failure and the command used rather than accepting a general statement that the work is complete.
For the sample search change, inspect what happens with punctuation, special characters and a very long search term. These are suggested review cases. They are not results from a Kingy.ai test run.
When the change is ready, review the diff and checks before committing or opening a pull request. Saved task state is temporary: OpenAI documents recoverability for up to seven days after the last turn or resume. Preserve important work in source control. Saved state and retention
Usage, access and current limitations
ChatGPT Work and Codex share usage. Cloud tasks can consume more allowance than short local messages, and consumption varies with the model and workload. Check the usage dashboard rather than estimating a fixed number of tasks from the subscription price. Current pricing and usage documentation
If you are comparing models for these tasks, our GPT-6.1 Sol guide explains the published model results, while our Pro 500 comparison covers plan and speed options. Neither establishes a universal cost or completion time for your project.
The new cloud environments currently have limits worth checking before setup: browser and computer use are unsupported, as are GitLab and self-hosted GitHub Enterprise Server. Repository skills are available, but personal skills on a local computer are not automatically synced. Network access and credentials also need configuration for private services. Cloud environment limitations
If a package download fails, capture the command and destination hostname. Ask whether the failure came from a missing dependency, blocked destination or authentication problem. Those require different fixes. Verify that the repaired setup can run the project’s normal checks before using it for another assignment.
For your first cloud task, retain the brief, changed files and check output together. That gives you a useful basis for deciding which project work to move into a reusable environment next.