Codex Cloud adds reusable environments for coding tasks
Codex Cloud adds reusable project environments. OpenAI documents machine sizes, a seven day recovery window and limits on browser use and code hosts.
OpenAI’s September 29 DevDay update gives Codex a reusable cloud setup for development work. Developers can prepare a project once, then start separate tasks that continue while their computer is asleep.
The practical change is where preparation happens. Dependencies, project tools and approved access can become part of a saved environment. That could reduce repeated setup, although a team still needs to check whether the configured machine can run its actual development workflow.
A prepared project becomes the starting point
The Codex Cloud overview describes a setup process built around GitHub repositories. Codex examines the selected projects, installs dependencies and tools, tests the workflow and asks for information or access it cannot supply itself. The developer reviews that preparation before publishing the environment.
Each task then receives its own workspace. Developers can examine changed files and test results, ask for revisions, and commit or open a pull request when ready. Tasks can be reviewed and continued across the web, mobile and desktop app.
This makes the setup report an important handoff point. Before delegating a substantial change, a team can ask whether the tests ran, whether services actually started and whether the selected tool versions match the project. A successful dependency installation alone would be weak evidence that the full workflow is ready.
Machine size and saved state have limits
OpenAI’s environment documentation gives different default resources by plan.
| Plan | Virtual CPUs | Memory | Disk |
|---|---|---|---|
| Plus and Edu Plus | 2 | 8 GiB | 8 GiB |
| Pro, Business and Enterprise | 4 | 16 GiB | 32 GiB |
| Edu and Edu Pro | 4 | 16 GiB | 32 GiB |
Saved machine state is recoverable for up to seven days after a task was last resumed or a turn was started. Republished environments apply to new tasks. Existing tasks retain their own state.
Current exclusions include browser and computer use, GitLab, and GitHub Enterprise Server hosted by the customer. Repository skills are available, while personal skills on a local computer are not synchronized. Enterprise customers can request larger machines.
These details narrow the first useful trial. A repository with a repeatable command line test suite is a different fit from a project that depends on manual browser interaction or an unsupported code host. Commit completed work rather than treating a saved task as the only copy.
Administrative controls still matter
The Agent Security documentation describes global policies with supported execution overrides for Codex Cloud. Requirements constrain what users can change, while defaults establish starting settings within those limits. Approval and web search controls remain in the global policy.
For a rollout, the operational question is who owns the reusable setup. Someone should review changes to its dependencies, service access and test commands. A shared starting point is useful only while it reflects how the project is meant to work.
Our coverage of GPT 6.1 Sol looks at the model economics behind coding agents. Codex Cloud adds a separate consideration, the reliability of the environment where those agents execute. Teams can evaluate both by tracking completed tasks, failed setup attempts and the review effort needed before code is ready to merge.
