# Codex Cloud Gets Persistent Environments — Set Up Once, Run Forever > Source: > Published: 2026-09-30 03:13:38+00:00 OpenAI’s Codex Cloud got a meaningful upgrade at DevDay 2026. The platform now supports persistent, reusable environments — you configure a project once, publish that state as a snapshot, and every future Codex Cloud task starts from it with dependencies already installed. That means less time waiting for setup scripts and more time watching the agent work. The headline feature: close your laptop mid-task and Codex keeps running. Pick it back up from your phone, browser, or another machine. For teams running frequent AI coding sessions, this changes the workflow. ## How to Set Up a Reusable Codex Cloud Environment The setup flow is straightforward, but the terminology is new, so it is worth walking through step by step. 1. **Connect a GitHub repo.** Open Codex Cloud in the ChatGPT app, desktop client, or browser. Select which repositories Codex may access — it does not get blanket access to your GitHub account. 2. **Let Codex read the codebase.** Codex analyzes the repo and identifies runtimes, package versions, and tooling. It then drafts two configurations: an*Install script* (commands to install dependencies and prepare dev files) and a*Start skill* (steps that launch services and verify they are running). 3. **Review and refine.** You can converse with Codex to adjust the generated configs, run the environment, test it, and iterate until setup is clean. 4. **Publish.** Hit Publish and that state is saved as a reusable snapshot. New tasks will start from this baseline — no reinstalling, no re-running setup. One important detail that is easy to miss: **changes made inside a task are not written back to the original environment.** Each task runs in isolation. If you update and republish the environment, existing tasks are unaffected — only new tasks pick up the change. This prevents one misbehaving task from corrupting the shared starting point. ## VM Specs: What You Get by Plan The hardware assigned to your cloud tasks depends on your ChatGPT plan. | Plan | vCPU | RAM | Disk | |---|---|---|---| | Plus / Edu Plus | 2 | 8 GiB | 8 GiB | | Pro / Business / Enterprise | 4 | 16 GiB | 32 GiB | Plus users also share a five-hour usage window across local and cloud tasks. If you are running heavy builds or large test suites, the Plus VM will feel constrained. Pro and above get the four-core machine with twice the memory. API key users get no cloud features at all — Codex Cloud environments are a ChatGPT plan feature only. ## Cross-Device Continuity A task started on your laptop continues running in the cloud after you close the lid. Task state is recoverable for up to seven days from last activity. You can reopen the same task from the web, the mobile app, or the desktop client and pick up exactly where things were — uncommitted changes included. For developers who work across multiple machines, or who need to hand off context without committing half-finished work, this is a genuine workflow improvement. A 40-minute task no longer needs an open laptop for 40 minutes. According to the [TechCrunch report on DevDay](https://techcrunch.com/2026/09/29/openai-gives-codex-reusable-cloud-environments-that-work-across-devices/), OpenAI framed the headline as: “You can finally close your laptop now.” ## What Does Not Work Yet A few constraints are worth knowing before you commit to this workflow. - **No computer use or browser use** inside cloud environments. These remain local-only features. - **GitLab support is beta.** Self-hosted GitHub Enterprise Server is not supported at all. - **No real-time diff view.** While a task runs, you see minimal output. The final diff appears when the task completes — there is no file-by-file live view. - **Local AGENTS.md and Personal Skills do not sync.** Cloud tasks do not inherit your local agent configuration. - **12-hour container cache.** After 12 hours, the container is rebuilt from the published snapshot. Anything not in the diff is gone. There are also pre-existing reliability issues with Codex Cloud that the community has flagged: cloud checkouts occasionally lose their Git remote mid-task, long-running processes can become inaccessible after being backgrounded, and safety checks can terminate tasks well into execution. These are not new to the reusable environments feature, but they are real friction for unattended cloud runs. The [OpenAI forum thread on Codex Cloud reliability](https://community.openai.com/t/codex-cloud-web-has-compounding-reliability-observability-and-workflow-problems-that-make-long-running-tasks-difficult-to-trust/1397259) is worth reading before you build critical workflows on top of it. ## Team Environments and Credentials For teams, shared environments live inside ChatGPT workspaces. Each team member runs independent tasks from the same published configuration without access to anyone else’s task state. Enterprise customers control who can edit environment settings. Individual users supply personal credentials through a “network secrets” mechanism: the agent receives a placeholder string, and OpenAI’s proxy injects the real value only for HTTPS requests to whitelisted domains. That model limits credential exposure, but it does not restrict what an agent can do once it has access. Review your domain allowlist carefully — especially for any credentials with write permissions to production systems. The [official Codex documentation](https://developers.openai.com/docs/codex) covers the network secrets configuration in detail. ## Who Should Set This Up Now If you are running one-off Codex tasks with a setup script under two minutes, this does not change much for you. But if you are working on a project with heavy dependencies, running multiple tasks per day, or coordinating cloud sessions across a team, persistent environments are worth configuring now. The infrastructure is still maturing, and the reliability issues are real — but the core capability is here and the setup is not complex. Full announcement details are at the [OpenAI DevDay 2026 recap](https://openai.com/index/devday-2026/).