# Codex Cloud
Codex Cloud is the cloud-hosted, async surface of [[OpenAI Codex]]; the place to send long-running or non-urgent agent work that does not need to occupy your laptop. Where [[Codex CLI]] is for tight feedback loops on the local filesystem, Cloud is for fire-and-forget jobs that come back as pull requests.
It is OpenAI's answer to [[Claude Managed Agents]] and to JetBrains Air's "cloud execution mode"; the same underlying pattern of `repo + issue → agent → PR`.
## What it actually is
A managed environment service plus an agent runtime. You connect a [[GitHub]] account, point Codex at a repo, optionally describe an environment, and dispatch tasks. The agent runs server-side on OpenAI infrastructure with internet access and tooling, then opens a pull request when done. No local Docker, no laptop time spent on the run.
## Capabilities
- **Repository-based task delegation**; describe what to do at the issue or task level; Codex Cloud handles checkout, branching, commit, PR.
- **GitHub integration**; native auth flow; PR descriptions, review threads, and re-runs all happen inside GitHub.
- **Cloud environments**; configure runtime, install steps, and dependencies once per repo; Codex reuses them.
- **Internet access controls**; per-repo allow/deny; mitigates the risk of an agent exfiltrating to arbitrary endpoints.
- **Secrets management**; environment variables and tokens scoped per environment.
- **IDE / CLI integration**; the [[Codex CLI]] and the IDE extension can both promote a job to Cloud (`Cloud delegation`); progress visible from the originating surface.
- **Async execution**; jobs queue, run, and notify. The originating session does not block.
## When to choose Cloud over CLI / App
- Long-running tasks; tests over a large suite, refactors that span hundreds of files, codebase migrations.
- Hands-off work; you want to fire something, close the laptop, get a PR.
- Multi-repo or batch jobs; sweeping the same change across many repos.
- Privacy-bounded work; the work needs internet and tools that are awkward to grant locally.
CLI / App stay better for tight feedback loops, exploration, and any work that benefits from a human-in-the-loop on every turn.
## Access
Bundled with ChatGPT Plus, Pro, Business, Edu, and Enterprise. Enterprise tenants may need workspace admins to enable it before users see it.
## Trade-offs to keep in mind
- **Visibility cost**; agents you cannot watch in real time are agents that surprise you in PR review. Treat Cloud output as untrusted until reviewed; the same `unreviewed AI code anti-pattern` applies.
- **Repo coupling**; the strongest UX is GitHub-shaped. GitLab and self-hosted git remain second-class as of 2026.
- **Compute cost is hidden**; not metered separately on consumer plans, which makes it tempting to dispatch sloppy work; tighten task descriptions or you pay in PR review time instead of dollars.
## References
- Codex Cloud entry point: https://chatgpt.com/codex
- Documentation: https://developers.openai.com/codex/
- OpenAI Codex umbrella: https://openai.com/codex/
## Related
- [[OpenAI Codex]]
- [[Codex CLI]]
- [[Codex App]]
- [[Codex IDE Extension]]
- [[AI Agent Harnesses (MoC)]]
- [[Claude Managed Agents]]
- [[JetBrains Air]]
- [[AI Agent Harness]]
- [[GitHub]]
- [[Unreviewed AI code anti-pattern]]