Codex Cloud & Mobile

Codex Cloud lets you delegate tasks to remote sandboxes hosted by OpenAI. Instead of running Codex locally against your filesystem, you push a task to the cloud, and it executes in an isolated environment with full compute resources. When it finishes, you review the results — diffs, test output, logs — and pull the changes locally if satisfied.

This is transformative for two use cases: long-running tasks that would tie up your local machine, and mobile workflows where you want to kick off work from your phone.

# Delegate a task to Codex Cloud
codex cloud "set up the CI pipeline with GitHub Actions, \
  including lint, test, build, and deploy stages"

# Execute a one-shot cloud task
codex cloud exec "run the full test suite and report any failures"

# Review a completed cloud task
codex cloud status
codex cloud diff <task-id>

Codex Cloud supports best-of-N attempts — it can run the same task multiple times with different approaches and let you pick the best result. This is useful for open-ended tasks where there are multiple valid solutions (UI design, architecture choices, optimization strategies).

You can also tag @codex in GitHub pull request comments to trigger cloud-based tasks directly from your PR workflow. This creates a seamless bridge between code review and AI-assisted development — spot an issue in a PR, tag Codex, and let it propose a fix.

Cloud execution is a significant architectural difference from Claude Code, which runs exclusively locally. Claude Code's model is "the agent runs where you run" — your machine, your filesystem, your sandbox. Codex offers both local and cloud execution, giving you flexibility to choose based on the task. Both approaches have tradeoffs: local execution gives you more control and faster iteration; cloud execution frees your machine and enables async workflows.