Cloud, worktrees, and automation

Codex supports parallel and recurring work across three execution locations:

Worktrees in the app

The desktop app can create a Worktree thread so an experiment does not collide with unfinished local changes. Codex keeps managed worktrees under $CODEX_HOME/worktrees and supports handoff between Local and Worktree threads.

Use worktrees for parallel feature attempts, risky upgrades, and automations that may create patches. Review and integrate the resulting branch as normal Git work; see Worktrees.

Cloud tasks

Cloud tasks run in isolated containers and can work in the background, create pull requests, and hand changes back to local tools.

codex cloud exec --env <environment-id> --attempts 3 \
  "Implement the agreed pagination plan and run the test suite."

codex apply <task-id>

The cloud service is current, but the codex cloud CLI surface can be marked Experimental. Verify the live feature maturity before scripting it into CI.

Cloud environments separate dependency setup from the agent phase. Network is off during agent work by default, and setup secrets are removed before that phase begins. See Cloud and environments.

Scheduled automations

The desktop app can run recurring tasks against a local checkout or an isolated worktree. Automations can invoke skills and plugins:

$pr-babysitter check the pull request, summarize new CI or review feedback, and
fix only issues with a clear reproduction.

Project automations are local automation, not cloud cron: the computer must be awake, the app must be running, and the project must remain available. Use a worktree when the automation may modify code. See Automations.

GitHub and CI

• @codex review requests a Codex review on a connected pull request.
• Another @codex PR comment can start a cloud task.
• openai/codex-action@v1 runs Codex in GitHub Actions.

Review repository access, action permissions, sandbox settings, and the exact credentials available to CI before enabling write actions. See GitHub and the GitHub Action.

SDK

Use the Codex SDK when a coding-focused Codex thread belongs inside your own automation:

npm install @openai/codex-sdk
pip install openai-codex

The TypeScript SDK is current; the Python SDK is Beta. Both should use the narrowest suitable sandbox (read_only or workspace_write). Enforce limits externally with cancellation, timeouts, and host-side usage tracking; the SDKs do not currently document built-in turn or budget limits. See Codex SDK.