Codex Cloud — Tasks That Outlive Your Terminal

~12 min · codex, cloud, long-running

Level 0🌱 Novice

0 XP0/70 lessons0/11 achievements

0/120 XP to next level120 XP to go0% complete

Hand a task to OpenAI's infrastructure

Codex Cloud (experimental) lets you submit a task to OpenAI's managed infrastructure — closing the laptop doesn't kill the job. Useful for: long test suites, multi-repo refactors, big migration jobs, anything you'd otherwise run on a Spot instance.

Enable via features.codex_cloud = true. Submit via codex cloud submit; check via codex cloud list / codex cloud status; pull results back to local with codex cloud pull. Sandbox + approval policy travel with the task.

Code

Submit, monitor, pull back·bash

# Enable in config
# [features]
# codex_cloud = true

# Submit a task
codex cloud submit --name nightly-bench \
  --sandbox workspace-write \
  --ask-for-approval never \
  "Run the full bench suite, write report to ./bench-$(date +%F).md"

# Monitor
codex cloud list
codex cloud status nightly-bench

# Pull results when done
codex cloud pull nightly-bench
# Files are merged into your local working copy; conflicts are flagged.

When Cloud is the right answer·text

Use Codex Cloud when:
- The task takes >30 min and you don't want to pin your laptop
- You want consistent execution environment (cloud > your laptop)
- Multiple parallel jobs (10 PRs to refactor — fan out to cloud)

Don't use it when:
- Task is short (<5 min)
- Task needs your local files outside the repo
- Task requires real-time iteration with you in the loop

External links

Codex Cloud (experimental)

Exercise

If you have access, submit a real long-running task to Codex Cloud (a slow test suite, a multi-file refactor). Walk away. Pull results back. Note where the experience was better/worse than your local + Spot+SSH alternative.