How to schedule Codex Cloud tasks: 4 ways that work today (October 2026)

Codex Cloud cannot schedule its own tasks yet. OpenAI's docs describe starting Codex Cloud tasks by hand, and a feature request for cron-style scheduling was opened on 22 October 2026. Until it ships, use ChatGPT scheduled tasks, a cron job that runs codex cloud exec, codex exec in GitHub Actions, or a hosted workspace that runs Codex on a schedule.

Can Codex Cloud run tasks on a schedule?

Not on its own, as of 25 October 2026. Codex Cloud runs each task in its own workspace, built from a published environment, and the task "can keep working while your computer is asleep". OpenAI's Codex Cloud docs describe starting tasks from ChatGPT on the web, mobile, the desktop app or the CLI. They describe no schedule.

The gap is on record. openai/codex #56123, opened 22 October 2026, asks for "first-class scheduling (cron/hourly) for Codex Cloud tasks launched from a published cloud environment, so tasks can run unattended while the user's computer is off". It asks for subscription usage, "not an API-key billed worker", and run history. It was still open with no replies when this page was written.

What are the four ways to schedule Codex today?

Way	Runs where	Sign-in	Computer can be off	Best for
ChatGPT scheduled tasks (web)	OpenAI's cloud	ChatGPT plan	Yes	Recurring work on connected tools: Gmail, Slack, GitHub
ChatGPT scheduled tasks (desktop)	Your computer	ChatGPT plan	No	Recurring work on a local project or worktree
Cron plus codex cloud exec	Codex Cloud, started from a machine that stays on	ChatGPT sign-in on that machine	Only the cloud task; the cron machine must stay on	True Codex Cloud runs on your published environment
codex exec in GitHub Actions	A GitHub runner	API key (OpenAI's recommendation)	Yes	Repo jobs: CI fixes, reviews, nightly checks
A hosted workspace (Kavela)	Kavela's cloud	Your Codex plan	Yes	Recurring work across your tools, checked before it reaches you

How do you schedule Codex with ChatGPT scheduled tasks?

Scheduled tasks are OpenAI's own way to repeat work, in ChatGPT and Codex.

1. Test the prompt in a normal chat first. OpenAI's docs say to "test its prompt in a regular web chat" before you schedule it.
2. Ask ChatGPT to create the task. Describe the work, when it runs, and whether each run starts a new chat or returns to the current one.
3. Pick the surface:
   • On the web, tasks can use uploads, connected tools, skills and plugins, but "can't work directly in a folder on your computer".
   • In the desktop app, tasks can run in a local project or an isolated worktree. You need to "keep the computer on and the app running".
4. For an unusual cadence, edit the task's recurrence rule, for example RRULE:FREQ=MONTHLY;BYMONTHDAY=1;BYHOUR=9;BYMINUTE=0.
5. To start from an event instead, use a Gmail, Slack or GitHub trigger on eligible plans. OpenAI's docs note one task "can't combine event triggers with a time-based schedule".
6. Review runs in Scheduled, which works as an inbox for runs with findings.

This is the simplest route. Note that OpenAI's docs do not say these tasks start a Codex Cloud task in a published environment. That is exactly what #56123 asks for.

How do you run Codex Cloud tasks from cron?

The Codex CLI can submit a Codex Cloud task directly. In the open-source CLI (codex-rs/cloud-tasks/src/cli.rs), codex cloud exec takes a prompt and a required --env environment ID, plus optional --branch and --attempts (1 to 4).

1. On a machine that stays on, install the Codex CLI and run codex login to sign in with ChatGPT. OpenAI says Codex Cloud requires a ChatGPT sign-in (auth docs).
2. Find your environment ID: run codex cloud and browse.
3. Add a cron line:

# Every weekday at 09:00 server time, start a Codex Cloud task on the main branch
0 9 * * 1-5  codex cloud exec --env <ENV_ID> --branch main "Triage issues opened since yesterday, label them, and draft a fix for the top one" >> ~/codex-cron.txt 2>&1

4. Check results with codex cloud list --json, or in ChatGPT.

Know the trade-offs. The cron machine must stay on, even though the task runs in the cloud. The ChatGPT sign-in on that machine must stay valid. OpenAI documents keeping a ChatGPT-managed login alive on a runner as "an advanced workflow for enterprise and other trusted private automation" (CI/CD auth). And nothing reads the result unless you do.

How do you run Codex on a schedule in GitHub Actions?

For repo work, OpenAI points to the Codex GitHub Action, openai/codex-action. It installs Codex and runs it behind a proxy so your API key is not exposed to other steps.

1. Add an OpenAI API key as a repository secret.
2. Add a workflow with a schedule: trigger (GitHub cron syntax) that runs the action with your prompt.
3. Have the job open a pull request or comment, so the result lands where someone reads it.

This runs Codex on a GitHub runner, not in Codex Cloud, and bills API rates. OpenAI says "API keys are the right default for automation" and warns not to set OPENAI_API_KEY or CODEX_API_KEY as a job-level variable in workflows that run repository code.

How do you schedule Codex on Kavela?

Kavela runs the Codex CLI in a cloud computer signed in to your Codex plan, and sets schedules from plain words in your own time zone (schedule a routine). Each run is checked against the workflow's checklist, and a run that falls short is reported.

One limit to know: Kavela routines cannot run on a day of the month. "The 1st of every month" is not supported today.

Connect your Codex plan once, then paste this into Kav's chat box:

Every weekday at 09:00 <your timezone>, use Codex on my connected Codex plan to work in <GitHub repo>:
triage issues opened since the last run, label each one, and draft a fix for the most urgent as a pull request.
A run is done when every new issue has a label and the draft pull request links its issue.
Send me a short summary in approvals. Never merge; I review every pull request.
Connect GitHub when you need it.
Run it once now so I can check the first result, then set the routine.

You still sign in to GitHub and your Codex plan, and you approve the first results. If your plan reaches its limit, the task waits for the reset (Kavela docs).

A note on rules: Kavela is not one of OpenAI's Sign in with ChatGPT partners. OpenAI has not said whether a third-party service running the Codex CLI on your ChatGPT sign-in is allowed. Read what Anthropic and OpenAI say about using a subscription for automation before you choose.

What breaks when Codex runs on a schedule?

• The sign-in lapses. A cron or runner login that expires stops every run until someone signs in again.
• The computer sleeps. Desktop scheduled tasks and the cron machine need power and a network.
• Runs stack up. In ChatGPT, matching events that arrive close together may be combined into one run.
• Nobody reads the result. A green run is not a correct one. Ask the run to prove its result: a passing test, a pull request, a file in a known place.

Frequently asked questions

Does OpenAI plan to add scheduling to Codex Cloud?

OpenAI has not announced it. The public request is openai/codex #56123, opened 22 October 2026.

Can ChatGPT scheduled tasks run while my computer is off?

Web tasks can. Desktop tasks that work in a local project need the computer on and the app running.

Does codex cloud exec use my ChatGPT plan?

Codex Cloud requires a ChatGPT sign-in, and the CLI's cloud commands use the same credentials as the main CLI. So the task runs on your plan's usage.

Should I use an API key or my ChatGPT plan for scheduled Codex jobs?

OpenAI recommends an API key for automation. Your ChatGPT sign-in is documented for Codex Cloud, scheduled tasks and trusted private runners.

How do I run Claude Code on a schedule instead?

Claude Code has cloud scheduled tasks on Pro and Max; see Claude calendar: how to run Claude Code on a schedule. To compare both agents, see run Claude Code or Codex in the cloud on your own subscription.

Stuck on setup? Ask in the Kavela Discord.