Routines and scheduled tasks

“Run Claude on a schedule” resolves to three different features, and picking the wrong one is a common mistake because they sound identical.

	Where it runs	Survives your machine being off	Triggered by
Routines	The cloud	Yes	Time, API call or GitHub event
Desktop scheduled tasks	Your machine	No	Time
/loop	Inside a running CLI session	No	Repetition within the session

Routines

Cloud-run, so they keep going when your computer is closed, and they can be triggered by things other than the clock — an API call or a GitHub event. Create them from the web, the desktop app, or by running /schedule in the CLI.

This is the one to use for anything the business depends on: morning PR reviews, overnight CI failure analysis, weekly dependency audits, doc sync after merges.

Desktop scheduled tasks

Run on your machine, with direct access to your local files and tools. That is the reason to choose them over routines — a task that needs your local environment cannot run in the cloud.

/loop

Repeats a prompt within a live CLI session. Good for polling something while you watch; not a scheduler.

What it does not do

• A routine does not have your local environment. If the task needs your machine, it is a desktop scheduled task.
• Scheduling does not remove the permission question. An unattended agent runs with whatever permissions you gave it, without anyone there to decline a prompt.
• Cowork has its own scheduling. Do not assume routines cover it.