What Routines Are — and How They Differ from /loop and Scheduled Agents
Routines are saved Claude Code configurations that execute on Anthropic's cloud infrastructure, not your machine. Understanding when to use Routines versus /loop versus local scheduled agents is the foundation for building reliable automation.
Quick Reference
- →Routines run on Anthropic's cloud — no machine, no terminal, no active session required
- →Three trigger types: Scheduled, API/webhook, GitHub events
- →Triggers are composable — one Routine can respond to multiple trigger types
- →/loop requires an active terminal; Routines survive laptop closure
- →Available on Pro, Max, Team, and Enterprise plans
- →Branch safety: Routines only push to branches prefixed with `claude/` by default
- →Configure Routines via Desktop app UI or the /schedule CLI command
What Routines Actually Are
A Routine is a saved Claude Code configuration — prompt, repository access, and connectors — that runs on Anthropic's cloud infrastructure on a schedule or in response to external events. The key word is 'cloud': unlike every other Claude Code execution model you've encountered so far, Routines have zero dependency on your machine being on, your terminal being open, or your session being active.
When a Routine fires, Anthropic's infrastructure provisions a fresh Claude Code environment, clones your repository, loads your connectors, and executes your prompt. The entire run happens remotely. You get a full execution transcript when it's done.
This changes the operational contract entirely. A Routine you set up at 9am to run nightly dependency audits will still run when you're on a flight, your laptop is in sleep mode, or you've switched machines. The execution environment is consistent and isolated across every run.
- ▸Routine = prompt + repository + connectors, saved as a reusable configuration
- ▸Execution happens on Anthropic's infrastructure, not your local environment
- ▸Each run gets a fresh, isolated environment — no state bleeds between runs
- ▸Full execution transcript available after every run, including failures
- ▸Configured once via Desktop app UI or /schedule CLI command