Skip to main content

Routines

Routines run an agent for you on a schedule. You choose how often it runs, which tools it may use each time, and how deep it should think—then you check the history when you want to see what it produced.

You create and edit routines from the same chat workspace you already use, so it should feel familiar if you chat with agents today.

📝 Note: Which schedules, tools, and plans are available depend on your organization and subscription. If you do not see routines, ask your organization admin.

What you can do​

  • Pick a schedule — Daily, weekly, monthly, or a custom pattern (for example every Monday morning or the first day of the month).
  • Limit tools per run — Give the routine only the tools it needs so each run stays predictable and lower risk than turning on every tool at once.
  • Adjust reasoning effort — Trade speed and cost against depth for automated runs versus when you are chatting live.
  • Review past runs — Open earlier runs to read outputs, timestamps, and errors when something failed.
  • Stay in chat — Set up and tweak routines next to the threads where you already work.

Common patterns​

  • Daily summaries — Inboxes, queues, or dashboards the agent can already reach with tools.
  • Weekly reports — Pull numbers or text from documents and connectors you trust.
  • Monthly checks — Same instructions you might type once in chat, repeated on a calendar.

How this fits with agents​

You still shape the agent in Creating agents, then refine models and data in Configuring model and data sources and tools in Configure extended capabilities. A routine reuses that agent and adds when it runs and how tightly each run is scoped.

⌘J