Automations

Create an automation that runs on its own

An automation asks an agent to repeat a defined piece of work at a scheduled time or on a one-time trigger. Use one when you already know the task, the expected result, and what the agent should do when it cannot complete the work.

What an automation is good for

Automations work well for recurring checks, reminders, reports, and other repeatable workflows—for example, “Every weekday morning, review new support conversations and tell me about issues that could affect more than five customers.”

Start with a one-off task when the request is new or still changing. Run it manually first so you can confirm the result, connected apps, and approval boundary before putting it on a schedule.

Before you begin

Make sure you know:

  • The owner: which agent should do the work.
  • The request: what information it may use, what it must return, and when it should stop or ask you for help.
  • The trigger: the schedule or one-time time at which it should run.
  • The access: whether it needs connected apps at all, and whether it only needs to read or must create or change something. The level you choose applies to every app connected to your Otto.
  • The controls: whether an action needs your approval and what usage or spending limit is appropriate.
  • The destination: where you want the result and whether you want a notification.

Connect required apps and review their access before saving the automation. An automation does not automatically inherit permission to send messages, publish, purchase, or change records. See Set automation permissions and limits.

Create the automation

  1. Open Automations.
  2. Choose Create automation.
  3. Select the agent that should own the work.
  4. Enter a clear task instruction. Include the source to check, the result to produce, and what the agent should do if something needs your decision.
  5. Choose a schedule or a one-time run, then set its time or recurrence.
  6. Choose the automation's session setting. Use the agent's Main session when the work should build on that conversation. Use a separate Automation conversation when the work is self-contained and should stay out of your active conversation.
  7. Review the connected-app access, approval settings, and limits.
  8. Choose where results should appear and whether Otto should notify you.
  9. Select Save.

If the form offers a choice of execution context, keep Branched for most scheduled work: it gives the run the automation instructions and a short recent conversation tail. Use Inline when the run needs the full current conversation. Use Clean only when the task is self-contained and its instructions and saved context are enough.

Write an instruction the agent can finish

State the input, output, and boundary in the same request. For example:

Every weekday at 9:00 AM, review new support conversations from the last business day. List only issues that could affect more than five customers, include links and the reason each issue matters, and notify me if one needs a decision. Do not reply to customers or change a ticket.

Avoid requests such as “keep an eye on support” without saying what counts as a meaningful change. Tell the agent whether it should report an unchanged result, where it should leave routine output, and what it must not send or modify.

Test and adjust the first run

After saving, open the automation and choose Run now when you want to test it immediately. Check the result, destination, and activity entry. Confirm that the agent used only the expected apps and that an action requiring approval stopped for your review rather than happening unexpectedly.

Change one thing at a time—such as the wording, delivery destination, or access— then run it again. If a run fails, inspect its history before creating a second automation. Read and recover failed runs explains the common fixes.

What happens next

The automation runs according to its saved trigger while it remains enabled. Results appear in the destination you selected. A notification is useful for an exception or a decision, but routine output is usually easier to review in the agent's Main session or the separate automation conversation.

You can reopen the automation to review its schedule, owner, delivery, access, approvals, and limits. Pause it when the work should stop temporarily; edit it when the workflow changes. Pausing prevents new scheduled runs but does not cancel a run that is already in progress.

Common questions

The automation needs to send or change something. Why did it stop?

That action may need a higher connected-app access level or your approval. Review the exact action first, then grant only the access and approval needed for this workflow. Read-only access is enough for monitoring and summaries; it cannot send, publish, purchase, or change records.

Where did the result go?

Open the automation's delivery settings and confirm its result destination and notification choice. Run it again after changing the setting so you can verify the new destination.

Can I stop it without deleting it?

Yes. Open the automation and choose Pause. Its settings and previous runs remain available; choose Resume when it is ready to run again.

Related guides