Triggers turn an agent from something you chat with into an automation that works in the background.
A trigger starts an agent automatically. Add one on the agent's Triggers tab with Add trigger.
Scheduled triggers run on a timetable ("every weekday at 8:00") or once at a set time. You can describe the schedule in plain language. Each run sends your written prompt to the agent, as if you had typed it.
App triggers fire when something happens in another service: a new email in a Gmail label, a new Slack channel message, a new Outlook email, a new Zendesk ticket, a Salesforce or HubSpot change, and more. Some react within seconds; others check for new events about every minute (called polling).
For app triggers you write a prompt template. Event details appear as variables (shown as badges), such as Sender, Subject and Email Body for Gmail. When the trigger fires, each badge is replaced with the real value. Example: "Classify this email from [Sender] with subject [Subject] and draft a reply." Alternatively, Pass Raw Data sends the whole event to the agent and lets it pick what matters.
Things to know:
- A trigger runs with the credentials of the person who created it.
- You can choose a model per trigger, often a cheaper one for routine work.
- If the previous run is still going, the new one is skipped.
- After 3 failed runs in a row a trigger switches itself off and notifies the owner.
- To pause an agent completely, deactivate all its triggers.
Remember
- Scheduled, one-time and app triggers start agents automatically
- Prompt templates fill event variables into the agent's instructions
- Triggers use their creator's credentials and turn off after repeated failures
Watch out for
- Writing a trigger prompt with no clear instructions, so the agent only receives raw event text and guesses what to do.