Automations

An automation combines instructions with a trigger so work can begin without a new chat message. Use one for recurring checks, reports, or responses to events from connected services.

Define the work

Give the automation a clear name and instructions. Explain what to inspect, what result to produce, and when to ask a person for more information. Choose the project, branch, and connections needed for the work.

For example, an automation can check open support requests each weekday and prepare a summary of the ones that need attention.

Choose a trigger

Run an automation manually to try it out, use a schedule for recurring work, or select an event from a connection. Available events depend on the connector. Event filters narrow which events should start a run.

Choose the conversation context

Runs can start separate threads or reuse a persistent thread. Separate threads keep each run focused. A persistent thread carries the conversation forward across runs.

An automation can use project model settings or override them. Its connections determine which external systems it can work with, subject to the applicable permissions.

Review and adjust runs

Inspect run history and the resulting threads to see what happened. Configure notifications for success, failure, or work that needs attention. Pause an automation when it should stop starting work from its trigger.

Try the automation with representative input before relying on it. Refine its instructions when results miss the goal, and use tasks for requests that need a person.