Agent

The Agent step runs a pre-configured AI Agent from your workspace inside an automation flow. Unlike single-prompt steps, it can use configured tools to complete multi-step tasks.

💡

You need to enable AI to use this feature.

Before you start

Create and test an Agent first, then reference it in your automation:

• Agent building 101
• Agents Config

Behavior

Automation agents run in a goal-oriented mode. Unlike interactive chat agents, they are optimized to focus on completing the specific task defined in the prompt rather than maintaining a conversational tone.

Permissions

Agents executed via an automation step can run using either the Requester's identity or an Admin identity.

• Requester (Default): The agent uses the identity and permissions of the user who triggered the automation. If the automation was not triggered by an authenticated user (e.g. a cron trigger), the step will fail if configured to run as the requester.
• Admin: The agent runs with full administrative permissions. This allows it to perform system tasks and access data without being constrained by the triggering user's role.

If the agent triggers a tool that requires human approval (escalation), the automation will pause. Once the action is approved by a human, the agent will resume execution using the permissions of the original automation requester.

Example use cases

• Looking up workspace data with tools, then returning an answer.
• Running an Agent to decide and execute follow-up actions in a workflow.
• Producing structured JSON output for downstream automation steps.

Inputs

• Agent (required) - The Agent configuration to run.
• Prompt (required) - The task or question sent to the Agent.
• Run as (required) - Choose which identity the agent uses to access data. Options are Requester (default) or Admin.
• Use structured output (optional) - Enables schema-constrained output.
• Output Schema (optional) - JSON schema used when structured output is enabled.

Output

• success - Whether the Agent completed successfully.
• response - The Agent's response text.
• output - Structured object output when an output schema is used.