An agent is a defined set of capabilities that can plan, research, and reason within a workflow. Each agent has a specific type that determines its behavior and domain of expertise — for example, a cip_orchestrator agent coordinates customer identification program reviews, while domain agents handle narrower tasks like address verification or sanctions screening.
Agents are predefined by Middesk; you don’t author your own. To use one, create a Run referencing the agent and its required context. The Run captures the agent’s plan and steps, while Actions capture moments where the agent pauses for human input.
To list the agents available to your account, see the Agents API. For an introduction to the agent model, see Work with agents.