An Agentic Process is a non-conversational system. It consists of a structured sequence of steps in which one or more Agents autonomously make decisions, interact with Tools, and perform actions to achieve a defined goal —optionally involving human input at key moments.
You can use an Agentic Process when you need to:
- Chain multiple Agents.
- Define specific tasks that each Agent must perform.
- Combine human input with automation (for example, validating results before proceeding).
- Reuse any Artifact generated as input for another Task or Agentic Process.
- Ensure full traceability by recording the model invoked, variables used, and Artifact produced at each node.
To define or update an Agentic Process, you can use the Agentic Processes API or work within The Lab. If you’re working in The Lab, follow the steps detailed in How to create an Agentic Process.
Since version 2025-04.