Skip to main content
An agent is one runnable harness-and-model combination. Select an agent ID when starting an eval run instead of configuring the harness and model separately.

Command

List agents

Example response:
Use the id value as the --agent-id when creating eval runs: