Evals

Scripted conversations that grade an agent’s replies and tool calls, without placing a real call.