Save scenarios that check the agent still answers well, and compare prompts and models before switching.
This website needs JavaScript. Log in to the dashboard.