A framework where multiple AI agents interact in a controlled environment to test specific behaviors or reasoning capabilities.