A standardized framework for agents that alternates between reasoning (thinking through problems) and acting (calling tools), used to isolate model capabilities from agent design choices.