One round trip inside an agent loop where the model produces tool calls, the runtime executes them, and results feed back as new messages.
↑ ContentsDefinition
A turn ends when tool results are returned to the model for the next inference. A simple question might take one turn; a complex task like refactoring a module can chain dozens. The system prompt, conversation history, and prior tool results are all available at the start of every turn. The conceptual phases people describe (gathering, acting, observing) are emergent properties of which tools appear in the response, not structural modes the runtime switches between.
↑ Contents