Agent

A system where an LLM drives its own actions in a loop — choosing tools, reading results, and deciding what to do next. Not just a prompt with a system message.