Agent
An Agent can run standalone — agent.run(task="…") — or sit inside a Chain as one step among many. Embedding it inside a Chain is the recommended pattern whenever you want the agent's output to flow into deterministic downstream work (formatting, translation, rendering, saving to disk).
Agent
Every knob that shapes how an Agent plans, acts, and spends its budget.
Agent
Every Agent has a shared key-value store — agent.memory — that every step reads from and writes to. It's the agent's scratchpad: seed data goes in, step outputs accumulate, and the full state is included in every orchestrator prompt so the planner always has complete context.
Agent
Agent "observability" from yaitio/aichain, covering observability & control (the step boundary), logging, hooks & events, convenience bases and event fields.
Agent
An Agent is a conversation that can act. You give it a task; it is asked what to do, the world answers, the answer is appended, and the loop goes round until it replies without asking for anything more. Use it when the sequence of steps is not known upfront. For a fixed pipeline, use a Chain — it spends no model call…