Skip to main content
In AgentScope, an agent is an independent entity that can receive outside information, process it, and take actions to achieve specific goals. To achieve this, an agent is abstracted into two main methods: During the reply process, the agent reasons based on the current state and takes actions accordingly — this is the “ReAct” (Reasoning + Acting) paradigm.
The reasoning-acting loop in replying
The observe method is useful for injecting context into the agent’s memory without triggering a reply, such as updating environment state or injecting background knowledge.