AS AgentScopeAPI 手册 / API Reference
包package Guide

指南Guides

本手册除了逐类逐方法的 API 参考，还提供一组与固定版本源码对应的概念指南。它们解释框架为什么这样分层、一次调用在内部如何流动，以及 Python 与 Java 两个版本如何一一对应。

阅读顺序

  1. 核心概念 — 消息、模型、智能体、工具、状态、中间件与 RAG 的整体模型。建议先读。
  2. Python 快速开始 — 安装 agentscope==2.0.11，构建第一个 ReAct 智能体。
  3. Java 快速开始 — 使用 Maven 依赖与 ReActAgent.Builder 构建同样的智能体。
  4. 架构 — 包层次、ReAct 执行生命周期、扩展点与持久化。
  5. Python ↔ Java 对照 — 概念、类型与代码的逐项映射。

使用建议

  • 需要查某个类或方法时，用页头的搜索（Ctrl/⌘ K）或左侧的包/类型导航直接进入 API 页。
  • API 页的签名和 Javadoc/docstring 保留英文原文；中文摘要通过翻译记忆叠加。页头右上角可在 中英混合 / 纯中文 / 纯英文 三种模式之间切换。
  • 指南中的代码示例与固定版本对应，但模型供应商的密钥、模型名与区域端点需要按你的环境替换。

Besides the exhaustive per-type API reference, this manual ships with a set of concept guides pinned to the same source versions. They explain why the framework is layered the way it is, how a single call flows internally, and how the Python and Java editions map onto each other.

Reading order

  1. Core concepts — messages, models, agents, tools, state, middleware and RAG. Start here.
  2. Python quick start — install agentscope==2.0.11 and build your first ReAct agent.
  3. Java quick start — do the same with Maven dependencies and ReActAgent.Builder.
  4. Architecture — package layers, the ReAct lifecycle, extension points and persistence.
  5. Python ↔ Java mapping — concept, type and code correspondences.

How to use this manual

  • To look up a class or method, use the header search (Ctrl/⌘ K) or the package/type navigation in the left sidebar.
  • Signatures and Javadoc/docstrings stay in their original English. Chinese summaries are layered on top through translation memory. Use the segmented control in the header to switch between mixed, Chinese-only and English-only.
  • Code samples match the pinned versions. Replace API keys, model names and regional endpoints with your own environment values.