跳到主要内容Skip to main content
← Frameworks

AgentScopeAgentScope

阿里巴巴(魔搭社区)Alibaba (ModelScope)
开源Open source Apache2.0Apache2.0 多智能体Multi-agent 国产China-origin 消息传递Message passing partial lifecycle: active 核验Checked 2026-10-30

AgentScope 是国产多智能体框架的主力线(Apache-2.0,阿里/魔搭系):Msg 消息传递抽象清晰, 带可视化工作台与分布式执行线。33k★、活跃维护。L5 口径:能力描述全部转述官方文档。

AgentScope is the flagship domestic multi-agent framework (Apache-2.0, Alibaba/ModelScope line): a clean Msg message-passing abstraction, with a visual Studio workbench and distributed-execution lines. 33k stars, actively maintained. Per rule L5: all capability descriptions paraphrase official docs.

适合与不适合When to use it, when not to

适合:多智能体协作应用、通义系栈、要 Studio 可视化调试。 不适合:单 agent 薄应用;要最大集成生态。

Fits: multi-agent collaboration apps, Tongyi-line stacks, Studio visual debugging. Not a fit: thin single-agent apps; needing the largest integration ecosystem.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-21

出品属地Origin
国产domestic 阿里巴巴集团 / 魔搭社区(ModelScope)线开源项目(官方口径:AgentScope 由阿里团队发起)。Alibaba Group / ModelScope-line open-source project (official positioning: initiated by an Alibaba team).
国内使用Direct access
大陆可直连direct OK agentscope.io 直连实测 200(2026-10-21);GitHub 仓大陆直连波动(同日代理可达)。agentscope.io measured 200 direct (2026-10-21); the GitHub repo's mainland reachability fluctuates (200 via proxy the same day).
备案状态CAC filing
备案未核验filing unknown:框架本体是开源库不面向公众提供服务,备案不适用;其关联模型服务(通义系)的备案状态属另一主体——本站未逐条核验,不下断言。the framework itself is an open-source library not serving the public, so filing does not apply; filing status of the associated model service (Tongyi line) belongs to another entity — not individually verified, no assertion made.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准(按批次滚动更新,本站只抄编号不解读);「大陆可直连」是本站从大陆网络直连实测的结果,不是厂商承诺。GitHub 类站点当日实测有波动,口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness

各家模型经 wrapper 接入(通义/DashScope 系为主,兼 OpenAI 兼容面与本地模型—— 以官方文档为准,未逐条清点);多智能体消息传递是它的核心抽象(Msg 对象在 agent 间流转)。

Each vendor's model connects via a wrapper (Tongyi/DashScope line first, plus the OpenAI-compatible surface and local models — following the official docs, not enumerated item by item); multi-agent message passing is its core abstraction (Msg objects flowing between agents).

运行位置Where it runs

Python 库(pip 装进你的进程);GitHub 34,871★,最近推送 2026-10-30(API 核验)。 官方叙事是「面向多智能体的生产框架」——含可视化工作台(Studio)与分布式执行线。

A Python library (pip-installed into your process); GitHub 34,871★, last push 2026-10-30 (API verified). The official narrative is "a production-oriented framework for multi-agent systems" — including a visual workbench (Studio) and a distributed-execution line.

本地文件Local files

库形态跑在你进程里;数据落点由你决定。官方 Studio 工作台是本地起的可视化面。

The library form runs inside your process; where data lands is your call. The official Studio workbench is a visualization surface started locally.

关机后的任务Tasks after shutdown

库无常驻语义;多智能体应用常驻与否取决于你的部署。分布式执行线细节未逐条核验。

A library has no resident semantics; whether a multi-agent app stays resident depends on your deployment. The distributed-execution line's details were not verified item by item.

工具与扩展Tools & extensions

工具调用与工具包在官方模块族内;MCP 支持状态未逐条核验(官方文档有相关线,细节未核)。

Tool calling and toolkits are within the official module family; MCP support status was not verified item by item (the official docs have related lines, details unverified).

上下文与记忆Context & memory

记忆模块(短期/长期记忆组件)是它的显式面——长期记忆要接外部存储;跨会话记忆给模式不给托管。

The memory modules (short-term/long-term memory components) are its explicit surface — long-term memory needs external storage attached; cross-session memory gives a pattern, not hosting.

权限与限制Permissions & limits

无账号体系(库形态);权限面=你接的各家 key 与工具能力。DashScope 系接入走阿里云账号线—— 那是模型服务的账号,不是框架的账号。

No account system (library form); the permission surface is whatever vendor keys and tool capabilities you connect. Tongyi/DashScope-line access runs on the Alibaba Cloud account line — that is the model service's account, not the framework's.

适合什么任务Best for

适合:多智能体协作应用(消息传递模型清晰)、通义系栈、要可视化调试工作台。 不适合:单 agent 薄应用(多智能体抽象是成本);要全网最大集成生态(LangChain 面更宽)。

Fits: multi-agent collaboration apps (the message-passing model is a clear abstraction), Tongyi-line stacks, wanting a visual debugging workbench. Does not fit: thin single-agent apps (the multi-agent abstraction is a cost); wanting the largest integration ecosystem (LangChain's surface is wider).

头号误解Top misconception

  • 把「多智能体框架」当「多智能体一定更好」——单任务单 agent 更简单,多 agent 是显式成本Reading 'multi-agent framework' as 'multi-agent is always better' — single-agent is simpler for single tasks; multi-agent is an explicit cost
  • 把 AgentScope 与通义千问模型混谈——框架开源 Apache2.0,模型服务是另一条账号线Blurring AgentScope with Tongyi Qianwen models — the framework is open-source Apache-2.0; the model service is a separate billing line
  • 分布式执行与 Studio 工作台的能力边界——按官方文档现状用,未核验部分别假设Assuming capabilities of the distributed-execution and Studio lines — use them per current official docs; do not assume unverified parts

未知项清单Known unknowns

  • 模型 wrapper 全目录(未逐条清点)
  • MCP 支持状态(官方文档有相关线,细节未逐条核验)
  • 分布式执行线当前完成度(未逐条核验)
  • The full model-wrapper catalog (not enumerated item by item)
  • MCP support status (the official docs have related lines; details not verified item by item)
  • The current completeness of the distributed-execution line (not verified item by item)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30,内容更新日content updated on 2026-10-21。

类型Type名称Name链接Link
repoagentscope-ai/agentscope · 仓库(Apache-2.0;34,871★,最近推送 2026-10-30,GitHub API 核验)agentscope-ai/agentscope · repository (Apache-2.0; 34,871★, last push 2026-10-30, GitHub API verified)https://github.com/agentscope-ai/agentscope
docsAgentScope · 官方文档站AgentScope · official docs sitehttps://agentscope.io
changelogagentscope-ai/agentscope · Releases(changelog)agentscope-ai/agentscope · Releases (changelog)https://github.com/agentscope-ai/agentscope/releases

实测记录Measured runs

本档未跑集成与性能实测;结论只覆盖 API 与官方文档披露的能力面。

This site has not completed measured runs.

The test protocol is at protocols/framework/_protocol.md.

相关条目Related entries

  • CrewAICrewAI多智能体对照:CrewAI 用「角色+任务」编排叙事,AgentScope 用消息传递模型。multi-agent contrast: CrewAI orchestrates with a "roles + tasks" narrative, AgentScope with a message-passing model
  • AG2AG2同为多智能体对话式框架:AG2 承 AutoGen 社区线,AgentScope 是阿里系独立设计。both are conversational multi-agent frameworks: AG2 carries the AutoGen community line, AgentScope is an independent Alibaba-line design
  • CursorCursor另一层 · AI 编程Other layer · AI coding
  • Claude Code IDE 扩展Claude Code (IDE extension)另一层 · AI 编程Other layer · AI coding

本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0). 方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.