登录

agentscope

agentscope-aiApache-2.0活跃维护

面向 LLM 应用的多智能体开发框架，支持多模态、语音与实时交互，让智能体的构建、运行和调试过程可见、可理解、可信。

Build and run agents you can see, understand and trust.

v2.0.11发布于 2026-10-27 02:54 UTC35,748star +188 / 7 天3,837fork同步于 2026-11-10 17:20 UTC
多智能体智能体框架多模态语音助手实时交互
对比 GitHub 官网agentchatbotlarge-language-modelsllmllm-agent

怎么上手

由 README 提炼，执行前请确认
原文 
uv pip install agentscope
硬件要求按 README 说明

README 未说明显存或 GPU 要求，仅要求 Python 3.11 及以上

支持 Apple Silicon
下载：源码 zip·GitHub Releases（安装包）本站不镜像安装包

项目说明

由 README 自动翻译提炼
没人工点评？写下你的经验

面向生产环境的 Agent 框架，用可组合抽象构建可观测、可信任的智能体

面向 LLM 应用的多智能体开发框架，支持多模态、语音与实时交互，让智能体的构建、运行和调试过程可见、可理解、可信。

核心能力

  • ReAct 循环支持结构化输出、实时打断恢复与批量工具调用
  • Toolkit 统一管理 Python 工具、MCP 服务器与技能
  • 内置上下文压缩、工具结果卸载与 RAG/记忆注入中间件
  • 事件总线流式推送推理、工具调用与多模态内容
  • 沙箱隔离执行，支持 Docker、E2B、K8s 等多种后端

容易踩的坑

需 Python 3.11 及以上版本；示例默认使用 DashScope 模型，需自行设置 DASHSCOPE_API_KEY；沙箱执行需额外准备 Docker 等后端环境

内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README

最近发版

5 个版本v2.0.11
全部 release
  1. v2.0.11最新正式版2026-10-27 · 2026-10-27 02:54 UTC

    What's Changed

  2. v2.0.102026-10-12 · 2026-10-12 08:15 UTC
    查看更新说明

    What's Changed

  3. v2.0.92026-09-28 · 2026-09-28 05:06 UTC
    查看更新说明

    Highlight

    • Standard Operating Procedures (Experimental): Added the agentscope.sop module for splitting a task into ordered milestones, each verified before the next one starts, with run state that can be persisted and resumed. The agent service also exposes /sop endpoints to store procedures, start runs, and review steps by hand.
    • Team Pipeline: Added a TeamPipeline class where a leader agent delegates tasks to member agents, which run concurrently and hand their results back.
    • More Realtime Voice Models: Added OpenAI Realtime, Gemini Live, and xAI Grok Voice models to the realtime voice agent.
    • Terminal UI: Introduced the agentscope.tui module, a full-screen terminal interface for chatting with agents and running realtime voice sessions via launch_realtime_ui.
    • AskUser Tool: Added AskUser, an external tool that lets agents ask users multiple-choice questions.
    • MiniMax Model and Jev-based Model Routing: Added the MiniMax chat model (M3 by default), and a ModelRouterMiddleware backed by the new agentscope.classifier module (with Jev support) that picks the chat model for each reply.
    • DingTalk Knowledge Base Tools: Agents on DingTalk channels can now search DingTalk knowledge bases.
    • Safer Shell Permission Checks: Tightened read-only command detection so destructive git, sed, and tee usages no longer skip permission confirmation.

    What's Changed

  4. v2.0.82026-09-08 · 2026-09-08 13:12 UTC
    查看更新说明

    Highlight

    • Realtime Voice Agent: Introduced a realtime voice agent with DashScope backends and local transport support.
    • A2A Protocol Support: Added A2AAgent class to the agent module, enabling agent-to-agent communication.
    • New Pipeline Module: Added a new pipeline module with a GoalPipeline class for goal-driven agent workflows.
    • Agent-Driven Context Compression: Introduced a context compression tool that allows agents to manage their own context window.
    • LLM Rerank for RAG: Added LLM-based reranking support to improve RAG retrieval quality.
    • Volcengine Ark Model: Added support for Volcengine Ark chat model.
    • MCP Enhancements: Support runtime MCP headers in Docker workspaces and improved MCP connection cleanup.
    • Web UI Polish: Chat experience improvements including session auto-naming, query caching, and better dark mode support.

    What's Changed

  5. v2.0.7.post12026-08-28 · 2026-08-28 02:42 UTC
    查看更新说明

    What's Changed

    New Contributors

    Full Changelog: https://github.com/agentscope-ai/agentscope/compare/v2.0.7...v2.0.7.post1

用户点评

登录用户的真实使用记录，可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译
    在 GitHub 打开

    中文主页 | Documentation | Roadmap

    What is AgentScope 2.0?

    AgentScope 2.0 is a production-ready, easy-to-use agent framework with essential abstractions that keep up with rising model capability.

    We design for increasingly agentic LLMs. Our approach leverages the models' reasoning and tool use abilities rather than constraining them with strict prompts and opinionated orchestrations.

    News

    • [2026-10] FEAT Experimental: Agent service supports realtime voice conversations over WebRTC, with text and voice sharing the same session history.
    • [2026-09] FEAT Experimental: SOP supported — run multi-step tasks as standard operating procedures, in both the framework and agent service. Docs | Service
    • [2026-09] FEAT: TeamPipeline supported — a leader agent delegates tasks to member agents. Docs
    • [2026-09] INTE: MiniMax chat models supported via MiniMaxChatModel (MiniMax-M3 by default).
    • [2026-09] FEAT: Model routing supported — use ModelRouterMiddleware to select an appropriate chat model for each reply.
    • [2026-09] INTE: Jev classifier model supported through the TypeSafe SDK via JevClassifierModel.
    • [2026-09] FEAT: agentscope-skill now supports AgentScope v2.
    • [2026-09] INTE: Support DashScope, OpenAI, Gemini and xAI realtime APIs in RealtimeAgent. Example | Docs
    • [2026-09] FEAT Experimental: Realtime voice agent supported. Example | Docs
    • [2026-09] FEAT: A2A protocol supported — chat with any remote A2A agent via A2AAgent. Example | Docs

    More news →

    Community

    Welcome to join our community on

    DiscordDingTalk

    Quickstart

    Installation

    AgentScope requires Python 3.11 or higher.

    From PyPI
    uv pip install agentscope
    From source
    # Pull the source code from GitHub
git clone -b main https://github.com/agentscope-ai/agentscope.git

# Install the package in editable mode
cd agentscope

uv pip install -e .

    Agent

    The SDK layer — compose an agent from a rich set of building blocks:

    Building blockWhat's inside
    ReActReasoning-acting loop with structured output, realtime interruption & resume, and batched (sequential / concurrent) tool acting
    ToolkitAgentic tool management over Python tools, MCP servers, and skills; ships with built-in coding tools (shell, file edit, search) and task/plan tools
    ModelLLM, embedding, and TTS across major providers (OpenAI, Anthropic, Gemini, DashScope, DeepSeek, Moonshot, Volcengine, xAI, Ollama)
    ContextAutomatic compaction, tool-result offload, and context injection (system prompt, RAG, memory) via built-in middleware
    Event SystemUnified event bus streaming reasoning, tool calls, and multimodal content (text, image, audio) to the frontend
    Permission & HITLFine-grained control over tools and resources, confirmation, bypass mode
    MiddlewareComposable hooks across the loop — reply, reasoning, acting, model calling, permission checking, context compression, system prompt
    MemoryAgentic memory with switchable backends (ReMe, Mem0)
    Workspace / SandboxIsolated tool & code execution — local, Docker, Apple Container, Bubblewrap, E2B, OpenSandbox, Daytona, K8s

    Start your first agent with AgentScope 2.0 in console:

    from agentscope.agent import Agent
from agentscope.console import launch_console
from agentscope.tool import Toolkit, Bash, Grep, Glob, Read, Write, Edit
from agentscope.credential import DashScopeCredential
from agentscope.model import DashScopeChatModel

import os, asyncio


async def main() -> None:
    agent = Agent(
        name="Friday",
        system_prompt="You're a helpful assistant named Friday.",
        model=DashScopeChatModel(
            credential=DashScopeCredential(
              api_key=os.environ["DASHSCOPE_API_KEY"]
            ),
            model="qwen3.6-plus",
        ),
        toolkit=Toolkit(
            tools=[
                Bash(),
                Grep(),
                Glob(),
                Read(),
                Write(),
                Edit(),
            ]
        ),
    )

    # Chat with the agent in the terminal — streamed output, tool-call
    # confirmation and Ctrl+C interruption are all handled for you
    await launch_console(agent)

asyncio.run(main())

    Agent Service — All You Need to Build Your App

    AgentScope ships a batteries-included agent service — a FastAPI backend with a pre-built Web UI (examples/web_ui) that turns your agents into a multi-tenant, multi-session application, with rich capabilities out of the box:

    CapabilityWhat you get
    ServingMulti-tenancy, multi-session isolation, FastAPI backend, pre-built Web UI
    Agent TeamLeader–worker orchestration, built-in team tools, task planning
    ChannelsConnect agents to IM platforms — Feishu (Lark), Discord, custom channels, message routing
    RAG ServiceBlob storage, index worker, multi-tenant retrieval
    MCP & Skill HubBrowse hubs (GitHub MCP Registry, ClawHub), install into your library, add to a workspace
    Resource SharingGroup- and org-level management for sharing models, MCP servers, skills, and workspaces
    PersistenceSQL & NoSQL persistence of agent state and sessions
    SchedulingScheduled tasks, agent wakeup, background task offloading

    Everything above is composable, so you can assemble your own application on top of the service with minimal glue code.

    Run the following commands to start the agent service backend and the web UI:

    git clone -b main https://github.com/agentscope-ai/agentscope.git
cd agentscope/examples/agent_service

# start the agent service backend
python main.py

    Then open another terminal to start the web UI:

    cd agentscope/examples/web_ui

# start the webui
pnpm install
pnpm dev

    Contributing

    We welcome contributions from the community! Please refer to our CONTRIBUTING.md for guidelines on how to contribute.

    License

    AgentScope is released under Apache License 2.0.

    Publications

    If you find our work helpful for your research or application, please cite our papers.

    @article{agentscope_v1,
    author  = {Dawei Gao, Zitao Li, Yuexiang Xie, Weirui Kuang, Liuyi Yao, Bingchen Qian, Zhijian Ma, Yue Cui, Haohao Luo, Shen Li, Lu Yi, Yi Yu, Shiqi He, Zhiling Luo, Wenmeng Zhou, Zhicheng Zhang, Xuguang He, Ziqian Chen, Weikai Liao, Farruh Isakulovich Kushnazarov, Yaliang Li, Bolin Ding, Jingren Zhou}
    title   = {AgentScope 1.0: A Developer-Centric Framework for Building Agentic Applications},
    journal = {CoRR},
    volume  = {abs/2508.16279},
    year    = {2025},
}

@article{agentscope,
    author  = {Dawei Gao, Zitao Li, Xuchen Pan, Weirui Kuang, Zhijian Ma, Bingchen Qian, Fei Wei, Wenhao Zhang, Yuexiang Xie, Daoyuan Chen, Liuyi Yao, Hongyi Peng, Zeyu Zhang, Lin Zhu, Chen Cheng, Hongzhu Shi, Yaliang Li, Bolin Ding, Jingren Zhou}
    title   = {AgentScope: A Flexible yet Robust Multi-Agent Platform},
    journal = {CoRR},
    volume  = {abs/2402.14034},
    year    = {2024},
}

    Contributors

    All thanks to our contributors:

    相关项目

    同分类里 star 最高的项目
    对比全部 →
    ECC
    affaan-m·JavaScript

    为 Claude Code、Codex、Cursor 等编码智能体提供技能、直觉、记忆与安全防护的性能优化层，让 AI 编程助手更稳定高效地干活。

    智能体框架+4Agent 框架
    300641+191444kv2.2.32026-11-09 04:56 UTC
    deepseek-harness
    deepseek-ai·TypeScript

    DeepSeek 开源的 Agent 编排框架，把模型、工具、记忆等所有能力都做成可插拔插件，方便按需拼装和扩展智能体。

    Agent 框架+3Agent 框架
    278086+288634kdsh-v0.2.2-rc.12026-11-04 13:06 UTC
    hermes-agent
    NousResearch·Python

    一个可随使用不断成长的 AI Agent 框架，支持接入 Claude、ChatGPT 等多种大模型，用于搭建和编排自主智能体。

    Agent 框架+3Agent 框架
    262361+108159kv0.23.02026-11-10 17:13 UTC
    firecrawl
    firecrawl·TypeScript

    把任意网页抓取并转成干净 Markdown 或结构化数据，直接喂给 LLM 和 AI Agent，省去自己写爬虫和清洗的麻烦。

    网页抓取+4Agent 框架
    200422+108311kv2.11.02026-11-10 17:12 UTC