AgentScope

1. Project/Product Overview

2. What does it mostly do?

AgentScope 2.0 is positioned as " a secure, efficient, flexible, and complete full-stack Agent framework ", covering the entire lifecycle from development to deployment:

Framework Architecture

! AgentScope 2.0 Architecture

• AgentScope 2.0 full stack architecture: from Agent SDK to multi-tenant services, Web UI, distributed deployment, a complete set of production-level solutions. *

3. Applicable Scenario

4. Not quite the scene

5. Core Competence List

5.1 Agent Core

5.2 Tool System

5.3 safety system (three-level protection)

5.4 Context and Memory

5.5 Middleware System

5.6 Production Services

6. Architecture/deployment/integration approach

Deployment Mode

Model Integration

-Ali Department:DashScope (Tongyi Thousand Questions Series, Deep Integration)

-OpenAI:GPT Series

-Anthropic:Claude Series

-DeepSeek:DeepSeek series

-Other: Access any LLM through Unified Model API

How to use #7.

Installation

# Python 3.11+
uv pip install agentscope
# 或 pip install agentscope

The First Agent

from agentscope.agent import Agent
from agentscope.model import DashScopeChatModel
from agentscope.credential import DashScopeCredential
from agentscope.message import UserMsg
import os

agent = Agent(
    name="Friday",
    system_prompt="你是一个有用的助手。",
    model=DashScopeChatModel(
        credential=DashScopeCredential(api_key=os.environ["DASHSCOPE_API_KEY"]),
        model="qwen3.6-plus",
    ),
)

async for evt in agent.reply_stream(UserMsg("Tony", "你好！")):
    # 处理事件流：REPLY_START、MODEL_CALL_START、TEXT_BLOCK_DELTA 等
    pass

Start Agent Service (Multi-Tenant Web UI)

git clone https://github.com/agentscope-ai/agentscope.git
cd agentscope/examples/agent_service
python main.py  # 启动后端

cd ../web_ui
pnpm install && pnpm dev  # 另一个终端启动前端

Key Screenshot

Agent Team - Leader creates a worker and coordinates execution:

! Agent Team

• Agent Team mode: Leader Agent automatically manages Worker pools, assigns and coordinates tasks. *

Task Planning-Agent automatically breaks down complex tasks and tracks progress:

! Mission Planning

• Task planning: The Agent splits complex tasks into traceable planning steps and updates them while executing them. *

8. What can I say before sales

8.1 a sentence positioning

" AgentScope is a production-level Agent framework produced by Ali-secure, full-stack, domestic controllable. "

8.2 customer pain points → solutions

8.3 Differentiated Selling Points

vs LlamaIndex / LangChain (Internationalization Framework):

-Produced by Ali, domestic controllable, Xinchuang friendly

-The security system (three-level protection sandbox) is not available in the competition.

-Multi-tenant multi-session service out of the box (competing products need to be built by themselves)

-DashScope/Tongyi Thousand Questions Deep Integration to Lower the Threshold for Domestic Customers

vs Agno:

-AgentScope security protection is more complete (level 3 vs single layer RBAC)

-AgentScope have web UI and complete front-end and back-end separation architecture

-Agno is more mature in versioning components and component management

-Different technical routes: AgentScope uses event-driven middleware, Agno uses primitive capability to mount

vs Dify / Coze (Low Code Platform):

-AgentScope is a code-level framework, more flexibility

-Dify/Coze is drag-and-drop low code for non-developers

-Different target users can be used in complementary ways

8.4 Customer Value Story Line

1. Cut in:"What are you most worried about when you are working on AI Agent application? Security? Multi-tenancy? Or is it the requirement of localization?"
2. Resonance :"Although the overseas framework is mature, it has shortcomings in security, compliance, Chinese support and Aliyun ecological integration."
3. Demo : Run through the first Agent in 10 minutes → Add sandbox → Start Agent Service → Open Web UI
4. Advanced : Single Agent → Code Tool Set → Agent Team → Distributed Deployment
5. Rest assured : Ali Open Source, Apache-2.0, Two Academic Papers, 27K Stars, Active Community

9. Frequently Asked Customer Questions

10. PoC Recommendations

Recommended PoC Direction: Enterprise Code Assistant Agent Platform

Validation Metrics:

-Sandbox isolation validity (Agent cannot access beyond authority)

-Multi-tenant data isolation (tenant A cannot see tenant B's data)

-Human-in-the-Loop Approval Process Availability

-Agent Team task assignment accuracy

-Service stability under concurrent sessions

11. Risks and Considerations

12. My Pre-Sales Judgment

• Recommendation: strongly recommended * (especially suitable for domestic enterprise customers with localization/safety requirements)

Reason:

1. Domestic Benchmark : Alibaba, 27K Stars, two academic papers-is the most technical background in the domestic Agent framework
2. Safety First : Three-level security protection sandbox isolation, the most complete security system in the competition-for financial, government and other customers is the killer
3. Production Ready :v2.0 is fully committed to multi-tenant, distributed, Web UI and other production-level features.
4. Full Stack Coverage : The SDK back-end front-end is distributed, and a set of frameworks can handle the full link.
5. Eco-friendly:DashScope deep integration, MCP support, flexible tool system expansion

Recommended Customer Persona:

-Domestic enterprises, with demand for localization/trust creation

-Security-sensitive industries (finance, government, healthcare, energy)

-Need a multi-tenant Agent platform (SaaS entrepreneur)

-Use Aliyun Ecology (DashScope, Tongyi Thousand Questions)

-The team has Python foundation and takes the code-level development route.

Not recommended situations:

-Pure overseas deployment (LlamaIndex/LangChain recommended)

-Simple scenes do not require a security sandbox (complexity is not worth it)

-Low-code platform required for non-technical staff (Dify/Coze recommended)

-v1.0 old users and unable to invest in migration costs

13. REFERENCE

-GitHub repository: https://github.com/agentscope-ai/agentscope

-Official Document: https://docs.agentscope.io

-README (Chinese):https://github.com/agentscope-ai/agentscope/blob/main/README_zh.md

-Roadmap:https://github.com/orgs/agentscope-ai/projects/2/views/1

-Academic Papers v1.0:https://arxiv.org/abs/2508.16279

-Academic paper original: https://arxiv.org/abs/2402.14034

-Discord Community: https://discord.gg/eYMpfnkG8h

-PyPI:https://pypi.org/project/agentscope/

• Analysis Date: 2026-06-02 | Data Aging: GitHub Information Pull in Real Time, Product Functions Based on Official Document v2.0.3 *