Qwen-Agent — Build AI Agents on Qwen Models

Qwen-Agent Architecture & Features

Core Components

┌─────────────────────────────────────┐
│           Agent Layer               │
│  Assistant │ ReActChat │ Router     │
├─────────────────────────────────────┤
│           Tool Layer                │
│  Function Calling │ Code Interpreter│
│  Web Search │ MCP Tools │ Custom    │
├─────────────────────────────────────┤
│           LLM Layer                 │
│  Qwen-Max │ Qwen-Plus │ Local Qwen │
└─────────────────────────────────────┘

Function Calling

Native function calling with automatic schema parsing:

from qwen_agent.tools import BaseTool, register_tool

@register_tool("get_weather")
class WeatherTool(BaseTool):
    description = "Get weather for a city"
    parameters = [{"name": "city", "type": "string", "required": True}]

    def call(self, params, **kwargs):
        city = params["city"]
        return f"Weather in {city}: 25°C, sunny"

Code Interpreter

Executes Python code in an isolated Docker sandbox:

• Matplotlib/Seaborn chart generation
• Pandas data analysis
• File I/O in sandboxed environment
• Automatic output capture and display

RAG (Retrieval Augmented Generation)

Built-in document understanding pipeline:

• PDF, DOCX, HTML, Markdown parsing
• Chunking with configurable overlap
• Vector similarity search
• Citation tracking in responses

MCP Support

Connect external tools via Model Context Protocol:

agent = Assistant(
    llm={"model": "qwen-max"},
    mcp_servers=[{
        "url": "http://localhost:8080/mcp",
        "tools": ["search", "calculator"]
    }]
)

Built-in GUI

Launch a Gradio chat interface instantly:

from qwen_agent.gui import WebUI

WebUI(agent).run(server_port=7860)

Multi-Agent Orchestration

Route tasks to specialized agents:

from qwen_agent.agents import GroupChat, Router

agents = [coder_agent, researcher_agent, writer_agent]
router = Router(llm=llm, agents=agents)

FAQ

Q: What is Qwen-Agent? A: Qwen-Agent is an open-source Python framework by Alibaba for building AI agents on Qwen models, featuring function calling, code interpretation, RAG, and MCP support. 15,800+ GitHub stars.

Q: Can I use Qwen-Agent with non-Qwen models? A: It's optimized for Qwen models but supports any OpenAI-compatible API endpoint. Best results come from Qwen-Max and Qwen-Plus which have native function calling.

Q: Is Qwen-Agent free? A: Yes, Apache-2.0 licensed. The Qwen models themselves have varying licenses — Qwen 3.0 is Apache-2.0 for most sizes.