Qwen-Agent Library

0.0.34 · active · verified Sun Apr 12

Qwen-Agent is a Python library designed to enhance Large Language Models (LLMs) with advanced capabilities such as Agent Workflows, Retrieval-Augmented Generation (RAG), Function Calling, and Code Interpreters. It's actively developed, with version 0.0.34 released recently, maintaining a rapid release cadence with frequent minor updates.

Warnings

gotcha API Key Configuration: Users often forget to set the correct environment variables (e.g., `DASHSCOPE_API_KEY` or `OPENAI_API_KEY`) or pass them incorrectly to the LLM configuration, leading to authentication errors.
Fix: Ensure `DASHSCOPE_API_KEY` or `OPENAI_API_KEY` is set in your environment or explicitly passed in the `llm_config` dictionary. Double-check `model_server` matches your `model` and API key.
gotcha Multi-Modal Compute Platform (MCP) is an optional dependency. Features relying on MCP (e.g., certain multi-modal capabilities) may not work without a specific setup or additional installation steps.
Fix: If using MCP-dependent features, refer to the official documentation for specific installation and configuration instructions, as it may not be part of the base `qwen-agent` install.
gotcha Default LLM call and token limits can lead to unexpected truncations or early termination of agent runs. Defaults are `QWEN_AGENT_MAX_LLM_CALL_PER_RUN=20` and `QWEN_AGENT_DEFAULT_MAX_INPUT_TOKENS=58k`.
Fix: For complex or lengthy tasks, consider adjusting these limits by setting the `QWEN_AGENT_MAX_LLM_CALL_PER_RUN` and `QWEN_AGENT_DEFAULT_MAX_INPUT_TOKENS` environment variables before running the agent.
gotcha The `gradio` dependency for GUI features has seen version modifications. If you encounter issues with the GUI, it might be due to `gradio` version incompatibility.
Fix: Install `qwen-agent[gui]` to ensure the correct `gradio` version is used. If problems persist, check the library's `pyproject.toml` or `setup.py` for the exact `gradio` version pinned and adjust your environment accordingly.

Install

pip install qwen-agent Basic Install
pip install qwen-agent[gui] Install with GUI support

Imports

AssistantAgent

from qwen_agent.agents import AssistantAgent

CodeInterpreter

from qwen_agent.tools.tool_code_interpreter import CodeInterpreter

RAG
```
from qwen_agent.tools.tool_rag import RAG
```
RAG tool path might vary slightly based on specific RAG components needed, but this is common.

Quickstart

This quickstart demonstrates how to initialize an `AssistantAgent` with a Code Interpreter tool and use it to execute a simple task. It requires an API key for DashScope (for Qwen models) or OpenAI, which should be set as an environment variable.

import os
from qwen_agent.agents import AssistantAgent
from qwen_agent.tools.tool_code_interpreter import CodeInterpreter

# Configure your LLM access
# Ensure DASHSCOPE_API_KEY or OPENAI_API_KEY is set in your environment variables.
llm_config = {
    'model': 'qwen-turbo',
    'model_server': 'dashscope', # or 'openai', 'ollama', etc.
    'api_key': os.environ.get('DASHSCOPE_API_KEY', os.environ.get('OPENAI_API_KEY', ''))
}

if not llm_config['api_key']:
    raise ValueError("Please set DASHSCOPE_API_KEY or OPENAI_API_KEY environment variable.")

# Initialize the agent with tools
agent = AssistantAgent(
    llm=llm_config,
    tools=[CodeInterpreter()]
)

# Run the agent with a prompt
response = agent.run("Please write a python code snippet to calculate the sum of 123 and 456.")
print(response)

view raw JSON →