Build and Debug LangGraph Workflows with Alibaba Qwen in Minutes

In the AI era, learning a new technology can be done quickly, and this tutorial demonstrates how to build a simple LangGraph workflow for a chatbot using Python.

1. Basic LangGraph with OpenAI

First, import the necessary packages and define a node that calls the OpenAI model:

from langchain_openai import ChatOpenAI
from langgraph.graph import StateGraph, MessagesState, START, END

def chatbot(state: MessagesState):
    return {"messages": [ChatOpenAI(model="gpt-5-nano").invoke(state["messages"]) ]}

graph = StateGraph(MessagesState)
graph.add_node("chatbot", chatbot)
graph.add_edge(START, "chatbot")
graph.add_edge("chatbot", END)
app = graph.compile()
res = app.invoke({"messages": [("user", "你好，请用一句话介绍 LangGraph") ]})
print(res["messages"][-1].content)

You can visualize the graph with:

from IPython.display import Image, display
display(Image(app.get_graph().draw_mermaid_png()))

2. Switching to Alibaba Qwen

Replace the OpenAI import with the Alibaba‑compatible ChatAnthropic class and provide your own API key and endpoint:

from langchain_anthropic import ChatAnthropic
from langgraph.graph import StateGraph, MessagesState, START, END

llm = ChatAnthropic(
    model="qwen3.5-plus",
    api_key="sk-05bd3dxxxxxxxxxxxxxxxxxxxxxxx",
    base_url="https://dashscope.aliyuncs.com/apps/anthropic",
    max_tokens=1024,
)

def chatbot(state: MessagesState):
    return {"messages": [llm.invoke(state["messages"]) ]}

graph = StateGraph(MessagesState)
graph.add_node("chatbot", chatbot)
graph.add_edge(START, "chatbot")
graph.add_edge("chatbot", END)
app = graph.compile()
res = app.invoke({"messages": [("user", "你好，请用一句话介绍 LangGraph") ]})
print(res["messages"][-1].content)

3. Handling Warnings and Filtering Thinking Output

To silence unrelated warnings and extract only the final text answer (ignoring the model’s internal "thinking" list), add a warning filter and post‑process the response:

import warnings
warnings.filterwarnings("ignore")

from langchain_anthropic import ChatAnthropic
from langgraph.graph import StateGraph, MessagesState, START, END

llm = ChatAnthropic(...)

def chatbot(state: MessagesState):
    response = llm.invoke(state["messages"])
    # If the response contains a list of parts, keep only the text part
    if isinstance(response.content, list):
        for item in response.content:
            if item["type"] == "text":
                response.content = item["text"]
                break
    return {"messages": [response]}

# Build and run the graph as before

4. Running the Final Workflow

After applying the warning filter and content‑filtering logic, the workflow runs cleanly and returns the expected one‑sentence description of LangGraph. The graph can be visualized in a Jupyter notebook using the same Image(app.get_graph().draw_mermaid_png()) call.

5. Key Takeaways

LangGraph lets you compose LLM calls as nodes in a directed graph.

Switching between providers (OpenAI → Alibaba Qwen) only requires changing the model class and credentials.

Suppressing irrelevant warnings and filtering out the model’s internal thinking stream yields concise user‑facing answers.

The entire process—from writing a few lines of Python to obtaining a working chatbot—can be completed in under 20 minutes, even for beginners.