从零开始理解如何使用Python搭建智能AI代理

import streamlit as st
import os
from typing import Annotated
from typing_extensions import TypedDict
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages
from langchain_community.llms import Tongyi

# ---- 环境变量输入区（可根据需求修改为硬编码） ----
st.set_page_config(page_title="Qwen 聊天机器人", page_icon="🤖", layout="centered")

with st.sidebar:
    st.title("配置参数")
    dashscope_api_key = st.text_input("DASHSCOPE_API_KEY", type="password", value=os.environ.get("DASHSCOPE_API_KEY", ""))
    langsmith_api_key = st.text_input("LANGSMITH_API_KEY（可选）", type="password", value=os.environ.get("LANGSMITH_API_KEY", ""))
    langsmith_project = st.text_input("LANGSMITH_PROJECT（可选）", value=os.environ.get("LANGSMITH_PROJECT", ""))
    langsmith_endpoint = st.text_input("LANGSMITH_ENDPOINT（可选）", value=os.environ.get("LANGSMITH_ENDPOINT", "https://api.smith.langchain.com"))
    st.markdown("> *如无需 LangSmith，可留空*")
    st.markdown("---")
    st.markdown("by [你的名字/团队]")

# 设置环境变量
if dashscope_api_key:
    os.environ["DASHSCOPE_API_KEY"] = dashscope_api_key
if langsmith_api_key:
    os.environ["LANGSMITH_API_KEY"] = langsmith_api_key
    os.environ["LANGSMITH_TRACING"] = "true"
if langsmith_project:
    os.environ["LANGSMITH_PROJECT"] = langsmith_project
if langsmith_endpoint:
    os.environ["LANGSMITH_ENDPOINT"] = langsmith_endpoint

# ---- 对话状态定义 ----
class State(TypedDict):
    messages: Annotated[list, add_messages]

# ---- 状态机组装 ----
@st.cache_resource(show_spinner="正在加载 Qwen 模型…")
def get_graph():
    graph_builder = StateGraph(State)
    llm = Tongyi(
        model="qwen-max",
        streaming=True,
        temperature=0.7,
        top_p=0.8,
        api_key=os.environ.get("DASHSCOPE_API_KEY")
    )
    def chatbot(state: State):
        return {"messages": [llm.invoke(state["messages"])]}
    graph_builder.add_node("chatbot", chatbot)
    graph_builder.add_edge(START, "chatbot")
    graph_builder.add_edge("chatbot", END)
    return graph_builder.compile()

graph = get_graph()

# ---- Streamlit 聊天 session 状态维护 ----
if "history" not in st.session_state:
    st.session_state.history = []

st.title("🤖 Qwen 聊天机器人（LangGraph 版）")

with st.form(key="chat-form", clear_on_submit=True):
    user_input = st.text_area("请输入你的问题：", value="", placeholder="和机器人聊点什么吧…", height=80)
    submitted = st.form_submit_button("发送")

def stream_graph_updates(user_input: str, history: list):
    yield from graph.stream({"messages": history + [{"role": "user", "content": user_input}]})

# 聊天历史展示
for msg in st.session_state.history:
    if msg["role"] == "user":
        st.chat_message("user").write(msg["content"])
    else:
        st.chat_message("assistant").write(msg["content"])

# 新问题处理
if submitted and user_input.strip():
    st.session_state.history.append({"role": "user", "content": user_input})
    response_placeholder = st.empty()
    full_response = ""
    # 流式显示
    with st.spinner("Assistant 正在思考..."):
        for event in stream_graph_updates(user_input, st.session_state.history[:-1]):
            for value in event.values():
                resp = value["messages"][-1]
                # 仅展示assistant内容
                # if resp["role"] == "assistant":
                #     full_response = resp
                response_placeholder.chat_message("assistant").write(resp)
    # 更新历史
    if full_response:
        st.session_state.history.append({"role": "assistant", "content": full_response})

# 清空历史按钮
st.sidebar.button("清空对话历史", on_click=lambda: st.session_state.pop("history", None))

st.info("提示：本机器人支持多轮上下文记忆，支持流式输出。输入 'quit'/'exit'/'q' 可结束对话（但不会关闭页面）。")