# 创建虚拟环境

python -m venv langchain-env

source langchain-env/bin/activate

# 安装LangChain核心库

pip install langchain langchain-core

# 安装大模型provider

pip install openai

# 安装必要工具

pip install requests duckduckgo-search

import os

# 设置OpenAI API Key

os.environ["OPENAI_API_KEY"] = "your-api-key-here"

⚠️注意：生产环境中请使用环境变量或密钥管理服务，避免硬编码API Key。

👤 用户："北京明天天气怎么样？"

🤖 AI："北京明天多云转晴，最高温度18℃，适合外出。"

👤 用户："周末想去上海旅行，带什么衣服合适？"

🤖 AI："上海本周末有小雨，气温15-22℃，建议带薄外套和雨伞。"

from langchain.agents import tool

import requests

@tool

def get_weather(city: str, days: int = 1) -> str:

    """

    查询城市天气的工具。

    参数:

        city: 城市名称（中文）

        days: 预报天数，默认1天

    返回:

        天气信息字符串

    """

    try:

        url = f"https://wttr.in/{city}?format=j1"

        response = requests.get(url, timeout=5)

        data = response.json()

        # ... 处理数据

    except Exception as e:

        return f"查询天气失败：{str(e)}"

from langchain.agents import AgentExecutor, create_react_agent

from langchain_openai import ChatOpenAI

from langchain import hub

# 1. 初始化大模型

llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)

# 2. 创建Agent

prompt = hub.pull("hwchase17/react")

agent = create_react_agent(llm, [get_weather], prompt)

# 3. 创建执行器

agent_executor = AgentExecutor(agent=agent, tools=[get_weather], verbose=True)

# 测试对话

queries = ["北京明天天气怎么样？", "周末去上海需要带伞吗？"]

for query in queries:

    print(f"\n{'='*50}")

    result = agent_executor.invoke({"input": query})

    print(result['output'])

from langchain.prompts import PromptTemplate

weather_template = PromptTemplate.from_template("""

你是一个专业的天气预报助手。

用户问题：{user_question}

请根据以下信息回答：{weather_info}

""")

from langchain.chains import LLMChain

chain = LLMChain(llm=llm, prompt=weather_template)

result = chain.run({"user_question": "...", "weather_info": "..."})

from langchain.memory import ConversationBufferMemory

memory = ConversationBufferMemory(memory_key="chat_history",

                      return_messages=True)

# 在Agent中使用记忆

agent_executor = AgentExecutor(agent=agent, memory=memory, ...)

from langchain.agents import tool

@tool

def search_web(query: str) -> str:

    """搜索互联网获取最新信息"""

    from duckduckgo_search import DDGS

    with DDGS() as ddgs:

        results = list(ddgs.text(query, max_results=5))

    return f"搜索结果：{results}"

# 创建包含多个工具的Agent

tools = [get_weather, search_web]

agent = create_react_agent(llm, tools, prompt)

agent_executor = AgentExecutor(agent=agent, tools=tools)

# Agent可以自动选择使用哪个工具

result = agent_executor.invoke({"input": "北京明天有雨吗？适合户外跑步吗？"})

class SafeAgentExecutor:

    def invoke(self, input_dict, retries=3):

        for i in range(retries):

            try:

                return self.executor.invoke(input_dict)

            except Exception as e:

                if i == retries - 1:

                    return {"output": f"处理失败：{str(e)}"}

            continue

def smart_router(query: str) -> str:

    # 简单查询用便宜模型

    simple_keywords = ["天气", "时间", "日期"]

    if any(k in query for k in simple_keywords):

        return get_weather.invoke({"city": extract_city(query)})

    # 复杂问题用高级模型

    return agent_executor.invoke({"input": query})