Langchain家庭助手Agent_langchain client 助手-优快云博客

本文链接：https://blog.youkuaiyun.com/weixin_44714085/article/details/146483068

1. 系统概述

1.1 功能需求

感知：接收用户的口头指令，并通过传感器获取环境信息（如温度、湿度等）。
行动：控制家庭电器（灯、窗帘、加湿器、电视、冰箱、空调等）。
思考：根据用户习惯和当前状态，提供优化建议（如时间管理、节能建议等），并在必要时请求用户确认。

1.2 技术栈

LangChain：用于构建对话 Agent 和工具链。
OpenAI 或其他大语言模型：作为核心推理引擎。
硬件接口：模拟或实际的家庭设备 API。
传感器数据：通过模拟或真实传感器获取环境信息。

2. 架构设计

2.1 感知模块

感知模块负责接收用户输入和获取环境信息：

用户输入可以通过语音转文字工具（如 Whisper API）转换为文本。
环境信息可以通过传感器获取（如温度、湿度、光线强度等）。

2.2 行动模块

行动模块封装了对家庭电器的控制逻辑：

每个设备都有对应的控制函数（如 toggle_light、control_curtains 等）。
这些函数被封装为 LangChain 的 Tool。

2.3 思考模块

思考模块负责以下任务：

分析用户习惯（如记录用户在特定时间的操作）。
提供优化建议（如提醒用户关闭不必要的电器以节省能源）。
根据环境信息调整设备状态（如根据湿度自动调节加湿器）。

3. 示例代码

3.1 定义硬件接口

我们首先定义一个模拟的家庭硬件接口：

class HomeHardwareAPI:
    def __init__(self):
        self.light_status = "off"
        self.curtain_status = "closed"
        self.humidifier_status = "off"
        self.tv_status = "off"
        self.fridge_temperature = 4.0  # 摄氏度
        self.air_conditioner_status = "off"
        self.temperature = 22.5  # 当前室内温度
        self.humidity = 50.0  # 当前湿度

    def toggle_light(self, action):
        if action.lower() == "on":
            self.light_status = "on"
            return "Light turned on."
        elif action.lower() == "off":
            self.light_status = "off"
            return "Light turned off."
        else:
            return "Invalid action."

    def control_curtains(self, action):
        if action.lower() == "open":
            self.curtain_status = "open"
            return "Curtains opened."
        elif action.lower() == "close":
            self.curtain_status = "closed"
            return "Curtains closed."
        else:
            return "Invalid action."

    def control_humidifier(self, action):
        if action.lower() == "on":
            self.humidifier_status = "on"
            return "Humidifier turned on."
        elif action.lower() == "off":
            self.humidifier_status = "off"
            return "Humidifier turned off."
        else:
            return "Invalid action."

    def control_tv(self, action):
        if action.lower() == "on":
            self.tv_status = "on"
            return "TV turned on."
        elif action.lower() == "off":
            self.tv_status = "off"
            return "TV turned off."
        else:
            return "Invalid action."

    def adjust_fridge_temperature(self, temperature):
        self.fridge_temperature = temperature
        return f"Fridge temperature set to {temperature}°C."

    def control_air_conditioner(self, action):
        if action.lower() == "on":
            self.air_conditioner_status = "on"
            return "Air conditioner turned on."
        elif action.lower() == "off":
            self.air_conditioner_status = "off"
            return "Air conditioner turned off."
        else:
            return "Invalid action."

    def get_temperature(self):
        return f"Current temperature is {self.temperature}°C."

    def get_humidity(self):
        return f"Current humidity is {self.humidity}%."

3.2 封装为 LangChain 工具

将上述硬件接口封装为 LangChain 的 Tool：

from langchain.agents import Tool

home_hardware_api = HomeHardwareAPI()

def toggle_light(action):
    return home_hardware_api.toggle_light(action)

def control_curtains(action):
    return home_hardware_api.control_curtains(action)

def control_humidifier(action):
    return home_hardware_api.control_humidifier(action)

def control_tv(action):
    return home_hardware_api.control_tv(action)

def adjust_fridge_temperature(temperature):
    return home_hardware_api.adjust_fridge_temperature(temperature)

def control_air_conditioner(action):
    return home_hardware_api.control_air_conditioner(action)

def get_temperature():
    return home_hardware_api.get_temperature()

def get_humidity():
    return home_hardware_api.get_humidity()

tools = [
    Tool(
        name="Toggle Light",
        func=toggle_light,
        description="Use this tool to turn the light on or off. Input should be 'on' or 'off'."
    ),
    Tool(
        name="Control Curtains",
        func=control_curtains,
        description="Use this tool to open or close the curtains. Input should be 'open' or 'close'."
    ),
    Tool(
        name="Control Humidifier",
        func=control_humidifier,
        description="Use this tool to turn the humidifier on or off. Input should be 'on' or 'off'."
    ),
    Tool(
        name="Control TV",
        func=control_tv,
        description="Use this tool to turn the TV on or off. Input should be 'on' or 'off'."
    ),
    Tool(
        name="Adjust Fridge Temperature",
        func=adjust_fridge_temperature,
        description="Use this tool to adjust the fridge temperature. Input should be a number representing the desired temperature in Celsius."
    ),
    Tool(
        name="Control Air Conditioner",
        func=control_air_conditioner,
        description="Use this tool to turn the air conditioner on or off. Input should be 'on' or 'off'."
    ),
    Tool(
        name="Get Temperature",
        func=get_temperature,
        description="Use this tool to get the current room temperature."
    ),
    Tool(
        name="Get Humidity",
        func=get_humidity,
        description="Use this tool to get the current room humidity."
    )
]

3.3 初始化 Agent

初始化 LangChain 的 Agent，并添加对话记忆功能：

from langchain.chat_models import ChatOpenAI
from langchain.agents import initialize_agent
from langchain.chains.conversation.memory import ConversationBufferMemory

# 初始化对话记忆
memory = ConversationBufferMemory(memory_key="chat_history")

# 使用 OpenAI 的 GPT-3.5-turbo 模型
llm = ChatOpenAI(temperature=0)

# 初始化 Agent
agent_chain = initialize_agent(
    tools, llm, agent="chat-conversational-react-description", verbose=True, memory=memory
)

3.4 添加思考模块

思考模块可以根据用户习惯和环境信息提供建议。例如：

def provide_suggestions():
    suggestions = []

    # 根据温度建议调整空调
    if home_hardware_api.temperature > 28:
        suggestions.append("It's quite hot. Do you want me to turn on the air conditioner?")

    # 根据湿度建议调整加湿器
    if home_hardware_api.humidity < 30:
        suggestions.append("The air is dry. Do you want me to turn on the humidifier?")

    # 节能建议
    if home_hardware_api.light_status == "on" and home_hardware_api.get_temperature().split()[2] > "25":
        suggestions.append("The light is on and it's warm. You might save energy by turning it off.")

    return suggestions

将建议集成到 Agent 中：

def run_agent(user_input):
    response = agent_chain.run(user_input)
    suggestions = provide_suggestions()
    if suggestions:
        response += "\n\nSuggestions for you:\n" + "\n".join(suggestions)
    return response

3.5 主人口头指令处理

假设主人口头指令通过语音转文字工具转换为文本，我们可以直接调用 run_agent 函数：

# 示例用户输入
user_input = "Turn on the light and open the curtains."
response = run_agent(user_input)
print(response)  # 输出: Light turned on. Curtains opened.

4. 用户习惯学习

为了根据用户习惯做出调整，可以记录用户的操作历史并分析模式。例如：

import datetime

class UserHabits:
    def __init__(self):
        self.history = []

    def log_action(self, action, timestamp=None):
        if timestamp is None:
            timestamp = datetime.datetime.now()
        self.history.append((action, timestamp))

    def analyze_habits(self):
        # 简单分析用户的日常习惯
        habits = {}
        for action, timestamp in self.history:
            time_of_day = timestamp.strftime("%H:%M")
            if time_of_day not in habits:
                habits[time_of_day] = []
            habits[time_of_day].append(action)
        return habits

user_habits = UserHabits()

# 记录用户操作
user_habits.log_action("Turned on the light")
user_habits.log_action("Opened the curtains")
print(user_habits.analyze_habits())