{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b1a31a19",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from langchain_openai import ChatOpenAI\n",
    "from langchain_core.tools import tool\n",
    "from langchain.agents import create_agent\n",
    "from dotenv import load_dotenv\n",
    "\n",
    "load_dotenv()\n",
    "\n",
    "# ============================================================\n",
    "# 第一步：初始化模型\n",
    "# ============================================================\n",
    "model = ChatOpenAI(\n",
    "    model_name=\"deepseek-chat\",\n",
    "    api_key=os.getenv(\"OPENAI_API_KEY\"),\n",
    "    base_url=\"https://api.deepseek.com\"\n",
    ")\n",
    "\n",
    "\n",
    "# ============================================================\n",
    "# 第二步：定义工具（只有两个，一清二楚）\n",
    "# docstring 就是给模型的\"使用说明书\"\n",
    "# ============================================================\n",
    "\n",
    "@tool\n",
    "def get_weather(city: str) -> str:\n",
    "    \"\"\"查询指定城市的实时天气。\n",
    "\n",
    "    当用户问\"某地天气怎么样\"、\"热不热\"、\"要不要带伞\"时调用。\n",
    "\n",
    "    Args:\n",
    "        city: 城市名称，例如\"北京\"、\"上海\"、\"深圳\"\n",
    "    \"\"\"\n",
    "    # 模拟天气数据（实际项目中对接真实天气 API）\n",
    "    weather_db = {\n",
    "        \"北京\": \"☀️ 晴天，32°C，湿度 45%，适合出行\",\n",
    "        \"上海\": \"🌧️ 小雨，28°C，湿度 80%，建议带伞\",\n",
    "        \"深圳\": \"⛅ 多云，30°C，湿度 65%\",\n",
    "    }\n",
    "    return weather_db.get(city, f\"抱歉，暂无「{city}」的天气数据\")\n",
    "\n",
    "\n",
    "@tool\n",
    "def calculator(expression: str) -> str:\n",
    "    \"\"\"执行数学计算，支持加减乘除和括号。\n",
    "\n",
    "    当用户说\"帮我算\"、\"等于多少\"、\"计算一下\"时调用。\n",
    "\n",
    "    Args:\n",
    "        expression: 数学表达式，例如 \"128 * 456\"、\"2 ** 10\"、\"(88 + 12) / 5\"\n",
    "    \"\"\"\n",
    "    try:\n",
    "        result = eval(expression)\n",
    "        return f\"🧮 {expression} = {result}\"\n",
    "    except Exception as e:\n",
    "        return f\"❌ 计算出错：{e}\"\n",
    "\n",
    "\n",
    "# ============================================================\n",
    "# 第三步：组装 Agent\n",
    "# create_agent 内部自动构建了 Think→Act→Observe 循环\n",
    "# ============================================================\n",
    "tools = [get_weather, calculator]\n",
    "\n",
    "agent = create_agent(\n",
    "    model=model,\n",
    "    tools=tools,\n",
    "    system_prompt=(\n",
    "        \"你是一个热心肠的生活助手，会查天气、会做算术。\\n\"\n",
    "        \"规则：\\n\"\n",
    "        \"1. 用户问天气 → 调用 get_weather。\\n\"\n",
    "        \"2. 用户要计算 → 调用 calculator。\\n\"\n",
    "        \"3. 一个请求涉及多个工具时，逐个调用，每次观察结果后再决定下一步。\\n\"\n",
    "        \"4. 闲聊类的问候直接回答，不用调工具。\"\n",
    "    ),\n",
    ")\n",
    "\n",
    "# ============================================================\n",
    "# 第四步：运行 Agent\n",
    "# ============================================================\n",
    "response = agent.invoke({\n",
    "    \"messages\": [\n",
    "        {\"role\": \"user\", \"content\": \"北京今天天气怎么样？顺便帮我算一下 356 × 128\"}\n",
    "    ]\n",
    "})\n",
    "\n",
    "print(response[\"messages\"][-1].content)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f213ee6d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# ============================================================\n",
    "# stream_mode=\"values\"\n",
    "# 每次迭代拿到的是图的完整状态（messages 的完整列表）\n",
    "# 适合需要\"每步都有完整上下文\"的场景，比如调试、审计日志\n",
    "# ============================================================\n",
    "\n",
    "# 沿用第一节的 agent 和 tools，直接调用 .stream()\n",
    "inputs = {\"messages\": [(\"human\", \"帮我查一下上海天气，再算算 999 除以 37 等于多少\")]}\n",
    "\n",
    "for chunk in agent.stream(inputs, stream_mode=\"values\"):\n",
    "    # chunk 是一个 dict，key 为 \"messages\"，value 是截至目前所有消息的完整列表\n",
    "    last_msg = chunk[\"messages\"][-1]\n",
    "    # pretty_print() 会根据消息类型（人类/AI/工具）用不同颜色格式化输出\n",
    "    last_msg.pretty_print()\n",
    "    print(\"---\")  # 分隔符，方便看到每次迭代的边界"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "09974934",
   "metadata": {},
   "outputs": [],
   "source": [
    "# ============================================================\n",
    "# stream_mode=\"updates\"\n",
    "# 只返回本轮节点产生的新内容，不是全量\n",
    "# 适合需要\"追踪每个节点的输入输出\"的场景，比如调试工具调用\n",
    "# ============================================================\n",
    "\n",
    "inputs = {\"messages\": [(\"human\", \"深圳天气怎么样？顺便帮我算一下 (88 + 12) × 5\")]}\n",
    "\n",
    "for chunk in agent.stream(inputs, stream_mode=\"updates\"):\n",
    "    # chunk 的结构：{ 节点名称: 该节点本次产出的数据 }\n",
    "    # 例如 {\"agent\": {...}} 或 {\"tools\": {...}}\n",
    "    for node_name, node_output in chunk.items():\n",
    "        print(f\"📍 节点: {node_name}\")\n",
    "        print(f\"📤 产出: {node_output}\")\n",
    "        print(\"=\" * 40)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d2749d93",
   "metadata": {},
   "outputs": [],
   "source": [
    "# ============================================================\n",
    "# stream_mode=\"messages\"\n",
    "# 最细粒度的流式：LLM 每生成一个 token，你就能拿到它\n",
    "# 适合聊天 UI 那种\"一个字一个字往外蹦\"的打字机效果\n",
    "# ============================================================\n",
    "\n",
    "inputs = {\"messages\": [(\"human\", \"简单介绍一下敏捷开发中的每日站会\")]}\n",
    "\n",
    "for chunk in agent.stream(inputs, stream_mode=\"messages\"):\n",
    "    # chunk 是一个元组：(message_obj, metadata_dict)\n",
    "    message, metadata = chunk\n",
    "\n",
    "    # 关键：message.content 可能为空字符串\n",
    "    # （比如工具调用的内部消息不产生可见文本）\n",
    "    # 所以必须判断 content 是否有值再打印\n",
    "    if message.content:\n",
    "        # end=\"\" 不换行，flush=True 立即刷新缓冲区 → 打字机效果\n",
    "        print(message.content, end=\"\", flush=True)\n",
    "\n",
    "print()  # 最后补一个换行"
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}