smolagents颠覆传统AI智能体！ollama本地部署Hugging Face开源全新AI智能体！重新定义AI智能体开发！从入门到精通只需10分钟

HuggingFace于2024年12月31日发布了一款全新的AI代理框架SmolAgents，这是一个极简但功能强大的AI Agents库，可以让开发者用最少的代码快速构建智能代理系统。该框架的核心逻辑仅用约1000行代码就实现了完整的代理功能，极大地简化了AI代理的开发流程。

本篇笔记所对应的视频：

• 👉👉👉 通过哔哩哔哩观看
• 👉👉👉 通过YouTube观看
• 👉👉👉 我的开源项目
• 👉👉👉 请我喝咖啡
• 👉👉👉 我的微信：stoeng
• 👉👉👉 承接大模型微调、RAG、AI智能体、AI相关应用开发等项目。

smolagents简介

smolagents是一款革命性的AI智能体框架，它的主要特点是:

• 设计简洁，使用方便，仅需几行代码即可创建功能强大的AI智能体
• 执行速度远超其他同类框架
• 学习曲线平缓，新手友好
• 代码精简但功能强大

核心优势

smolagents的设计理念是”极简代码，最大功能”。它提供了：

• 安全的Python解释器和E2B沙箱环境
• 原生支持工具调用功能
• 支持多种LLM模型集成（ollama、ChatGPT、Claude等）
• 支持搜索引擎等实用工具集成
• 支持自定义工具开发
• 通过CodeAgent类实现直观高效的代码形式工具调用
• 支持Python代码直接执行工具函数
• 可通过Hugging Face Hub轻松共享和加载工具

发展前景

SmolAgents将在未来取代现有的AI智能体框架。其简洁的设计理念和强大的功能正在吸引越来越多开发者的关注，有望成为2025年AI代理开发领域的重要工具。

快速入门

安装依赖

pip install smolagents sqlalchemy google-search-results

基础使用示例

以下是一个使用ollama模型的简单示例：

from smolagents import LiteLLMModel, DuckDuckGoSearchTool
from smolagents import CodeAgent

# 使用ollama api
model = LiteLLMModel(
    model_id="ollama/mistral-nemo",
    api_base="http://127.0.0.1:11434",
    api_key="ollama",
)

agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=model)

# 测试智能体
agent.run("What's the 20th number in the Fibonacci sequence?")

搜索引擎工具实现

下面是一个集成Google新闻搜索的示例：

# 导入所需的库和模块
from smolagents.agents import ToolCallingAgent
from smolagents import tool, LiteLLMModel
from serpapi import GoogleSearch
import os
from google.colab import userdata

os.environ["OPENAI_API_KEY"] = userdata.get('OPENAI_API_KEY')

model = LiteLLMModel(model_id="gpt-4o-mini")

# # 初始化LLM模型
# model = LiteLLMModel(
#     model_id="ollama/mistral-nemo",
#     api_base="http://127.0.0.1:11434",
#     api_key="ollama",
# )


@tool
def search_ai_news(query: str) -> str:
    """
    搜索AI相关新闻并返回结果。

    Args:
        query: 要搜索的AI新闻关键词

    Returns:
        新闻搜索结果的字符串
    """
    params = {
        "engine": "google_news",
        "q": query,
        "gl": "us",
        "hl": "en",
        "api_key": "Your_API_KEY"  # 替换为你的 SerpAPI key
    }

    try:
        search = GoogleSearch(params)
        results = search.get_dict()

        if "news_results" not in results:
            return "No news results found."

        news_articles = results["news_results"][:3]  # 获取前3条新闻

        formatted_results = []
        for article in news_articles:
            title = article.get("title", "No title")
            source = article.get("source", {}).get("name", "Unknown source")
            date = article.get("date", "No date")
            link = article.get("link", "No link")

            formatted_results.append(
                f"Title: {title}\n"
                f"Source: {source}\n"
                f"Date: {date}\n"
                f"Link: {link}\n"
            )

        return "\n".join(formatted_results)

    except Exception as e:
        return f"Error occurred while fetching news: {str(e)}"


agent = ToolCallingAgent(tools=[search_ai_news], model=model)

# 执行
print(agent.run("What are the latest news about artificial intelligence?"))

天气查询工具实现

from smolagents.agents import ToolCallingAgent
from smolagents import tool, LiteLLMModel
from typing import Optional
import requests
import json
import os
from google.colab import userdata

os.environ["OPENAI_API_KEY"] = userdata.get('OPENAI_API_KEY')

model = LiteLLMModel(model_id="gpt-4o-mini")

# 设置API key，从此处获取API key：https://home.openweathermap.org/api_keys
OPENWEATHER_API_KEY = userdata.get('OPENWEATHER_API_KEY')

@tool
def get_weather(location: str, celsius: Optional[bool] = True) -> str:
    """
    获取指定位置的实时天气信息
    
    Args:
        location: 位置名称(城市名)
        celsius: 是否使用摄氏度(默认True)
    """
    try:
        # 1. 先通过地理编码API获取位置的经纬度
        geocoding_url = f"http://api.openweathermap.org/geo/1.0/direct?q={location}&limit=1&appid={OPENWEATHER_API_KEY}"
        geo_response = requests.get(geocoding_url)
        geo_data = geo_response.json()
        
        if not geo_data:
            return f"无法找到位置: {location}"
            
        lat = geo_data[0]['lat']
        lon = geo_data[0]['lon']
        
        # 2. 调用天气API获取天气数据
        weather_url = f"https://api.openweathermap.org/data/2.5/weather?lat={lat}&lon={lon}&appid={OPENWEATHER_API_KEY}"
        response = requests.get(weather_url)
        data = response.json()
        
        # 3. 温度转换(开氏度转换为摄氏度或华氏度)
        temp_k = data['main']['temp']
        if celsius:
            temp = temp_k - 273.15
            temp_unit = "°C"
        else:
            temp = (temp_k - 273.15) * 9/5 + 32
            temp_unit = "°F"
            
        # 4. 格式化天气信息
        weather_desc = data['weather'][0]['description']
        humidity = data['main']['humidity']
        wind_speed = data['wind']['speed']
        
        weather_info = (
            f"当前{location}的天气情况:\n"
            f"- 天气状况: {weather_desc}\n"
            f"- 温度: {temp:.1f}{temp_unit}\n"
            f"- 湿度: {humidity}%\n"
            f"- 风速: {wind_speed}m/s"
        )
        
        return weather_info
        
    except Exception as e:
        return f"获取天气信息时发生错误: {str(e)}"

agent = ToolCallingAgent(tools=[get_weather], model=model)

# 测试
print(agent.run("What's the weather like in Paris?"))
print(agent.run("What's the weather like in New York? Use Fahrenheit."))

Text to SQL实现

from sqlalchemy import (
    create_engine,
    MetaData,
    Table,
    Column,
    String,
    Integer,
    Float,
    ForeignKey,
    insert,
    inspect,
    text,
)
from smolagents import tool, CodeAgent, LiteLLMModel

# 创建内存数据库
engine = create_engine("sqlite:///:memory:")
metadata_obj = MetaData()

# 创建商品表
products = Table(
    "products",
    metadata_obj,
    Column("product_id", Integer, primary_key=True),
    Column("product_name", String(50), nullable=False),
    Column("category", String(20)),
    Column("unit_price", Float, nullable=False),
    Column("stock_quantity", Integer, nullable=False),
)

# 创建订单表
orders = Table(
    "orders",
    metadata_obj,
    Column("order_id", Integer, primary_key=True),
    Column("product_id", Integer, ForeignKey("products.product_id")),
    Column("customer_name", String(50)),
    Column("order_date", String(10)),
    Column("quantity", Integer),
    Column("total_amount", Float),
)

metadata_obj.create_all(engine)

# 插入示例商品数据
product_data = [
    {"product_id": 1, "product_name": "iPhone 14", "category": "手机", "unit_price": 5999.0, "stock_quantity": 100},
    {"product_id": 2, "product_name": "MacBook Pro", "category": "笔记本", "unit_price": 12999.0, "stock_quantity": 50},
    {"product_id": 3, "product_name": "AirPods Pro", "category": "配件", "unit_price": 1999.0, "stock_quantity": 200},
    {"product_id": 4, "product_name": "iPad Air", "category": "平板", "unit_price": 4799.0, "stock_quantity": 80},
    {"product_id": 5, "product_name": "Apple Watch", "category": "智能手表", "unit_price": 3299.0,
     "stock_quantity": 150},
    {"product_id": 6, "product_name": "小米13", "category": "手机", "unit_price": 3999.0, "stock_quantity": 120},
    {"product_id": 7, "product_name": "华为Mate60", "category": "手机", "unit_price": 6999.0, "stock_quantity": 90},
    {"product_id": 8, "product_name": "联想小新Pro", "category": "笔记本", "unit_price": 5999.0, "stock_quantity": 45},
    {"product_id": 9, "product_name": "华为MateBook", "category": "笔记本", "unit_price": 6599.0, "stock_quantity": 60},
    {"product_id": 10, "product_name": "小米手环", "category": "智能手表", "unit_price": 249.0, "stock_quantity": 300},
    {"product_id": 11, "product_name": "华为手表", "category": "智能手表", "unit_price": 1499.0, "stock_quantity": 180},
    {"product_id": 12, "product_name": "荣耀手机", "category": "手机", "unit_price": 2999.0, "stock_quantity": 150},
    {"product_id": 13, "product_name": "戴尔XPS", "category": "笔记本", "unit_price": 9999.0, "stock_quantity": 30},
    {"product_id": 14, "product_name": "机械键盘", "category": "配件", "unit_price": 399.0, "stock_quantity": 250},
    {"product_id": 15, "product_name": "无线鼠标", "category": "配件", "unit_price": 129.0, "stock_quantity": 400},
    {"product_id": 16, "product_name": "显示器", "category": "配件", "unit_price": 1299.0, "stock_quantity": 120},
    {"product_id": 17, "product_name": "小米平板", "category": "平板", "unit_price": 2299.0, "stock_quantity": 100},
    {"product_id": 18, "product_name": "华为平板", "category": "平板", "unit_price": 3599.0, "stock_quantity": 85},
    {"product_id": 19, "product_name": "蓝牙耳机", "category": "配件", "unit_price": 499.0, "stock_quantity": 300},
    {"product_id": 20, "product_name": "手机壳", "category": "配件", "unit_price": 49.0, "stock_quantity": 1000},
    {"product_id": 21, "product_name": "充电器", "category": "配件", "unit_price": 129.0, "stock_quantity": 800},
    {"product_id": 22, "product_name": "移动电源", "category": "配件", "unit_price": 199.0, "stock_quantity": 600},
    {"product_id": 23, "product_name": "ThinkPad", "category": "笔记本", "unit_price": 8999.0, "stock_quantity": 40},
]

for row in product_data:
    stmt = insert(products).values(**row)
    with engine.begin() as connection:
        connection.execute(stmt)

# 插入示例订单数据
order_data = [
    {"order_id": 1, "product_id": 1, "customer_name": "张三", "order_date": "2024-01-01", "quantity": 2,
     "total_amount": 11998.0},
    {"order_id": 2, "product_id": 2, "customer_name": "李四", "order_date": "2024-01-02", "quantity": 1,
     "total_amount": 12999.0},
    {"order_id": 3, "product_id": 3, "customer_name": "王五", "order_date": "2024-01-03", "quantity": 3,
     "total_amount": 5997.0},
    {"order_id": 4, "product_id": 4, "customer_name": "赵六", "order_date": "2024-01-03", "quantity": 1,
     "total_amount": 4799.0},
    {"order_id": 5, "product_id": 5, "customer_name": "钱七", "order_date": "2024-01-04", "quantity": 2,
     "total_amount": 6598.0},
    {"order_id": 6, "product_id": 1, "customer_name": "孙八", "order_date": "2024-01-04", "quantity": 1,
     "total_amount": 5999.0},
    {"order_id": 7, "product_id": 7, "customer_name": "周九", "order_date": "2024-01-05", "quantity": 1,
     "total_amount": 6999.0},
    {"order_id": 8, "product_id": 10, "customer_name": "吴十", "order_date": "2024-01-05", "quantity": 5,
     "total_amount": 1245.0},
    {"order_id": 9, "product_id": 14, "customer_name": "郑十一", "order_date": "2024-01-06", "quantity": 3,
     "total_amount": 1197.0},
    {"order_id": 10, "product_id": 6, "customer_name": "王婷婷", "order_date": "2024-01-06", "quantity": 2,
     "total_amount": 7998.0},
    {"order_id": 11, "product_id": 8, "customer_name": "李明", "order_date": "2024-01-07", "quantity": 1,
     "total_amount": 5999.0},
    {"order_id": 12, "product_id": 12, "customer_name": "张华", "order_date": "2024-01-07", "quantity": 2,
     "total_amount": 5998.0},
    {"order_id": 13, "product_id": 15, "customer_name": "刘芳", "order_date": "2024-01-08", "quantity": 4,
     "total_amount": 516.0},
    {"order_id": 14, "product_id": 17, "customer_name": "陈明", "order_date": "2024-01-08", "quantity": 1,
     "total_amount": 2299.0},
    {"order_id": 15, "product_id": 20, "customer_name": "王磊", "order_date": "2024-01-09", "quantity": 10,
     "total_amount": 490.0},
    {"order_id": 16, "product_id": 3, "customer_name": "李娜", "order_date": "2024-01-09", "quantity": 2,
     "total_amount": 3998.0},
    {"order_id": 17, "product_id": 9, "customer_name": "张伟", "order_date": "2024-01-10", "quantity": 1,
     "total_amount": 6599.0},
    {"order_id": 18, "product_id": 11, "customer_name": "王秀英", "order_date": "2024-01-10", "quantity": 2,
     "total_amount": 2998.0},
    {"order_id": 19, "product_id": 19, "customer_name": "李军", "order_date": "2024-01-11", "quantity": 3,
     "total_amount": 1497.0},
    {"order_id": 20, "product_id": 13, "customer_name": "刘洋", "order_date": "2024-01-11", "quantity": 1,
     "total_amount": 9999.0},
    {"order_id": 21, "product_id": 21, "customer_name": "张萍", "order_date": "2024-01-12", "quantity": 5,
     "total_amount": 645.0},
    {"order_id": 22, "product_id": 22, "customer_name": "王强", "order_date": "2024-01-12", "quantity": 3,
     "total_amount": 597.0},
    {"order_id": 23, "product_id": 16, "customer_name": "李艳", "order_date": "2024-01-13", "quantity": 2,
     "total_amount": 2598.0},
]

for row in order_data:
    stmt = insert(orders).values(**row)
    with engine.begin() as connection:
        connection.execute(stmt)


# 获取表结构信息
def get_table_info(table_name):
    inspector = inspect(engine)
    columns_info = [(col["name"], col["type"]) for col in inspector.get_columns(table_name)]
    return "Columns:\n" + "\n".join([f"  - {name}: {col_type}" for name, col_type in columns_info])


@tool
def sql_engine(query: str) -> str:
    """
    执行SQL查询,支持商品表(products)和订单表(orders)的联合查询。

    商品表结构:
    - product_id: INTEGER (主键)
    - product_name: VARCHAR(50)
    - category: VARCHAR(20)
    - unit_price: FLOAT
    - stock_quantity: INTEGER

    订单表结构:
    - order_id: INTEGER (主键)
    - product_id: INTEGER (外键关联products表)
    - customer_name: VARCHAR(50)
    - order_date: VARCHAR(10)
    - quantity: INTEGER
    - total_amount: FLOAT

    Args:
        query: SQL查询语句
    """
    output = ""
    with engine.connect() as con:
        rows = con.execute(text(query))
        for row in rows:
            output += "\n" + str(row)
    return output


# 创建agent实例
agent = CodeAgent(
    tools=[sql_engine],
    model=LiteLLMModel(model_id="gpt-4o-mini"),
)

# 示例查询
# print("===示例查询1: 获取销量最高的商品信息===")
# agent.run("""
# 查询销量最高的商品的详细信息,包括商品名称、销量和销售总额。
# 需要联合查询products和orders表,按订单数量统计。
# """)

# print("\n===示例查询2: 获取库存紧张的商品===")
# agent.run("""
# 查询当前库存数量低于50件的商品信息,
# 返回商品名称、类别、当前库存数量,按库存数量升序排序。
# """)

# print("===示例查询3: 各类别商品的销售统计===")
# agent.run("""
# 统计每个商品类别的总销售额和销售数量,
# 按销售总额降序排序,同时显示该类别的商品数量。
# """)

print("\n===示例查询4: 热销商品分析===")
agent.run("""
查找单价超过1000元且销量超过2件的商品,
显示商品名称、单价、销量、销售总额,
按销售总额降序排序。
""")

print("\n===示例查询5: 客户购买力分析===")
agent.run("""
统计每个客户的消费总额和购买次数,
只显示消费总额超过10000元的客户,
按消费总额降序排序。
""")

print("\n===示例查询6: 商品库存周转分析===")
agent.run("""
计算每个商品的销量与库存比率(周转率),
显示商品名称、销量、当前库存、周转率,
只显示周转率超过10%的商品,
按周转率降序排序。
""")

print("\n===示例查询7: 日销售趋势分析===")
agent.run("""
统计每天的订单数量、销售总额、平均订单金额,
按日期升序排序。
""")

结语

smolagents的出现让AI智能体开发变得前所未有的简单和高效。无论你是AI开发新手还是经验丰富的开发者，都能快速上手并构建强大的AI应用。它的开源发布必将推动AI智能体技术的普及和发展。