如何开发AI Agent?

import math
import requests

def get_weather(city: str) -> str:
    # 模拟调用天气 API
    return f"今天 {city} 晴，25°C"

def calculate(expression: str) -> float:
    # 安全计算（仅支持简单表达式）
    try:
        return eval(expression, {"__builtins__": {}}, {"math": math})
    except:
        return "计算错误"

from qwen_agent import Agent  # 假设使用通义千问官方 Agent 框架

# 定义工具描述（供 LLM 理解）
tools = [
    {
        "name": "get_weather",
        "description": "查询某城市的天气",
        "parameters": {"city": "城市名，如'北京'"}
    },
    {
        "name": "calculate",
        "description": "计算数学表达式",
        "parameters": {"expression": "合法的数学表达式，如'2+3*4'"}
    }
]

# 创建 Agent
agent = Agent(
    model="qwen-max",
    tools=tools,
    system_prompt="你是一个智能助手，能回答问题、查天气、做计算。"
)

# 运行
response = agent.run("北京今天天气怎么样？")
print(response)  # → 调用 get_weather("北京") → 返回天气

history = [
    {"role": "user", "content": "北京天气？"},
    {"role": "assistant", "tool_calls": [{"name": "get_weather", "args": {"city": "北京"}}]},
    {"role": "tool", "content": "今天北京晴，25°C"},
    {"role": "assistant", "content": "北京今天晴，25°C"}
]

from chromadb import Client

client = Client()
collection = client.create_collection("memory")

# 存记忆
collection.add(
    documents=["用户喜欢喝美式咖啡"],
    metadatas=[{"user_id": "123"}],
    ids=["mem1"]
)

# 查记忆
results = collection.query(query_texts=["用户喜欢什么咖啡？"], n_results=1)

pip install qwen-agent tushare TA-Lib pandas numpy

import tushare as ts
import talib
import pandas as pd

ts.set_token('YOUR_TUSHARE_TOKEN')
pro = ts.pro_api()

def get_stock_price(symbol: str) -> dict:
    df = ts.get_realtime_quotes(symbol)
    return {
        "price": float(df['price'].iloc[0]),
        "high_52w": float(df['high'].iloc[0]),
        "low_52w": float(df['low'].iloc[0])
    }

def get_technical_indicators(symbol: str, days=60) -> dict:
    df = ts.pro_bar(ts_code=symbol, adj='qfq', freq='D', start_date='20240101')
    df = df.sort_index(ascending=True)
    close = df['close'].values
    rsi = talib.RSI(close, timeperiod=14)[-1]
    macd, signal, hist = talib.MACD(close)
    ma20 = talib.SMA(close, 20)[-1]
    ma60 = talib.SMA(close, 60)[-1]
    return {
        "rsi": rsi,
        "macd": macd[-1],
        "macd_signal": signal[-1],
        "ma20": ma20,
        "ma60": ma60,
        "price": close[-1]
    }

def get_fundamentals(symbol: str) -> dict:
    # 简化：假设 symbol 是 ts_code，如 '600519.SH'
    df = pro.daily_basic(ts_code=symbol, trade_date='20240614')
    ifnot df.empty:
        return {
            "pe": df['pe'].iloc[0],
            "pb": df['pb'].iloc[0],
            "total_mv": df['total_mv'].iloc[0]  # 亿元
        }
    return {"pe": None, "pb": None}

from qwen_agent import Agent

tools = [
    {
        "name": "get_stock_price",
        "description": "获取股票当前价格和52周高低",
        "parameters": {"symbol": "股票代码，如 '600519'"}
    },
    {
        "name": "get_technical_indicators",
        "description": "获取技术指标：RSI、MACD、均线",
        "parameters": {"symbol": "股票代码"}
    },
    {
        "name": "get_fundamentals",
        "description": "获取基本面数据：PE、PB",
        "parameters": {"symbol": "股票代码（带交易所后缀，如 '600519.SH'）"}
    }
]

system_prompt = """
你是一个专业的股票投资顾问AI，擅长结合技术面与基本面给出操作建议。
请根据工具返回的数据，回答以下问题：
1. 是否值得买入？当前是否是好时机？
2. 建议买入区间、止损位、目标卖出价。
3. 适合短线（<5天）、波段（1-3月）还是长线（>6月）？
4. 风险提示（如高估值、行业利空等）。
"""

agent = Agent(
    model="qwen-max",
    tools=tools,
    system_prompt=system_prompt
)

# 用户提问
response = agent.run("分析贵州茅台（600519）现在是否适合买入？")
print(response)