OpenAI 推动面向大众的 AI 智能体建设

import json
import requests

# 模拟工具：一个简单的计算器
def calculate(expression: str) -> str:
    try:
        # 安全地评估基础数学表达式
        allowed_chars = "0123456789+-*/(). "
        if all(char in allowed_chars for char in expression):
            return str(eval(expression))
        return "错误：表达式中包含非法字符。"
    except Exception as e:
        return f"错误：{str(e)}"

# 使用 n1n.ai 端点的统一 API 请求辅助函数
def call_llm(messages, tools):
    # 开发者可以使用单个 API 密钥通过 n1n.ai 路由到多个模型
    url = "https://api.n1n.ai/v1/chat/completions"
    headers = {
        "Authorization": "Bearer YOUR_N1N_API_KEY",
        "Content-Type": "application/json"
    }
    payload = {
        "model": "gpt-4o",
        "messages": messages,
        "tools": tools,
        "tool_choice": "auto"
    }
    response = requests.post(url, json=payload, headers=headers)
    return response.json()

# 为 LLM 定义工具元数据
tools = [
    {
        "type": "function",
        "function": {
            "name": "calculate",
            "description": "计算数学表达式。适用于任何数学计算。",
            "parameters": {
                "type": "object",
                "properties": {
                    "expression": {
                        "type": "string",
                        "description": "要计算的数学表达式，例如 '2 + 2'"
                    }
                },
                "required": ["expression"]
            }
        }
    }
]

# 初始化对话状态
messages = [
    {"role": "user", "content": "1459 乘以 32，然后除以 4 的结果是多少？"}
]

# 运行智能体循环
print("正在启动智能体循环...")
response_data = call_llm(messages, tools)
choice = response_data["choices"][0]["message"]

if choice.get("tool_calls"):
    for tool_call in choice["tool_calls"]:
        function_name = tool_call["function"]["name"]
        arguments = json.loads(tool_call["function"]["arguments"])

        print(f"智能体决定调用工具: {function_name}，参数: {arguments}")

        if function_name == "calculate":
            tool_result = calculate(arguments["expression"])
            print(f"工具输出结果: {tool_result}")

            # 将助手的决策和工具的输出附加到对话历史记录中
            messages.append(choice)
            messages.append({
                "role": "tool",
                "tool_call_id": tool_call["id"],
                "name": function_name,
                "content": tool_result
            })

            # 再次调用 LLM 以合成最终答案
            final_response = call_llm(messages, tools)
            print("智能体最终输出:", final_response["choices"][0]["message"]["content"])
else:
    print("智能体直接回答:", choice["content"])