概述

Function Calling 是 LLM 与外部系统交互的核心能力。通过结构化输出和正确的 Tool 定义，Agent 可以可靠地调用工具。

结构化输出配置

OpenAI

from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "查询北京天气"}],
    tools=[{"type": "function", "function": {"name": "get_weather", "parameters": {"type": "object", "properties": {"location": {"type": "string"}}, "required": ["location"]}}}]
)
tool_call = response.choices[0].message.tool_calls[0]

Anthropic

from anthropic import Anthropic
client = Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "查询北京天气"}],
    tools=[{"name": "get_weather", "input_schema": {"type": "object", "properties": {"location": {"type": "string"}}, "required": ["location"]}}]
)

Tool 定义最佳实践

1. 清晰的描述：每个参数都需要详细的 description
2. 类型约束：使用 JSON Schema 定义严格的类型
3. 枚举限制：使用 enum 限制可选值范围

调用失败处理

def execute_tool_call(tool_call):
    try:
        return execute_function(tool_call)
    except ValidationError as e:
        return f"参数验证失败: {e}"

参考资料

• OpenAI Function Calling
• Anthropic Tool Use