函数调用 | OpenModels Documentation

函数调用让模型决定何时调用你的应用暴露的工具。OpenModels 在支持工具调用的模型上支持兼容 OpenAI 的 tools 格式。

并非每个模型都支持工具。在生产环境使用函数调用前，请验证所选模型和供应商路由的工具支持情况，因为参数支持可能因上游供应商而异。

下面的示例使用 <tool-capable-model> 作为占位符。请将其替换为你已验证支持工具调用的模型和路由。

工作方式

使用 JSON schemas 定义一个或多个工具。
将用户消息和 tools 发送到 POST /chat/completions。
检查 assistant 是否返回了 tool_calls。
在你的应用中执行匹配的 function。
将工具结果作为 role: "tool" 消息发回。
请求模型生成最终回答。

请求

{
  "model": "<tool-capable-model>",
  "route": { "provider": "<provider-for-this-model>" },
  "route_mode": "balanced",
  "messages": [
    {
      "role": "user",
      "content": "What is the weather in San Francisco?"
    }
  ],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get current weather for a city",
        "parameters": {
          "type": "object",
          "properties": {
            "city": {
              "type": "string",
              "description": "City name, for example San Francisco"
            }
          },
          "required": ["city"],
          "additionalProperties": false
        }
      }
    }
  ],
  "tool_choice": "auto"
}

如果模型选择工具，assistant 消息会包含 tool_calls：

{
  "role": "assistant",
  "content": null,
  "tool_calls": [
    {
      "id": "call_123",
      "type": "function",
      "function": {
        "name": "get_weather",
        "arguments": "{\"city\":\"San Francisco\"}"
      }
    }
  ]
}

你的应用执行 get_weather，然后发回结果：

{
  "role": "tool",
  "tool_call_id": "call_123",
  "content": "{\"city\":\"San Francisco\",\"temperature\":\"18C\",\"condition\":\"clear\"}"
}

Python

import json
import os
from openai import OpenAI
client = OpenAI(
    base_url="https://api.getopenmodels.com/v1",
    api_key=os.environ["OM_API_KEY"],
)
def get_weather(city: str) -> dict:
    return {
        "city": city,
        "temperature": "18C",
        "condition": "clear",
    }
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {
                        "type": "string",
                        "description": "City name, for example San Francisco",
                    }
                },
                "required": ["city"],
                "additionalProperties": False,
            },
        },
    }
]
messages = [
    {
        "role": "user",
        "content": "What is the weather in San Francisco?",
    }
]
response = client.chat.completions.create(
    model="<tool-capable-model>",
    messages=messages,
    tools=tools,
    tool_choice="auto",
    extra_body={
        "route": {"provider": "<provider-for-this-model>"},
        "route_mode": "balanced",
    },
)
assistant_message = response.choices[0].message
messages.append(assistant_message.model_dump(exclude_none=True))
for tool_call in assistant_message.tool_calls or []:
    if tool_call.function.name != "get_weather":
        continue
    arguments = json.loads(tool_call.function.arguments)
    result = get_weather(city=arguments["city"])
    messages.append(
        {
            "role": "tool",
            "tool_call_id": tool_call.id,
            "content": json.dumps(result),
        }
    )
final_response = client.chat.completions.create(
    model="<tool-capable-model>",
    messages=messages,
    extra_body={
        "route": {"provider": "<provider-for-this-model>"},
        "route_mode": "balanced",
    },
)
print(final_response.choices[0].message.content)

Node.js

import OpenAI from "openai";
const client = new OpenAI({
  baseURL: "https://api.getopenmodels.com/v1",
  apiKey: process.env.OM_API_KEY,
});
function getWeather({ city }) {
  return {
    city,
    temperature: "18C",
    condition: "clear",
  };
}
const tools = [
  {
    type: "function",
    function: {
      name: "get_weather",
      description: "Get current weather for a city",
      parameters: {
        type: "object",
        properties: {
          city: {
            type: "string",
            description: "City name, for example San Francisco",
          },
        },
        required: ["city"],
        additionalProperties: false,
      },
    },
  },
];
const messages = [
  {
    role: "user",
    content: "What is the weather in San Francisco?",
  },
];
const response = await client.chat.completions.create({
  model: "<tool-capable-model>",
  route: { provider: "<provider-for-this-model>" },
  route_mode: "balanced",
  messages,
  tools,
  tool_choice: "auto",
});
const assistantMessage = response.choices[0].message;
messages.push(assistantMessage);
for (const toolCall of assistantMessage.tool_calls ?? []) {
  if (toolCall.function.name !== "get_weather") {
    continue;
  }
  const args = JSON.parse(toolCall.function.arguments);
  const result = getWeather(args);
  messages.push({
    role: "tool",
    tool_call_id: toolCall.id,
    content: JSON.stringify(result),
  });
}
const finalResponse = await client.chat.completions.create({
  model: "<tool-capable-model>",
  route: { provider: "<provider-for-this-model>" },
  route_mode: "balanced",
  messages,
});
console.log(finalResponse.choices[0].message.content);

工具选择

使用 tool_choice 控制模型是否可以调用工具：

值	行为
`auto`	模型决定是否调用工具
`none`	模型应在不调用工具的情况下回答
`{"type":"function","function":{"name":"get_weather"}}`	在模型支持时强制调用指定函数

生产指南

只执行应用明确 allowlist 的工具。
在调用任何内部系统前，验证并解析 function.arguments。
保持工具 schemas 小巧、类型明确且具体。
将工具输出发回模型时，把它视为不可信输入。
如果模型没有返回 tool_calls，请将 assistant 消息作为普通文本响应处理。
优先先使用非流式函数调用。流式工具调用需要在执行前累积部分 delta.tool_calls 参数。

计费

函数调用使用与普通聊天补全相同的积分计费方式。积分会根据输入 token、生成的输出 token、工具调用消息以及会话中包含的工具结果消耗。