组合使用内置工具和函数调用

Gemini 允许在单个生成中组合使用 内置工具（例如 google_search）和 函数调用（也称为“自定义工具”），方法是保留和公开工具调用的上下文历史记录。内置工具和自定义工具组合支持复杂的代理工作流，例如，模型可以在调用您的特定业务逻辑之前，根据实时网络数据进行自我定位。

以下示例展示了如何使用 google_search 和自定义函数 getWeather 启用内置工具和自定义工具组合：

Python

from google import genai
from google.genai import types

client = genai.Client()

getWeather = {
    "name": "getWeather",
    "description": "Gets the weather for a requested city.",
    "parameters": {
        "type": "object",
        "properties": {
            "city": {
                "type": "string",
                "description": "The city and state, e.g. Utqiaġvik, Alaska",
            },
        },
        "required": ["city"],
    },
}

# Turn 1: Initial request with Google Search (built-in) and getWeather (custom) tools enabled
response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="What is the northernmost city in the United States? What's the weather like there today?",
    config=types.GenerateContentConfig(
      tools=[
        types.Tool(
          google_search=types.ToolGoogleSearch(),  # Built-in tool
          function_declarations=[getWeather]       # Custom tool
        ),
      ],
      include_server_side_tool_invocations=True
    ),
)

for part in response.candidates[0].content.parts:
    if part.tool_call:
        print(f"Tool call: {part.tool_call.tool_type} (ID: {part.tool_call.id})")
    if part.tool_response:
        print(f"Tool response: {part.tool_response.tool_type} (ID: {part.tool_response.id})")
    if part.function_call:
        print(f"Function call: {part.function_call.name} (ID: {part.function_call.id})")

# Turn 2: Manually build history to circulate both tool and function context
history = [
    types.Content(
        role="user",
        parts=[types.Part(text="What is the northernmost city in the United States? What's the weather like there today?")]
    ),
    # Response from Turn 1 includes tool_call, tool_response, and thought_signatures
    response.candidates[0].content,
    # Return the function_response
    types.Content(
        role="user",
        parts=[types.Part(
            function_response=types.FunctionResponse(
                name="getWeather",
                response={"response": "Very cold. 22 degrees Fahrenheit."},
                id=response.candidates[0].content.parts[2].function_call.id # Match the ID from the function_call
            )
        )]
    )
]

response_2 = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents=history,
    config=types.GenerateContentConfig(
      tools=[
        types.Tool(
          google_search=types.ToolGoogleSearch(),
          function_declarations=[getWeather]
        ),
      ],
      # This flag needs to be enabled for built-in tool context circulation and tool combination
      include_server_side_tool_invocations=True
    ),
)

for part in response_2.candidates[0].content.parts:
    if part.text:
        print(part.text)

JavaScript

import { GoogleGenAI } from '@google/genai';

const client = new GoogleGenAI({});

const getWeather = {
    name: "getWeather",
    description: "Get the weather in a given location",
    parameters: {
        type: "OBJECT",
        properties: {
            location: {
                type: "STRING",
                description: "The city and state, e.g. San Francisco, CA"
            }
        },
        required: ["location"]
    }
};

async function run() {
    const model = client.getGenerativeModel({
        model: "gemini-3-flash-preview",
    });

    const tools = [
      { googleSearch: {} },
      { functionDeclarations: [getWeather] }
    ];
    // This flag needs to be enabled for built-in tool context circulation and tool combination
    const toolConfig = { includeServerSideToolInvocations: true };

    // Turn 1: Initial request with Google Search (built-in) and getWeather (custom) tools enabled
    const result1 = await model.generateContent({
        contents: [{role: "user", parts: [{text: "What is the northernmost city in the United States? What's the weather like there today?"}]}],
        tools: tools,
        toolConfig: toolConfig,
    });

    const response1 = result1.response;

    for (const part of response1.candidates[0].content.parts) {
        if (part.toolCall) {
            console.log(`Tool call: ${part.toolCall.toolType} (ID: ${part.toolCall.id})`);
        }
        if (part.toolResponse) {
            console.log(`Tool response: ${part.toolResponse.toolType} (ID: ${part.toolResponse.id})`);
        }
        if (part.functionCall) {
            console.log(`Function call: ${part.functionCall.name} (ID: ${part.functionCall.id})`);
        }
    }

    const functionCallId = response1.candidates[0].content.parts.find(p => p.functionCall)?.functionCall?.id;

    // Turn 2: Manually build history to circulate both tool and function context
    const history = [
        {
            role: "user",
            parts:[{text: "What is the northernmost city in the United States? What's the weather like there today?"}]
        },
        // Response from Turn 1 includes tool_call, tool_response, and thought_signatures
        response1.candidates[0].content,
        // Return the function_response
        {
            role: "user",
            parts: [{
                functionResponse: {
                    name: "getWeather",
                    response: {response: "Very cold. 22 degrees Fahrenheit."},
                    id: functionCallId // Match the ID from the function_call
                }
            }]
        }
    ];

    const result2 = await model.generateContent({
        contents: history,
        tools: tools,
        toolConfig: toolConfig,
    });

    for (const part of result2.response.candidates[0].content.parts) {
        if (part.text) {
            console.log(part.text);
        }
    }
}

run();

Go

package main

import (
    "context"
    "fmt"
    "log"
    "os"

    "github.com/google/generative-ai-go/genai"
    "google.golang.org/api/option"
)

func main() {
    ctx := context.Background()
    client, err := genai.NewClient(ctx, option.WithAPIKey(os.Getenv("GEMINI_API_KEY")))
    if err != nil {
        log.Exit(err)
    }
    defer client.Close()

    getWeather := &genai.FunctionDeclaration{
        Name:        "getWeather",
        Description: "Get the weather in a given location",
        Parameters: &genai.Schema{
            Type: genai.Object,
            Properties: map[string]*genai.Schema{
                "location": {
                    Type:        genai.String,
                    Description: "The city and state, e.g. San Francisco, CA",
                },
            },
            Required: []string{"location"},
        },
    }

    model := client.GenerativeModel("gemini-3-flash-preview")
    model.Tools = []*genai.Tool{
        {GoogleSearch: &genai.GoogleSearch{}}, // Built-in tool
        {FunctionDeclarations: []*genai.FunctionDeclaration{getWeather}}, // Custom tool
    }
    ist := true
    model.ToolConfig = &genai.ToolConfig{
        IncludeServerSideToolInvocations: &ist, // This flag needs to be enabled for built-in tool context circulation and tool combination
    }

    chat := model.StartChat()

    // Turn 1: Initial request with Google Search (built-in) and getWeather (custom) tools enabled
    prompt := genai.Text("What is the northernmost city in the United States? What's the weather like there today?")
    resp1, err := chat.SendMessage(ctx, prompt)
    if err != nil {
        log.Exitf("SendMessage failed: %v", err)
    }

    if resp1 == nil || len(resp1.Candidates) == 0 || resp1.Candidates[0].Content == nil {
        log.Exit("empty response from model")
    }

    var functionCallID string
    for _, part := range resp1.Candidates[0].Content.Parts {
        switch p := part.(type) {
        case genai.FunctionCall:
            fmt.Printf("Function call: %s (ID: %s)\n", p.Name, p.ID)
            if p.Name == "getWeather" {
                functionCallID = p.ID
            }
        case genai.ToolCallPart:
            fmt.Printf("Tool call: %s (ID: %s)\n", p.ToolType, p.ID)
        case genai.ToolResponsePart:
            fmt.Printf("Tool response: %s (ID: %s)\n", p.ToolType, p.ID)
        }
    }

    if functionCallID == "" {
        log.Exit("no getWeather function call in response")
    }

    // Turn 2: Provide function result back to model.
    // Chat history automatically includes tool_call, tool_response, and thought_signatures from Turn 1.
    fr := genai.FunctionResponse{
        Name: "getWeather",
        ID:   functionCallID,
        Response: map[string]any{
            "response": "Very cold. 22 degrees Fahrenheit.",
        },
    }

    resp2, err := chat.SendMessage(ctx, fr)
    if err != nil {
        log.Exitf("SendMessage for turn 2 failed: %v", err)
    }

    if resp2 == nil || len(resp2.Candidates) == 0 || resp2.Candidates[0].Content == nil {
        log.Exit("empty response from model in turn 2")
    }

    for _, part := range resp2.Candidates[0].Content.Parts {
        if txt, ok := part.(genai.Text); ok {
            fmt.Println(string(txt))
        }
    }
}

REST

# Turn 1: Initial request with Google Search (built-in) and getWeather (custom) tools enabled
curl -X POST "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-flash-preview:generateContent" \
-H "Content-Type: application/json" \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-d '{
  "contents": [{
    "role": "user",
    "parts": [{
      "text": "What is the northernmost city in the United States? What'\''s the weather like there today?"
    }]
  }],
  "tools": [{
    "googleSearch": {}
  }, {
    "functionDeclarations": [{
      "name": "getWeather",
      "description": "Get the weather in a given location",
      "parameters": {
          "type": "OBJECT",
          "properties": {
              "location": {
                  "type": "STRING",
                  "description": "The city and state, e.g. San Francisco, CA"
              }
          },
          "required": ["location"]
      }
    }]
  }],
  "toolConfig": {
    "includeServerSideToolInvocations": true
  }
}'

# Turn 2: Manually build history to circulate both tool and function context
# The following request assumes you have captured candidates[0].content from Turn 1 response,
# and extracted function_call.id for getWeather.
# Replace FUNCTION_CALL_ID and insert candidate content from turn 1.
curl -X POST "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-flash-preview:generateContent" \
-H "Content-Type: application/json" \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-d '{
  "contents": [
    {
      "role": "user",
      "parts": [{"text": "What is the northernmost city in the United States? What'\''s the weather like there today?"}]
    },
    YOUR_CANDIDATE_CONTENT_FROM_TURN_1_RESPONSE,
    {
      "role": "user",
      "parts": [{
        "functionResponse": {
          "name": "getWeather",
          "id": "FUNCTION_CALL_ID",
          "response": {"response": "Very cold. 22 degrees Fahrenheit."}
        }
      }]
    }
  ],
  "tools": [{
    "googleSearch": {}
  }, {
    "functionDeclarations": [{
      "name": "getWeather",
      "description": "Get the weather in a given location",
      "parameters": {
          "type": "OBJECT",
          "properties": {
              "location": {
                  "type": "STRING",
                  "description": "The city and state, e.g. San Francisco, CA"
              }
          },
          "required": ["location"]
      }
    }]
  }],
  "toolConfig": {
    "includeServerSideToolInvocations": true
  }
}'

运作方式

Gemini 3 模型使用工具上下文循环来启用内置工具和自定义工具组合。 借助工具上下文循环，您可以保留和公开内置工具的上下文，并在同一调用中逐轮与自定义工具共享该上下文。

启用工具组合

  • 您必须将 include_server_side_tool_invocations 标志设置为 true，才能启用工具上下文循环。
  • 请添加 function_declarations 以及您要使用的 内置工具，以触发组合行为。
    • 如果您未添加 function_declarations，只要设置了该标志，工具上下文循环仍会对添加的内置工具起作用。

API 返回部分

在单个响应中，API 会返回内置工具调用的 toolCall 和 toolResponse 部分。对于函数（自定义工具）调用，API 会返回 functionCall 调用部分，用户会在下一轮中提供 functionResponse 部分。

  • toolCall 和 toolResponse：API 会返回这些部分，以保留在服务器端运行的工具的上下文及其执行结果，供下一轮使用。
  • functionCall 和 functionResponse：API 会将函数调用发送给 用户以供填写，用户会在 函数响应中将结果发送回去（这些部分是 Gemini API 中所有 函数调用的标准部分，并非 工具组合功能独有）。
  • （代码执行工具专用） executableCode 和 codeExecutionResult： 使用代码执行工具时，API 会返回 executableCode（模型生成的要执行的代码）和 codeExecutionResult（可执行代码的结果），而不是 functionCall 和 functionResponse。

您必须在每一轮中将所有部分（包括其中包含的所有字段）返回给模型，以保持上下文并启用工具 组合。

返回部分中的关键字段

API 返回的某些 部分 将包含 id、 tool_type 和 thought_signature 字段。这些字段对于维护工具上下文至关重要（因此对于工具组合也至关重要）；您需要在后续请求中返回响应中提供的所有部分。

  • id：将调用映射到其响应的唯一标识符。id 无论工具上下文循环如何，**所有函数调用响应都会设置**。 您必须在函数响应中提供与 API 在函数调用中提供的 id 相同的 id。 内置工具会在工具调用和工具响应之间自动共享 id。
    • 在所有与工具相关的部分中找到：toolCall、toolResponse、functionCall、functionResponse、executableCode、 codeExecutionResult
  • tool_type：标识正在使用的特定工具；字面内置工具（例如 URL_CONTEXT）或函数（例如 getWeather）名称。
    • 在 toolCall 和 toolResponse 部分中找到。
  • thought_signature：嵌入在 API 返回的每个部分 中的实际加密上下文。如果没有思维签名，则无法重建上下文；如果您未在每一轮中返回所有部分的思维签名，模型将出错。
    • 在所有部分中找到。

工具专用数据

某些内置工具会返回特定于工具类型的用户可见数据实参。

工具 用户可见的工具调用实参（如果有） 用户可见的工具响应（如果有）
GOOGLE_SEARCH queries search_suggestions
GOOGLE_MAPS queries places
google_maps_widget_context_token
URL_CONTEXT urls
要浏览的网址		 urls_metadata
retrieved_url：浏览的网址
url_retrieval_status：浏览状态
FILE_SEARCH 无 无

示例工具组合请求结构

以下请求结构展示了提示“美国最北端的城市是哪个？今天那里的天气怎么样？”的请求结构。它组合了三个工具：内置 Gemini 工具 google_search 和 code_execution，以及自定义函数 get_weather。

{
  "model": "models/gemini-3-flash-preview",
  "contents": [{
    "parts": [{
      "text": "What is the northernmost city in the United States? What's the weather like there today?"
    }],
    "role": "user"
  }, {
    "parts": [{
      "thoughtSignature": "...",
      "toolCall": {
        "toolType": "GOOGLE_SEARCH_WEB",
        "args": {
          "queries": ["northernmost city in the United States"]
        },
        "id": "a7b3k9p2"
      }
    }, {
      "thoughtSignature": "...",
      "toolResponse": {
        "toolType": "GOOGLE_SEARCH_WEB",
        "response": {
          "search_suggestions": "..."
        },
        "id": "a7b3k9p2"
      }
    }, {
      "functionCall": {
        "name": "getWeather",
        "args": {
          "city": "Utqiaġvik, Alaska"
        },
        "id": "m4q8z1v6"
      },
      "thoughtSignature": "..."
    }],
    "role": "model"
  }, {
    "parts": [{
      "functionResponse": {
        "name": "getWeather",
        "response": {
          "response": "Very cold. 22 degrees Fahrenheit."
        },
        "id": "m4q8z1v6"
      }
    }],
    "role": "user"
  }],
  "tools": [{
    "functionDeclarations": [{
      "name": "getWeather"
    }]
  }, {
    "googleSearch": {
    }
  }, {
    "codeExecution": {
    }
  }],
  "toolConfig": {
    "includeServerSideToolInvocations": true
  }
}

令牌和价格

请注意，请求中的 toolCall 和 toolResponse 部分会计入 prompt_token_count。由于这些中间工具步骤现在可见并返回给您，因此它们是对话历史记录的一部分。这种情况仅适用于请求，而不适用于响应。

Google 搜索工具是此规则的例外。Google 搜索已在查询级别应用自己的定价模式，因此不会重复收取令牌费用（请参阅价格页面）。

如需了解详情，请参阅令牌页面。

限制

  • 启用 include_server_side_tool_invocations 标志时，默认使用 VALIDATED 模式（不支持 AUTO 模式）
  • google_search 等内置工具依赖于位置和当前时间信息，因此，如果您的 system_instruction 或 function_declaration.description 包含冲突的位置和时间信息，工具组合功能可能无法正常运行。

支持的工具

标准工具上下文循环适用于服务器端（内置）工具。代码执行也是服务器端工具，但有自己的内置上下文循环解决方案。计算机使用和函数调用是客户端工具，也有内置的上下文循环解决方案。

工具 执行端 上下文循环支持
Google 搜索 服务器端 支持
Google 地图 服务器端 支持
网址上下文 服务器端 支持
文件搜索 服务器端 支持
代码执行 服务器端 支持（内置，使用 executableCode 和 codeExecutionResult 部分）
计算机使用 客户端 支持（内置，使用 functionCall 和 functionResponse 部分）
自定义函数 客户端 支持（内置，使用 functionCall 和 functionResponse 部分）

后续步骤