@librechat/agents 3.4.6 → 3.5.0

package/src/llm/google/llm.spec.ts

@@ -1,1388 +0,0 @@
-import { config } from 'dotenv';
-config();
-import { beforeEach, test, jest } from '@jest/globals';
-
-jest.setTimeout(Number(process.env.GOOGLE_TEST_TIMEOUT_MS ?? 120000));
-import * as fs from 'node:fs/promises';
-import * as path from 'node:path';
-import {
-  AIMessage,
-  AIMessageChunk,
-  HumanMessage,
-  SystemMessage,
-  ToolMessage,
-} from '@langchain/core/messages';
-import {
-  ChatPromptTemplate,
-  MessagesPlaceholder,
-} from '@langchain/core/prompts';
-import { StructuredTool, tool } from '@langchain/core/tools';
-import { z } from 'zod/v3';
-import {
-  DynamicRetrievalMode,
-  SchemaType as FunctionDeclarationSchemaType,
-} from '@google/generative-ai';
-import { concat } from '@langchain/core/utils/stream';
-import type {
-  CodeExecutionTool,
-  GoogleSearchRetrievalTool,
-} from '@google/generative-ai';
-import type { ContentBlock } from '@langchain/core/messages';
-import { CustomChatGoogleGenerativeAI as ChatGoogleGenerativeAI } from './index';
-import {
-  convertMessageContentToParts,
-  convertResponseContentToChatGenerationChunk,
-  mapGenerateContentResultToChatResult,
-  _FUNCTION_CALL_THOUGHT_SIGNATURES_MAP_KEY,
-} from './utils/common';
-
-// Save the original value of the 'LANGCHAIN_CALLBACKS_BACKGROUND' environment variable
-const originalBackground = process.env.LANGCHAIN_CALLBACKS_BACKGROUND;
-
-const GOOGLE_TEST_MAX_RETRIES = Number(
-  process.env.GOOGLE_TEST_MAX_RETRIES ?? 2
-);
-const GOOGLE_TEST_CALL_GAP_MS = Number(
-  process.env.GOOGLE_TEST_CALL_GAP_MS ?? 1000
-);
-const GOOGLE_TEST_MAX_RETRY_DELAY_MS = Number(
-  process.env.GOOGLE_TEST_MAX_RETRY_DELAY_MS ?? 50000
-);
-let lastGoogleTestCallAt = 0;
-
-const wait = async (ms: number): Promise<void> =>
-  new Promise((resolve) => {
-    setTimeout(resolve, ms);
-  });
-
-async function waitForGoogleTestSlot(): Promise<void> {
-  const waitMs = Math.max(
-    0,
-    lastGoogleTestCallAt + GOOGLE_TEST_CALL_GAP_MS - Date.now()
-  );
-  if (waitMs > 0) {
-    await wait(waitMs);
-  }
-  lastGoogleTestCallAt = Date.now();
-}
-
-const GOOGLE_NO_RETRY_STATUSES = new Set([
-  400, 401, 402, 403, 404, 405, 406, 407, 409,
-]);
-
-interface GoogleRetryInfoDetail {
-  '@type'?: string;
-  retryDelay?: string;
-}
-
-interface GoogleFetchErrorLike extends Error {
-  status?: number;
-  errorDetails?: GoogleRetryInfoDetail[];
-}
-
-interface ErrorWithCode extends Error {
-  code?: string;
-}
-
-function throwIfNonRetryableError(error: unknown): void {
-  if (!(error instanceof Error)) {
-    return;
-  }
-  if (
-    error.message.startsWith('Cancel') ||
-    error.message.startsWith('AbortError') ||
-    error.name === 'AbortError'
-  ) {
-    throw error;
-  }
-  const errorCode = (error as Partial<ErrorWithCode>).code;
-  if (errorCode === 'ECONNABORTED') {
-    throw error;
-  }
-}
-
-function getGoogleFetchError(error: unknown): GoogleFetchErrorLike | undefined {
-  if (!(error instanceof Error)) {
-    return undefined;
-  }
-  const possibleError = error as Partial<GoogleFetchErrorLike>;
-  if (
-    typeof possibleError.status !== 'number' &&
-    !Array.isArray(possibleError.errorDetails)
-  ) {
-    return undefined;
-  }
-  return possibleError as GoogleFetchErrorLike;
-}
-
-function parseGoogleRetryDelayMs(retryDelay: string): number | undefined {
-  const retryDelaySeconds = /^(\d+(?:\.\d+)?)s$/.exec(retryDelay)?.[1];
-  if (retryDelaySeconds === undefined) {
-    return undefined;
-  }
-  return Math.ceil(Number(retryDelaySeconds) * 1000);
-}
-
-function getGoogleRetryDelayMs(
-  error: GoogleFetchErrorLike
-): number | undefined {
-  const retryInfoDelay = error.errorDetails?.find(
-    (detail) => detail['@type'] === 'type.googleapis.com/google.rpc.RetryInfo'
-  )?.retryDelay;
-  if (retryInfoDelay !== undefined) {
-    return parseGoogleRetryDelayMs(retryInfoDelay);
-  }
-  const messageDelay = /Please retry in (\d+(?:\.\d+)?)s/.exec(
-    error.message
-  )?.[1];
-  if (messageDelay === undefined) {
-    return undefined;
-  }
-  return Math.ceil(Number(messageDelay) * 1000);
-}
-
-async function handleGoogleFailedAttempt(error: unknown): Promise<void> {
-  throwIfNonRetryableError(error);
-  const googleError = getGoogleFetchError(error);
-  if (googleError === undefined) {
-    return;
-  }
-  if (
-    googleError.status !== undefined &&
-    GOOGLE_NO_RETRY_STATUSES.has(googleError.status)
-  ) {
-    throw googleError;
-  }
-  if (googleError.status !== 429) {
-    return;
-  }
-  if (/\blimit:\s*0\b/.test(googleError.message)) {
-    throw googleError;
-  }
-  const retryDelayMs = getGoogleRetryDelayMs(googleError);
-  if (retryDelayMs === undefined) {
-    return;
-  }
-  await wait(Math.min(retryDelayMs, GOOGLE_TEST_MAX_RETRY_DELAY_MS));
-}
-
-function createGoogleModel(
-  fields: ConstructorParameters<typeof ChatGoogleGenerativeAI>[0]
-): ChatGoogleGenerativeAI {
-  return new ChatGoogleGenerativeAI({
-    ...fields,
-    maxRetries: GOOGLE_TEST_MAX_RETRIES,
-    onFailedAttempt: handleGoogleFailedAttempt,
-  });
-}
-
-beforeEach(async () => {
-  await waitForGoogleTestSlot();
-});
-
-const dummyToolResponse =
-  "[{\"title\":\"Weather in New York City\",\"url\":\"https://www.weatherapi.com/\",\"content\":\"{'location': {'name': 'New York', 'region': 'New York', 'country': 'United States of America', 'lat': 40.71, 'lon': -74.01, 'tz_id': 'America/New_York', 'localtime_epoch': 1718659486, 'localtime': '2024-06-17 17:24'}, 'current': {'last_updated_epoch': 1718658900, 'last_updated': '2024-06-17 17:15', 'temp_c': 27.8, 'temp_f': 82.0, 'is_day': 1, 'condition': {'text': 'Partly cloudy', 'icon': '//cdn.weatherapi.com/weather/64x64/day/116.png', 'code': 1003}, 'wind_mph': 2.2, 'wind_kph': 3.6, 'wind_degree': 159, 'wind_dir': 'SSE', 'pressure_mb': 1021.0, 'pressure_in': 30.15, 'precip_mm': 0.0, 'precip_in': 0.0, 'humidity': 58, 'cloud': 25, 'feelslike_c': 29.0, 'feelslike_f': 84.2, 'windchill_c': 26.9, 'windchill_f': 80.5, 'heatindex_c': 27.9, 'heatindex_f': 82.2, 'dewpoint_c': 17.1, 'dewpoint_f': 62.8, 'vis_km': 16.0, 'vis_miles': 9.0, 'uv': 7.0, 'gust_mph': 18.3, 'gust_kph': 29.4}}\",\"score\":0.98192,\"raw_content\":null},{\"title\":\"New York, NY Monthly Weather | AccuWeather\",\"url\":\"https://www.accuweather.com/en/us/new-york/10021/june-weather/349727\",\"content\":\"Get the monthly weather forecast for New York, NY, including daily high/low, historical averages, to help you plan ahead.\",\"score\":0.97504,\"raw_content\":null}]";
-
-test('Test Google AI', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-  const res = await model.invoke('what is 1 + 1?');
-  expect(res).toBeTruthy();
-});
-
-test('Test Google AI generation', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-  const res = await model.generate([
-    [['human', 'Translate "I love programming" into Korean.']],
-  ]);
-  expect(res).toBeTruthy();
-});
-
-test('Test Google AI generation with a stop sequence', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-    stopSequences: ['two', '2'],
-  });
-  const res = await model.invoke([
-    ['human', 'What are the first three positive whole numbers?'],
-  ]);
-  expect(res).toBeTruthy();
-  expect(res.additional_kwargs.finishReason).toBe('STOP');
-  expect(res.content).not.toContain('2');
-  expect(res.content).not.toContain('two');
-});
-
-test('Test Google AI generation with a system message', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-  const res = await model.generate([
-    [
-      ['system', 'You are an amazing translator.'],
-      ['human', 'Translate "I love programming" into Korean.'],
-    ],
-  ]);
-  expect(res).toBeTruthy();
-});
-
-test('Test Google AI multimodal generation', async () => {
-  const imageData = (
-    await fs.readFile(path.join(__dirname, '/data/hotdog.jpg'))
-  ).toString('base64');
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-  const res = await model.invoke([
-    new HumanMessage({
-      content: [
-        {
-          type: 'text',
-          text: 'Describe the following image:',
-        },
-        {
-          type: 'image_url',
-          image_url: `data:image/png;base64,${imageData}`,
-        },
-      ],
-    }),
-  ]);
-  expect(res).toBeTruthy();
-});
-
-test('Test Google AI handleLLMNewToken callback', async () => {
-  // Running LangChain callbacks in the background will sometimes cause the callbackManager to execute
-  // after the test/llm call has already finished & returned. Set that environment variable to false
-  // to prevent that from happening.
-  process.env.LANGCHAIN_CALLBACKS_BACKGROUND = 'false';
-
-  try {
-    const model = createGoogleModel({
-      model: 'gemini-2.0-flash',
-    });
-    let tokens = '';
-    const res = await model.invoke([new HumanMessage('what is 1 + 1?')], {
-      callbacks: [
-        {
-          handleLLMNewToken(token: string): void {
-            tokens += token;
-          },
-        },
-      ],
-    });
-    const responseContent = typeof res.content === 'string' ? res.content : '';
-    expect(tokens).toBe(responseContent);
-  } finally {
-    // Reset the environment variable
-    process.env.LANGCHAIN_CALLBACKS_BACKGROUND = originalBackground;
-  }
-});
-
-test('Test Google AI handleLLMNewToken callback with streaming', async () => {
-  // Running LangChain callbacks in the background will sometimes cause the callbackManager to execute
-  // after the test/llm call has already finished & returned. Set that environment variable to false
-  // to prevent that from happening.
-  process.env.LANGCHAIN_CALLBACKS_BACKGROUND = 'false';
-
-  try {
-    const model = createGoogleModel({
-      model: 'gemini-2.0-flash',
-    });
-    let tokens = '';
-    const res = await model.stream([new HumanMessage('what is 1 + 1?')], {
-      callbacks: [
-        {
-          handleLLMNewToken(token: string): void {
-            tokens += token;
-          },
-        },
-      ],
-    });
-    let responseContent = '';
-    for await (const streamItem of res) {
-      responseContent += streamItem.content;
-    }
-    expect(tokens).toBe(responseContent);
-  } finally {
-    // Reset the environment variable
-    process.env.LANGCHAIN_CALLBACKS_BACKGROUND = originalBackground;
-  }
-});
-
-test('Test Google AI in streaming mode', async () => {
-  // Running LangChain callbacks in the background will sometimes cause the callbackManager to execute
-  // after the test/llm call has already finished & returned. Set that environment variable to false
-  // to prevent that from happening.
-  process.env.LANGCHAIN_CALLBACKS_BACKGROUND = 'false';
-
-  try {
-    const model = createGoogleModel({
-      model: 'gemini-2.0-flash',
-      streaming: true,
-    });
-    let tokens = '';
-    let nrNewTokens = 0;
-    const res = await model.invoke([new HumanMessage('Write a haiku?')], {
-      callbacks: [
-        {
-          handleLLMNewToken(token: string): void {
-            nrNewTokens += 1;
-            tokens += token;
-          },
-        },
-      ],
-    });
-    expect(nrNewTokens).toBeGreaterThanOrEqual(1);
-    expect(res.content).toBe(tokens);
-  } finally {
-    // Reset the environment variable
-    process.env.LANGCHAIN_CALLBACKS_BACKGROUND = originalBackground;
-  }
-});
-
-async function fileToBase64(filePath: string): Promise<string> {
-  const fileData = await fs.readFile(filePath);
-  // eslint-disable-next-line @typescript-eslint/ban-ts-comment
-  /** @ts-ignore */
-  const base64String = Buffer.from(fileData).toString('base64');
-  return base64String;
-}
-
-test('Gemini can understand audio', async () => {
-  // Update this with the correct path to an audio file on your machine.
-  const audioPath = path.join(__dirname, 'data/gettysburg10.wav');
-  const audioMimeType = 'audio/wav';
-
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-    temperature: 0,
-  });
-
-  const audioBase64 = await fileToBase64(audioPath);
-
-  const prompt = ChatPromptTemplate.fromMessages([
-    new MessagesPlaceholder('audio'),
-  ]);
-
-  const chain = prompt.pipe(model);
-  const response = await chain.invoke({
-    audio: new HumanMessage({
-      content: [
-        {
-          type: 'media',
-          mimeType: audioMimeType,
-          data: audioBase64,
-        },
-        {
-          type: 'text',
-          text: "Summarize the content in this audio. ALso, what is the speaker's tone?",
-        },
-      ],
-    }),
-  });
-
-  expect(typeof response.content).toBe('string');
-  expect((response.content as string).length).toBeGreaterThan(15);
-});
-
-class FakeBrowserTool extends StructuredTool {
-  schema = z.object({
-    url: z.string(),
-    query: z.string().optional(),
-  });
-
-  name = 'fake_browser_tool';
-
-  description =
-    'useful for when you need to find something on the web or summarize a webpage.';
-
-  async _call(_: z.infer<this['schema']>): Promise<string> {
-    return 'fake_browser_tool';
-  }
-}
-const googleGenAITool = {
-  functionDeclarations: [
-    {
-      name: 'fake_browser_tool',
-      description:
-        'useful for when you need to find something on the web or summarize a webpage.',
-      parameters: {
-        type: FunctionDeclarationSchemaType.OBJECT,
-        required: ['url'],
-        properties: {
-          url: {
-            type: FunctionDeclarationSchemaType.STRING,
-          },
-          query: {
-            type: FunctionDeclarationSchemaType.STRING,
-          },
-        },
-      },
-    },
-  ],
-};
-const prompt = new HumanMessage(
-  'Search the web and tell me what the weather will be like tonight in new york. use weather.com'
-);
-
-test('ChatGoogleGenerativeAI can bind and invoke langchain tools', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const modelWithTools = model.bindTools([new FakeBrowserTool()]);
-  const res = await modelWithTools.invoke([prompt]);
-  const toolCalls = res.tool_calls;
-  expect(toolCalls).toBeDefined();
-  if (!toolCalls) {
-    throw new Error('tool_calls not in response');
-  }
-  expect(toolCalls.length).toBe(1);
-  expect(toolCalls[0].name).toBe('fake_browser_tool');
-  expect('url' in toolCalls[0].args).toBe(true);
-});
-
-test('ChatGoogleGenerativeAI can bind and stream langchain tools', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const modelWithTools = model.bindTools([new FakeBrowserTool()]);
-  let finalChunk: AIMessageChunk | undefined;
-  for await (const chunk of await modelWithTools.stream([prompt])) {
-    if (!finalChunk) {
-      finalChunk = chunk;
-    } else {
-      finalChunk = finalChunk.concat(chunk);
-    }
-  }
-  if (!finalChunk) {
-    throw new Error('finalChunk is undefined');
-  }
-  const toolCalls = finalChunk.tool_calls;
-  expect(toolCalls).toBeDefined();
-  if (!toolCalls) {
-    throw new Error('tool_calls not in response');
-  }
-  expect(toolCalls.length).toBe(1);
-  expect(toolCalls[0].name).toBe('fake_browser_tool');
-  expect(toolCalls[0].id).toBeDefined();
-  expect('url' in toolCalls[0].args).toBe(true);
-});
-
-test('ChatGoogleGenerativeAI can handle streaming tool messages.', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const browserTool = new FakeBrowserTool();
-
-  const modelWithTools = model.bindTools([browserTool]);
-  let finalChunk: AIMessageChunk | undefined;
-  const fullPrompt = [
-    new SystemMessage(
-      'You are a helpful assistant. If the chat history contains the tool results, you should use that and not call the tool again.'
-    ),
-    prompt,
-    new AIMessage({
-      content: '',
-      tool_calls: [
-        {
-          name: browserTool.name,
-          args: {
-            query: 'weather tonight new york',
-            url: 'https://weather.com',
-          },
-        },
-      ],
-    }),
-    new ToolMessage(dummyToolResponse, 'id', browserTool.name),
-  ];
-  for await (const chunk of await modelWithTools.stream(fullPrompt)) {
-    if (!finalChunk) {
-      finalChunk = chunk;
-    } else {
-      finalChunk = finalChunk.concat(chunk);
-    }
-  }
-  if (!finalChunk) {
-    throw new Error('finalChunk is undefined');
-  }
-  expect(typeof finalChunk.content).toBe('string');
-  expect(finalChunk.content.length).toBeGreaterThan(1);
-  expect(finalChunk.tool_calls).toHaveLength(0);
-});
-
-test('ChatGoogleGenerativeAI can handle invoking tool messages.', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const browserTool = new FakeBrowserTool();
-
-  const modelWithTools = model.bindTools([browserTool]);
-  const fullPrompt = [
-    new SystemMessage(
-      'You are a helpful assistant. If the chat history contains the tool results, you should use that and not call the tool again.'
-    ),
-    prompt,
-    new AIMessage({
-      content: '',
-      tool_calls: [
-        {
-          name: browserTool.name,
-          args: {
-            query: 'weather tonight new york',
-            url: 'https://weather.com',
-          },
-        },
-      ],
-    }),
-    new ToolMessage(dummyToolResponse, 'id', browserTool.name),
-  ];
-  const response = await modelWithTools.invoke(fullPrompt);
-  expect(typeof response.content).toBe('string');
-  expect(response.content.length).toBeGreaterThan(1);
-  expect(response.tool_calls).toHaveLength(0);
-});
-
-test('ChatGoogleGenerativeAI can bind and invoke genai tools', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const modelWithTools = model.bindTools([googleGenAITool]);
-  const res = await modelWithTools.invoke([prompt]);
-  const toolCalls = res.tool_calls;
-  expect(toolCalls).toBeDefined();
-  if (!toolCalls) {
-    throw new Error('tool_calls not in response');
-  }
-  expect(toolCalls.length).toBe(1);
-  expect(toolCalls[0].name).toBe('fake_browser_tool');
-  expect('url' in toolCalls[0].args).toBe(true);
-});
-
-test('ChatGoogleGenerativeAI can bindTools with langchain tools and invoke', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const modelWithTools = model.bindTools([new FakeBrowserTool()]);
-  const res = await modelWithTools.invoke([prompt]);
-  const toolCalls = res.tool_calls;
-  expect(toolCalls).toBeDefined();
-  if (!toolCalls) {
-    throw new Error('tool_calls not in response');
-  }
-  expect(toolCalls.length).toBe(1);
-  expect(toolCalls[0].name).toBe('fake_browser_tool');
-  expect('url' in toolCalls[0].args).toBe(true);
-});
-
-test('ChatGoogleGenerativeAI can bindTools with genai tools and invoke', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.5-flash',
-  });
-
-  const modelWithTools = model.bindTools([googleGenAITool]);
-  const res = await modelWithTools.invoke([prompt]);
-  const toolCalls = res.tool_calls;
-  expect(toolCalls).toBeDefined();
-  if (!toolCalls) {
-    throw new Error('tool_calls not in response');
-  }
-  expect(toolCalls.length).toBe(1);
-  expect(toolCalls[0].name).toBe('fake_browser_tool');
-  expect('url' in toolCalls[0].args).toBe(true);
-});
-
-test('ChatGoogleGenerativeAI can call withStructuredOutput langchain tools and invoke', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-
-  const modelWithTools = model.withStructuredOutput(
-    z.object({
-      zomg: z.string(),
-      omg: z.number().optional(),
-    })
-  );
-  const res = await modelWithTools.invoke([prompt]);
-  expect(typeof res.zomg === 'string').toBe(true);
-});
-
-test('ChatGoogleGenerativeAI can call withStructuredOutput genai tools and invoke', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-
-  type GeminiTool = {
-    url: string;
-    query?: string;
-  };
-
-  const modelWithTools = model.withStructuredOutput<GeminiTool>(
-    googleGenAITool.functionDeclarations[0].parameters
-  );
-  const res = await modelWithTools.invoke([prompt]);
-  expect(typeof res.url === 'string').toBe(true);
-});
-
-test('Stream token count usage_metadata', async () => {
-  const model = createGoogleModel({
-    temperature: 0,
-    model: 'gemini-2.0-flash',
-    maxOutputTokens: 10,
-  });
-  let res: AIMessageChunk | null = null;
-  for await (const chunk of await model.stream(
-    'Why is the sky blue? Be concise.'
-  )) {
-    if (!res) {
-      res = chunk;
-    } else {
-      res = res.concat(chunk);
-    }
-  }
-  expect(res?.usage_metadata).toBeDefined();
-  if (!res?.usage_metadata) {
-    return;
-  }
-  expect(res.usage_metadata.input_tokens).toBeGreaterThan(1);
-  expect(res.usage_metadata.output_tokens).toBeGreaterThan(1);
-  expect(res.usage_metadata.total_tokens).toBe(
-    res.usage_metadata.input_tokens + res.usage_metadata.output_tokens
-  );
-});
-
-describe('ChatGoogleGenerativeAI should count tokens correctly', () => {
-  describe('when streaming', () => {
-    test.each(['gemini-2.5-flash', 'gemini-2.5-pro'])(
-      'with %s',
-      async (modelName) => {
-        const model = createGoogleModel({
-          model: modelName,
-          temperature: 0,
-        });
-        const res = await model.stream('Why is the sky blue? Be concise.');
-        let full: AIMessageChunk | undefined;
-        for await (const chunk of res) {
-          full ??= chunk;
-          full = full.concat(chunk);
-        }
-        // expect(full?.usage_metadata);
-        // expect(res.usage_metadata).toBeDefined();
-      }
-    );
-  });
-});
-
-test('streamUsage excludes token usage', async () => {
-  const model = createGoogleModel({
-    temperature: 0,
-    model: 'gemini-2.0-flash',
-    streamUsage: false,
-  });
-  let res: AIMessageChunk | null = null;
-  for await (const chunk of await model.stream(
-    'Why is the sky blue? Be concise.'
-  )) {
-    if (!res) {
-      res = chunk;
-    } else {
-      res = res.concat(chunk);
-    }
-  }
-  expect(res?.usage_metadata).not.toBeDefined();
-});
-
-test('Invoke token count usage_metadata', async () => {
-  const model = createGoogleModel({
-    temperature: 0,
-    model: 'gemini-2.0-flash',
-    maxOutputTokens: 10,
-  });
-  const res = await model.invoke('Why is the sky blue? Be concise.');
-  expect(res?.usage_metadata).toBeDefined();
-  if (!res?.usage_metadata) {
-    return;
-  }
-  expect(res.usage_metadata.input_tokens).toBeGreaterThan(1);
-  expect(res.usage_metadata.output_tokens).toBeGreaterThan(1);
-  expect(res.usage_metadata.total_tokens).toBe(
-    res.usage_metadata.input_tokens + res.usage_metadata.output_tokens
-  );
-});
-
-test('Invoke with JSON mode', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-    temperature: 0,
-    maxOutputTokens: 10,
-    json: true,
-  });
-  const res = await model.invoke('Why is the sky blue? Be concise.');
-  expect(res?.usage_metadata).toBeDefined();
-  if (!res?.usage_metadata) {
-    return;
-  }
-  expect(res.usage_metadata.input_tokens).toBeGreaterThan(1);
-  expect(res.usage_metadata.output_tokens).toBeGreaterThan(1);
-  expect(res.usage_metadata.total_tokens).toBe(
-    res.usage_metadata.input_tokens + res.usage_metadata.output_tokens
-  );
-});
-
-test('Includes server-side tool invocation config when enabled', () => {
-  const model = new ChatGoogleGenerativeAI({
-    apiKey: 'test-key',
-    model: 'gemini-3.5-flash',
-    includeServerSideToolInvocations: true,
-  });
-  const getWeather = {
-    name: 'get_weather',
-    description: 'Get the weather',
-  };
-
-  const request = model.invocationParams({
-    tools: [{ googleSearch: {} }, { functionDeclarations: [getWeather] }],
-  });
-
-  expect(request.toolConfig).toEqual(
-    expect.objectContaining({
-      includeServerSideToolInvocations: true,
-    })
-  );
-});
-
-test('Normalizes auto tool choice when server-side tool invocation config is enabled', () => {
-  const model = new ChatGoogleGenerativeAI({
-    apiKey: 'test-key',
-    model: 'gemini-3.5-flash',
-    includeServerSideToolInvocations: true,
-  });
-  const getWeather = {
-    name: 'get_weather',
-    description: 'Get the weather',
-  };
-
-  const request = model.invocationParams({
-    tools: [{ googleSearch: {} }, { functionDeclarations: [getWeather] }],
-    tool_choice: 'auto',
-  });
-  const toolConfig = request.toolConfig as
-    | {
-        includeServerSideToolInvocations?: boolean;
-        functionCallingConfig?: { mode?: string };
-      }
-    | undefined;
-
-  expect(toolConfig).toEqual(
-    expect.objectContaining({
-      includeServerSideToolInvocations: true,
-    })
-  );
-  expect(toolConfig?.functionCallingConfig?.mode).toBe('VALIDATED');
-});
-
-test('Preserves Gemini server-side tool context parts in history', () => {
-  const toolCallPart = {
-    type: 'toolCall',
-    agentId: 'agent_1',
-    groupId: 1,
-    thoughtSignature: 'signature-1',
-    toolCall: {
-      id: 'server-search-1',
-      name: 'google_search',
-      args: {},
-    },
-  } as const;
-  const toolResponsePart = {
-    type: 'toolResponse',
-    agentId: 'agent_1',
-    groupId: 1,
-    toolResponse: {
-      id: 'server-search-1',
-      name: 'google_search',
-      response: { results: [] },
-    },
-  } as const;
-  const message = new AIMessage({
-    content: [toolCallPart, toolResponsePart],
-  });
-
-  expect(convertMessageContentToParts(message, true, [])).toEqual([
-    {
-      toolCall: toolCallPart.toolCall,
-      thoughtSignature: 'signature-1',
-    },
-    {
-      toolResponse: toolResponsePart.toolResponse,
-    },
-  ]);
-});
-
-test('Preserves Gemini function-call ids in history', () => {
-  const aiMessage = new AIMessage({
-    content: '',
-    tool_calls: [
-      {
-        id: 'call_weather_1',
-        name: 'get_weather',
-        args: { city: 'NYC' },
-      },
-    ],
-  });
-  const toolMessage = new ToolMessage({
-    content: 'sunny',
-    name: 'get_weather',
-    tool_call_id: 'call_weather_1',
-  });
-
-  expect(convertMessageContentToParts(aiMessage, true, [])).toEqual([
-    {
-      functionCall: {
-        id: 'call_weather_1',
-        name: 'get_weather',
-        args: { city: 'NYC' },
-      },
-    },
-  ]);
-  expect(convertMessageContentToParts(toolMessage, true, [aiMessage])).toEqual([
-    {
-      functionResponse: {
-        id: 'call_weather_1',
-        name: 'get_weather',
-        response: { result: 'sunny' },
-      },
-    },
-  ]);
-});
-
-test('Preserves Gemini server-side tool context parts from responses', () => {
-  const toolCallPart = {
-    toolCall: {
-      id: 'server-search-1',
-      name: 'google_search',
-      args: {},
-    },
-  };
-  const toolResponsePart = {
-    toolResponse: {
-      id: 'server-search-1',
-      name: 'google_search',
-      response: { results: [] },
-    },
-  };
-  const response = {
-    candidates: [
-      {
-        content: {
-          role: 'model',
-          parts: [toolCallPart, toolResponsePart],
-        },
-      },
-    ],
-  } as unknown as Parameters<
-    typeof convertResponseContentToChatGenerationChunk
-  >[0];
-
-  const chunk = convertResponseContentToChatGenerationChunk(response, {
-    index: 0,
-  });
-  const result = mapGenerateContentResultToChatResult(response);
-
-  const expectedContent = [
-    { ...toolCallPart, type: 'toolCall' },
-    { ...toolResponsePart, type: 'toolResponse' },
-  ];
-  expect(chunk?.message.content).toEqual(expectedContent);
-  expect(result.generations[0].message.content).toEqual(expectedContent);
-});
-
-test('Supports tool_choice', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-  const tools = [
-    {
-      name: 'get_weather',
-      description: 'Get the weather',
-      schema: z.object({
-        location: z.string(),
-      }),
-    },
-    {
-      name: 'calculator',
-      description: 'Preform calculations',
-      schema: z.object({
-        expression: z.string(),
-      }),
-    },
-  ];
-
-  const modelWithTools = model.bindTools(tools, {
-    tool_choice: 'calculator',
-    allowedFunctionNames: ['calculator'],
-  });
-  const response = await modelWithTools.invoke(
-    'What is 27725327 times 283683? Also whats the weather in New York?'
-  );
-  expect(response.tool_calls?.length).toBe(1);
-});
-
-describe('GoogleSearchRetrievalTool', () => {
-  test('Supports GoogleSearchRetrievalTool', async () => {
-    const searchRetrievalTool: GoogleSearchRetrievalTool = {
-      googleSearchRetrieval: {
-        dynamicRetrievalConfig: {
-          mode: DynamicRetrievalMode.MODE_DYNAMIC,
-          dynamicThreshold: 0.7,
-        },
-      },
-    };
-    const model = createGoogleModel({
-      model: 'gemini-1.5-pro',
-      temperature: 0,
-    }).bindTools([searchRetrievalTool]);
-
-    const result = await model.invoke('Who won the 2024 MLB World Series?');
-
-    expect(result.response_metadata?.groundingMetadata).toBeDefined();
-    expect(result.content as string).toContain('Dodgers');
-  });
-
-  test('Can stream GoogleSearchRetrievalTool', async () => {
-    const searchRetrievalTool: GoogleSearchRetrievalTool = {
-      googleSearchRetrieval: {
-        dynamicRetrievalConfig: {
-          mode: DynamicRetrievalMode.MODE_DYNAMIC,
-          dynamicThreshold: 0.7,
-        },
-      },
-    };
-    const model = createGoogleModel({
-      model: 'gemini-1.5-pro',
-      temperature: 0,
-    }).bindTools([searchRetrievalTool]);
-
-    const stream = await model.stream('Who won the 2024 MLB World Series?');
-    let finalMsg: AIMessageChunk | undefined;
-    for await (const msg of stream) {
-      finalMsg = finalMsg ? concat(finalMsg, msg) : msg;
-    }
-    if (!finalMsg) {
-      throw new Error('finalMsg is undefined');
-    }
-    expect(finalMsg.response_metadata?.groundingMetadata).toBeDefined();
-    expect(finalMsg.content as string).toContain('Dodgers');
-  });
-});
-
-describe('GoogleSearch (new API)', () => {
-  test('Supports GoogleSearch tool', async () => {
-    // New google_search tool for Gemini 2.0+ models
-    const googleSearchTool = {
-      googleSearch: {},
-    };
-    const model = createGoogleModel({
-      model: 'gemini-2.5-flash',
-      temperature: 0,
-    }).bindTools([googleSearchTool]);
-
-    // Ask about something that requires current web data beyond training cutoff
-    const result = await model.invoke(
-      'What was the closing price of NVIDIA stock yesterday? Use web search to find the exact current price.'
-    );
-
-    expect(result.content).toBeDefined();
-    expect(
-      typeof result.content === 'string' || Array.isArray(result.content)
-    ).toBe(true);
-
-    // Grounding metadata should be present when Google Search is used
-    expect(result.response_metadata?.groundingMetadata).toBeDefined();
-    expect(result.response_metadata.groundingMetadata).toHaveProperty(
-      'groundingChunks'
-    );
-    expect(result.response_metadata.groundingMetadata).toHaveProperty(
-      'webSearchQueries'
-    );
-  });
-
-  test('Can stream GoogleSearch tool', async () => {
-    const googleSearchTool = {
-      googleSearch: {},
-    };
-    const model = createGoogleModel({
-      model: 'gemini-2.5-flash',
-      temperature: 0,
-    }).bindTools([googleSearchTool]);
-
-    const stream = await model.stream(
-      'What was the closing price of NVIDIA stock yesterday? Use web search to find the exact current price.'
-    );
-    let finalMsg: AIMessageChunk | undefined;
-    for await (const msg of stream) {
-      finalMsg = finalMsg ? concat(finalMsg, msg) : msg;
-    }
-    if (!finalMsg) {
-      throw new Error('finalMsg is undefined');
-    }
-
-    expect(finalMsg.content).toBeDefined();
-    expect(
-      typeof finalMsg.content === 'string' || Array.isArray(finalMsg.content)
-    ).toBe(true);
-
-    // Grounding metadata should be present when Google Search is used
-    expect(finalMsg.response_metadata?.groundingMetadata).toBeDefined();
-    expect(finalMsg.response_metadata.groundingMetadata).toHaveProperty(
-      'groundingChunks'
-    );
-    expect(finalMsg.response_metadata.groundingMetadata).toHaveProperty(
-      'webSearchQueries'
-    );
-  });
-});
-
-describe('CodeExecutionTool', () => {
-  test('Supports CodeExecutionTool', async () => {
-    const codeExecutionTool: CodeExecutionTool = {
-      codeExecution: {}, // Simply pass an empty object to enable it.
-    };
-    const model = createGoogleModel({
-      model: 'gemini-2.5-flash',
-      temperature: 0,
-    }).bindTools([codeExecutionTool]);
-
-    const result = await model.invoke(
-      'Use code execution to find the sum of the first and last 3 numbers in the following list: [1, 2, 3, 72638, 8, 727, 4, 5, 6]'
-    );
-
-    expect(Array.isArray(result.content)).toBeTruthy();
-    if (!Array.isArray(result.content)) {
-      throw new Error('Content is not an array');
-    }
-    const texts = result.content
-      .flatMap((item) => ('text' in item ? [item.text] : []))
-      .join('\n');
-    expect(texts).toContain('21');
-
-    const executableCode = result.content.find(
-      (item) => item.type === 'executableCode'
-    );
-    expect(executableCode).toBeDefined();
-    const codeResult = result.content.find(
-      (item) => item.type === 'codeExecutionResult'
-    );
-    expect(codeResult).toBeDefined();
-  });
-
-  test('CodeExecutionTool contents can be passed in chat history', async () => {
-    const codeExecutionTool: CodeExecutionTool = {
-      codeExecution: {}, // Simply pass an empty object to enable it.
-    };
-    const model = createGoogleModel({
-      model: 'gemini-2.5-flash',
-      temperature: 0,
-    }).bindTools([codeExecutionTool]);
-
-    const codeResult = await model.invoke(
-      'Use code execution to find the sum of the first and last 3 numbers in the following list: [1, 2, 3, 72638, 8, 727, 4, 5, 6]'
-    );
-
-    const explanation = await model.invoke([
-      codeResult,
-      {
-        role: 'user',
-        content:
-          'Please explain the question I asked, the code you wrote, and the answer you got.',
-      },
-    ]);
-
-    // Content can be string or array depending on response format
-    if (typeof explanation.content === 'string') {
-      expect(explanation.content.length).toBeGreaterThan(10);
-    } else if (Array.isArray(explanation.content)) {
-      expect(explanation.content.length).toBeGreaterThan(0);
-    } else {
-      expect(explanation.content).toBeDefined();
-    }
-  });
-
-  test('Can stream CodeExecutionTool', async () => {
-    const codeExecutionTool: CodeExecutionTool = {
-      codeExecution: {}, // Simply pass an empty object to enable it.
-    };
-    const model = createGoogleModel({
-      model: 'gemini-2.5-flash',
-      temperature: 0,
-    }).bindTools([codeExecutionTool]);
-
-    const stream = await model.stream(
-      'Use code execution to find the sum of the first and last 3 numbers in the following list: [1, 2, 3, 72638, 8, 727, 4, 5, 6]'
-    );
-    let finalMsg: AIMessageChunk | undefined;
-    for await (const msg of stream) {
-      finalMsg = finalMsg ? concat(finalMsg, msg) : msg;
-    }
-
-    if (!finalMsg) {
-      throw new Error('finalMsg is undefined');
-    }
-    expect(Array.isArray(finalMsg.content)).toBeTruthy();
-    if (!Array.isArray(finalMsg.content)) {
-      throw new Error('Content is not an array');
-    }
-    const texts = finalMsg.content
-      .flatMap((item) => ('text' in item ? [item.text] : []))
-      .join('\n');
-    expect(texts).toContain('21');
-
-    const executableCode = finalMsg.content.find(
-      (item) => item.type === 'executableCode'
-    );
-    expect(executableCode).toBeDefined();
-    const codeResult = finalMsg.content.find(
-      (item) => item.type === 'codeExecutionResult'
-    );
-    expect(codeResult).toBeDefined();
-  });
-});
-
-test('pass pdf to request', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-2.0-flash-exp',
-    temperature: 0,
-  });
-  const pdfPath = path.join(
-    __dirname,
-    '../anthropic/Jacob_Lee_Resume_2023.pdf'
-  );
-  const pdfBase64 = await fs.readFile(pdfPath, 'base64');
-
-  const response = await model.invoke([
-    ['system', 'Use the provided documents to answer the question'],
-    [
-      'user',
-      [
-        {
-          type: 'application/pdf',
-          data: pdfBase64,
-        },
-        {
-          type: 'text',
-          text: 'Summarize the contents of this PDF',
-        },
-      ],
-    ],
-  ]);
-
-  expect(response.content.length).toBeGreaterThan(10);
-});
-
-test('calling tool with no args should work', async () => {
-  const llm = createGoogleModel({
-    model: 'gemini-2.0-flash',
-  });
-  const sfWeatherTool = tool(
-    async () => 'The weather is 80 degrees and sunny',
-    {
-      name: 'sf_weather',
-      description: 'Gets the weather in SF',
-      schema: z.object({}),
-    }
-  );
-  const llmWithTools = llm.bindTools([sfWeatherTool]);
-  const result = await llmWithTools.invoke([
-    {
-      role: 'user',
-      content: 'What is the current weather in SF?',
-    },
-  ]);
-  const nextMessage = await sfWeatherTool.invoke(result.tool_calls![0]);
-  delete nextMessage.name; // Should work even if name is not present
-  const finalResult = await llmWithTools.invoke([
-    {
-      role: 'user',
-      content: 'What is the current weather in SF?',
-    },
-    result,
-    nextMessage,
-  ]);
-  expect(finalResult.content).toContain('80');
-});
-
-describe('tool calling with thought signatures', () => {
-  const model = createGoogleModel({
-    model: 'gemini-3-pro-preview',
-  });
-  const weatherTool = tool(async () => 'The weather is 80 degrees and sunny', {
-    name: 'weather',
-    description: 'Gets the weather in SF',
-    schema: z.object({}),
-  });
-  const modelWithTools = model.bindTools([weatherTool]);
-
-  test('works when invoking', async () => {
-    const result = await modelWithTools.invoke(
-      'What is the current weather in SF?'
-    );
-    expect(result.tool_calls).toBeDefined();
-    expect(result.tool_calls!.length).toBe(1);
-    expect(result.tool_calls![0].id).toBeDefined();
-
-    const toolMessage = new ToolMessage({
-      content: 'The weather is 80 degrees and sunny',
-      tool_call_id: result.tool_calls![0].id ?? '',
-    });
-
-    // Thought signatures are stored in additional_kwargs
-    const thoughtSignatures = result.additional_kwargs?.[
-      _FUNCTION_CALL_THOUGHT_SIGNATURES_MAP_KEY
-    ] as Record<string, string> | undefined;
-
-    // Gemini 3 REQUIRES thought signatures - not optional
-    expect(thoughtSignatures).toBeDefined();
-    expect(thoughtSignatures![result.tool_calls![0].id ?? '']).toBeDefined();
-
-    const finalResult = await model.invoke([
-      new HumanMessage('What is the current weather in SF?'),
-      result,
-      toolMessage,
-    ]);
-    expect(finalResult.content).toBeDefined();
-  });
-
-  test('works when streaming', async () => {
-    let finalChunk: AIMessageChunk | undefined;
-    for await (const chunk of await modelWithTools.stream(
-      'What is the current weather in SF?'
-    )) {
-      finalChunk = finalChunk ? finalChunk.concat(chunk) : chunk;
-    }
-    expect(finalChunk).toBeDefined();
-    expect(finalChunk?.tool_calls).toBeDefined();
-    expect(finalChunk?.tool_calls!.length).toBe(1);
-    const toolMessage = new ToolMessage({
-      content: 'The weather is 80 degrees and sunny',
-      tool_call_id: finalChunk?.tool_calls![0].id ?? '',
-    });
-
-    // Thought signatures are stored in additional_kwargs
-    const thoughtSignatures = finalChunk?.additional_kwargs[
-      _FUNCTION_CALL_THOUGHT_SIGNATURES_MAP_KEY
-    ] as Record<string, string> | undefined;
-
-    // Only check if thought signatures exist (may not be present for all models/responses)
-    if (thoughtSignatures) {
-      expect(
-        thoughtSignatures[finalChunk?.tool_calls![0].id ?? '']
-      ).toBeDefined();
-    }
-
-    const finalResult = await model.invoke([
-      new HumanMessage('What is the current weather in SF?'),
-      finalChunk!,
-      toolMessage,
-    ]);
-    expect(finalResult.content).toBeDefined();
-  });
-});
-
-test('works with thinking config', async () => {
-  const model = createGoogleModel({
-    model: 'gemini-3-pro-preview',
-    thinkingConfig: {
-      includeThoughts: true,
-      thinkingBudget: 100,
-    },
-  });
-  const result = await model.invoke('What is 2+2?');
-  expect(result.content).toBeDefined();
-
-  if (Array.isArray(result.content)) {
-    const thinkingBlocks = result.content.filter(
-      (block): block is ContentBlock =>
-        typeof block === 'object' &&
-        block !== null &&
-        'type' in block &&
-        block.type === 'thinking'
-    );
-    const textBlocks = result.content.filter(
-      (block): block is ContentBlock =>
-        typeof block === 'object' &&
-        block !== null &&
-        'type' in block &&
-        block.type === 'text'
-    );
-
-    expect(thinkingBlocks.length).toBeGreaterThan(0);
-
-    thinkingBlocks.forEach((block) => {
-      expect(block).toHaveProperty('thinking');
-      expect(typeof block.thinking).toBe('string');
-    });
-
-    textBlocks.forEach((block) => {
-      expect(block).toHaveProperty('text');
-      expect(typeof block.text).toBe('string');
-    });
-  }
-});
-
-describe('Google GenAI Reasoning with contentBlocks', () => {
-  test('invoke returns thinking as reasoning in contentBlocks', async () => {
-    const model = createGoogleModel({
-      model: 'gemini-3-pro-preview',
-      thinkingConfig: {
-        includeThoughts: true,
-        thinkingBudget: 100,
-      },
-    });
-
-    const result = await model.invoke('What is 2 + 2?');
-    const blocks = result.contentBlocks;
-
-    expect(blocks.length).toBeGreaterThan(0);
-
-    const reasoningBlocks = blocks.filter(
-      (block): block is ContentBlock.Reasoning => block.type === 'reasoning'
-    );
-    expect(reasoningBlocks.length).toBeGreaterThan(0);
-    expect(reasoningBlocks[0].reasoning.length).toBeGreaterThan(0);
-
-    const textBlocks = blocks.filter((block) => block.type === 'text');
-    expect(textBlocks.length).toBeGreaterThan(0);
-  });
-
-  test('stream returns thinking as reasoning in contentBlocks', async () => {
-    const model = createGoogleModel({
-      model: 'gemini-3-pro-preview',
-      thinkingConfig: {
-        includeThoughts: true,
-        thinkingBudget: 100,
-      },
-    });
-
-    let fullMessage: AIMessageChunk | null = null;
-    for await (const chunk of await model.stream('What is 3 + 3?')) {
-      fullMessage = fullMessage ? concat(fullMessage, chunk) : chunk;
-    }
-
-    expect(fullMessage).toBeDefined();
-
-    const blocks = fullMessage!.contentBlocks;
-    expect(blocks.length).toBeGreaterThan(0);
-
-    const reasoningBlocks = blocks.filter(
-      (block): block is ContentBlock.Reasoning => block.type === 'reasoning'
-    );
-    expect(reasoningBlocks.length).toBeGreaterThan(0);
-    expect(reasoningBlocks[0].reasoning.length).toBeGreaterThan(0);
-  });
-});