@lobehub/chat 1.60.8 → 1.61.0

package/locales/zh-CN/models.json

@@ -371,9 +371,6 @@
   "abab6.5t-chat": {
     "description": "针对中文人设对话场景优化，提供流畅且符合中文表达习惯的对话生成能力。"
   },
-  "abab7-chat-preview": {
-    "description": "相对于abab6.5系列模型在长文、数学、写作等能力有大幅度提升。"
-  },
   "accounts/fireworks/models/deepseek-r1": {
     "description": "DeepSeek-R1 是一款最先进的大型语言模型，经过强化学习和冷启动数据的优化，具有出色的推理、数学和编程性能。"
   },
@@ -728,6 +725,9 @@
   "gemini-1.0-pro-latest": {
     "description": "Gemini 1.0 Pro 是Google的高性能AI模型，专为广泛任务扩展而设计。"
   },
+  "gemini-1.5-flash": {
+    "description": "Gemini 1.5 Flash 是Google最新的多模态AI模型，具备快速处理能力，支持文本、图像和视频输入，适用于多种任务的高效扩展。"
+  },
   "gemini-1.5-flash-001": {
     "description": "Gemini 1.5 Flash 001 是一款高效的多模态模型，支持广泛应用的扩展。"
   },
@@ -767,24 +767,15 @@
   "gemini-2.0-flash-001": {
     "description": "Gemini 2.0 Flash 提供下一代功能和改进，包括卓越的速度、原生工具使用、多模态生成和1M令牌上下文窗口。"
   },
-  "gemini-2.0-flash-exp": {
-    "description": "Gemini 2.0 Flash Exp 是 Google 的实验性多模态AI模型，拥有下一代特性，卓越的速度，原生工具调用以及多模态生成。"
-  },
   "gemini-2.0-flash-lite-preview-02-05": {
     "description": "一个 Gemini 2.0 Flash 模型，针对成本效益和低延迟等目标进行了优化。"
   },
   "gemini-2.0-flash-thinking-exp-01-21": {
     "description": "Gemini 2.0 Flash Thinking Exp 是 Google 的实验性多模态推理AI模型，能对复杂问题进行推理，拥有新的思维能力。"
   },
-  "gemini-2.0-flash-thinking-exp-1219": {
-    "description": "Gemini 2.0 Flash Thinking Exp 是 Google 的实验性多模态推理AI模型，能对复杂问题进行推理，拥有新的思维能力。"
-  },
   "gemini-2.0-pro-exp-02-05": {
     "description": "Gemini 2.0 Pro Experimental 是 Google 最新的实验性多模态AI模型，与历史版本相比有一定的质量提升，特别是对于世界知识、代码和长上下文。"
   },
-  "gemini-exp-1206": {
-    "description": "Gemini Exp 1206 是 Google 的实验性多模态AI模型，与历史版本相比有一定的质量提升。"
-  },
   "gemma-7b-it": {
     "description": "Gemma 7B 适合中小规模任务处理，兼具成本效益。"
   },

package/locales/zh-CN/providers.json

@@ -119,6 +119,9 @@
   "upstage": {
     "description": "Upstage 专注于为各种商业需求开发AI模型，包括 Solar LLM 和文档 AI，旨在实现工作的人造通用智能（AGI）。通过 Chat API 创建简单的对话代理，并支持功能调用、翻译、嵌入以及特定领域应用。"
   },
+  "vertexai": {
+    "description": "Google 的 Gemini 系列是其最先进、通用的 AI模型，由 Google DeepMind 打造，专为多模态设计，支持文本、代码、图像、音频和视频的无缝理解与处理。适用于从数据中心到移动设备的多种环境，极大提升了AI模型的效率与应用广泛性。"
+  },
   "vllm": {
     "description": "vLLM 是一个快速且易于使用的库，用于 LLM 推理和服务。"
   },

package/locales/zh-TW/error.json

@@ -76,6 +76,7 @@
     "InvalidGithubToken": "Github 個人存取權杖不正確或為空，請檢查 Github 個人存取權杖後再試一次",
     "InvalidOllamaArgs": "Ollama 配置不正確，請檢查 Ollama 配置後重試",
     "InvalidProviderAPIKey": "{{provider}} API 金鑰不正確或為空，請檢查 {{provider}} API 金鑰後重試",
+    "InvalidVertexCredentials": "Vertex 認證未通過，請檢查認證憑證後重試",
     "LocationNotSupportError": "很抱歉，你的所在位置不支持此模型服務，可能是由於地區限制或服務未開通。請確認當前位置是否支持使用此服務，或嘗試使用其他位置信息。",
     "NoOpenAIAPIKey": "OpenAI API 金鑰為空，請添加自訂 OpenAI API 金鑰",
     "OllamaBizError": "請求 Ollama 服務出錯，請根據以下資訊排查或重試",

package/locales/zh-TW/modelProvider.json

@@ -322,6 +322,13 @@
     "tooltip": "更新服務商基礎配置",
     "updateSuccess": "更新成功"
   },
+  "vertexai": {
+    "apiKey": {
+      "desc": "填入你的 Vertex AI 金鑰",
+      "placeholder": "{ \"type\": \"service_account\", \"project_id\": \"xxx\", \"private_key_id\": ... }",
+      "title": "Vertex AI 金鑰"
+    }
+  },
   "zeroone": {
     "title": "01.AI 零一萬物"
   },

package/locales/zh-TW/models.json

@@ -371,9 +371,6 @@
   "abab6.5t-chat": {
     "description": "針對中文人設對話場景優化，提供流暢且符合中文表達習慣的對話生成能力。"
   },
-  "abab7-chat-preview": {
-    "description": "相對於abab6.5系列模型在長文、數學、寫作等能力有大幅度提升。"
-  },
   "accounts/fireworks/models/deepseek-r1": {
     "description": "DeepSeek-R1 是一款最先進的大型語言模型，經過強化學習和冷啟動數據的優化，具有出色的推理、數學和編程性能。"
   },
@@ -728,6 +725,9 @@
   "gemini-1.0-pro-latest": {
     "description": "Gemini 1.0 Pro 是 Google 的高性能 AI 模型，專為廣泛任務擴展而設計。"
   },
+  "gemini-1.5-flash": {
+    "description": "Gemini 1.5 Flash 是 Google 最新的多模態 AI 模型，具備快速處理能力，支持文本、圖像和視頻輸入，適用於多種任務的高效擴展。"
+  },
   "gemini-1.5-flash-001": {
     "description": "Gemini 1.5 Flash 001 是一款高效的多模態模型，支持廣泛應用的擴展。"
   },
@@ -767,24 +767,15 @@
   "gemini-2.0-flash-001": {
     "description": "Gemini 2.0 Flash 提供下一代功能和改進，包括卓越的速度、原生工具使用、多模態生成和1M令牌上下文窗口。"
   },
-  "gemini-2.0-flash-exp": {
-    "description": "Gemini 2.0 Flash Exp 是 Google 最新的實驗性多模態AI模型，擁有下一代特性，卓越的速度，原生工具調用以及多模態生成。"
-  },
   "gemini-2.0-flash-lite-preview-02-05": {
     "description": "一個 Gemini 2.0 Flash 模型，針對成本效益和低延遲等目標進行了優化。"
   },
   "gemini-2.0-flash-thinking-exp-01-21": {
     "description": "Gemini 2.0 Flash Exp 是 Google 最新的實驗性多模態AI模型，擁有下一代特性，卓越的速度，原生工具調用以及多模態生成。"
   },
-  "gemini-2.0-flash-thinking-exp-1219": {
-    "description": "Gemini 2.0 Flash Exp 是 Google 最新的實驗性多模態AI模型，擁有下一代特性，卓越的速度，原生工具調用以及多模態生成。"
-  },
   "gemini-2.0-pro-exp-02-05": {
     "description": "Gemini 2.0 Pro Experimental 是 Google 最新的實驗性多模態AI模型，與歷史版本相比有一定的質量提升，特別是對於世界知識、代碼和長上下文。"
   },
-  "gemini-exp-1206": {
-    "description": "Gemini Exp 1206 是 Google 最新的實驗性多模態 AI 模型，與歷史版本相比有一定的質量提升。"
-  },
   "gemma-7b-it": {
     "description": "Gemma 7B 適合中小規模任務處理，兼具成本效益。"
   },

package/locales/zh-TW/providers.json

@@ -119,6 +119,9 @@
   "upstage": {
     "description": "Upstage 專注於為各種商業需求開發 AI 模型，包括 Solar LLM 和文檔 AI，旨在實現工作的人工通用智能（AGI）。通過 Chat API 創建簡單的對話代理，並支持功能調用、翻譯、嵌入以及特定領域應用。"
   },
+  "vertexai": {
+    "description": "Google 的 Gemini 系列是其最先進、通用的 AI 模型，由 Google DeepMind 打造，專為多模態設計，支持文本、程式碼、圖像、音訊和視頻的無縫理解與處理。適用於從數據中心到行動裝置的多種環境，極大提升了 AI 模型的效率與應用廣泛性。"
+  },
   "vllm": {
     "description": "vLLM 是一個快速且易於使用的庫，用於 LLM 推理和服務。"
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.60.8",
+  "version": "1.61.0",
   "description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
   "keywords": [
     "framework",
@@ -118,6 +118,7 @@
     "@codesandbox/sandpack-react": "^2.19.10",
     "@cyntler/react-doc-viewer": "^1.17.0",
     "@electric-sql/pglite": "0.2.13",
+    "@google-cloud/vertexai": "^1.9.2",
     "@google/generative-ai": "^0.21.0",
     "@huggingface/inference": "^2.8.1",
     "@icons-pack/react-simple-icons": "9.6.0",

package/src/app/(backend)/webapi/chat/vertexai/route.ts

@@ -0,0 +1,35 @@
+import { checkAuth } from '@/app/(backend)/middleware/auth';
+import { AgentRuntime, ModelProvider } from '@/libs/agent-runtime';
+import { LobeVertexAI } from '@/libs/agent-runtime/vertexai';
+import { safeParseJSON } from '@/utils/safeParseJSON';
+
+import { POST as UniverseRoute } from '../[provider]/route';
+
+// due to the Chinese region does not support accessing Google
+// we need to use proxy to access it
+// refs: https://github.com/google/generative-ai-js/issues/29#issuecomment-1866246513
+// if (process.env.HTTP_PROXY_URL) {
+//   const { setGlobalDispatcher, ProxyAgent } = require('undici');
+//
+//   setGlobalDispatcher(new ProxyAgent({ uri: process.env.HTTP_PROXY_URL }));
+// }
+
+export const POST = checkAuth(async (req: Request, { jwtPayload }) =>
+  UniverseRoute(req, {
+    createRuntime: () => {
+      const googleAuthStr = jwtPayload.apiKey ?? process.env.VERTEXAI_CREDENTIALS ?? undefined;
+
+      const credentials = safeParseJSON(googleAuthStr);
+      const googleAuthOptions = credentials ? { credentials } : undefined;
+
+      const instance = LobeVertexAI.initFromVertexAI({
+        googleAuthOptions,
+        location: process.env.VERTEXAI_LOCATION,
+        project: !!credentials?.project_id ? credentials?.project_id : process.env.VERTEXAI_PROJECT,
+      });
+
+      return new AgentRuntime(instance);
+    },
+    params: Promise.resolve({ provider: ModelProvider.VertexAI }),
+  }),
+);

package/src/app/[variants]/(main)/settings/provider/(detail)/huggingface/page.tsx

@@ -43,15 +43,15 @@ const useProviderCard = (): ProviderItem => {
         ) : (
           <FormPassword
             autoComplete={'new-password'}
-            placeholder={t(`huggingface.accessToken.placeholder`)}
+            placeholder={t('huggingface.accessToken.placeholder')}
           />
         ),
         desc: (
           <Markdown className={styles.markdown} fontSize={12} variant={'chat'}>
-            {t(`huggingface.accessToken.desc`)}
+            {t('huggingface.accessToken.desc')}
           </Markdown>
         ),
-        label: t(`huggingface.accessToken.title`),
+        label: t('huggingface.accessToken.title'),
         name: [KeyVaultsConfigKey, LLMProviderApiTokenKey],
       },
     ],

package/src/app/[variants]/(main)/settings/provider/(detail)/vertexai/page.tsx

@@ -0,0 +1,67 @@
+'use client';
+
+import { Markdown } from '@lobehub/ui';
+import { createStyles } from 'antd-style';
+import { useTranslation } from 'react-i18next';
+
+import { FormPassword } from '@/components/FormInput';
+import { VertexAIProviderCard } from '@/config/modelProviders';
+import { aiProviderSelectors, useAiInfraStore } from '@/store/aiInfra';
+import { GlobalLLMProviderKey } from '@/types/user/settings';
+
+import { KeyVaultsConfigKey, LLMProviderApiTokenKey } from '../../const';
+import { SkeletonInput } from '../../features/ProviderConfig';
+import { ProviderItem } from '../../type';
+import ProviderDetail from '../[id]';
+
+const useStyles = createStyles(({ css, token }) => ({
+  markdown: css`
+    p {
+      color: ${token.colorTextDescription} !important;
+    }
+  `,
+  tip: css`
+    font-size: 12px;
+    color: ${token.colorTextDescription};
+  `,
+}));
+
+const providerKey: GlobalLLMProviderKey = 'vertexai';
+
+// Same as OpenAIProvider, but replace API Key with HuggingFace Access Token
+const useProviderCard = (): ProviderItem => {
+  const { t } = useTranslation('modelProvider');
+  const { styles } = useStyles();
+  const isLoading = useAiInfraStore(aiProviderSelectors.isAiProviderConfigLoading(providerKey));
+
+  return {
+    ...VertexAIProviderCard,
+    apiKeyItems: [
+      {
+        children: isLoading ? (
+          <SkeletonInput />
+        ) : (
+          <FormPassword
+            autoComplete={'new-password'}
+            placeholder={t('vertexai.apiKey.placeholder')}
+          />
+        ),
+        desc: (
+          <Markdown className={styles.markdown} fontSize={12} variant={'chat'}>
+            {t('vertexai.apiKey.desc')}
+          </Markdown>
+        ),
+        label: t('vertexai.apiKey.title'),
+        name: [KeyVaultsConfigKey, LLMProviderApiTokenKey],
+      },
+    ],
+  };
+};
+
+const Page = () => {
+  const card = useProviderCard();
+
+  return <ProviderDetail {...card} />;
+};
+
+export default Page;

package/src/config/aiModels/index.ts

@@ -40,6 +40,7 @@ import { default as taichu } from './taichu';
 import { default as tencentcloud } from './tencentcloud';
 import { default as togetherai } from './togetherai';
 import { default as upstage } from './upstage';
+import { default as vertexai } from './vertexai';
 import { default as vllm } from './vllm';
 import { default as volcengine } from './volcengine';
 import { default as wenxin } from './wenxin';
@@ -107,6 +108,7 @@ export const LOBE_DEFAULT_MODEL_LIST = buildDefaultModelList({
   tencentcloud,
   togetherai,
   upstage,
+  vertexai,
   vllm,
   volcengine,
   wenxin,
@@ -155,6 +157,7 @@ export { default as taichu } from './taichu';
 export { default as tencentcloud } from './tencentcloud';
 export { default as togetherai } from './togetherai';
 export { default as upstage } from './upstage';
+export { default as vertexai } from './vertexai';
 export { default as vllm } from './vllm';
 export { default as volcengine } from './volcengine';
 export { default as wenxin } from './wenxin';

package/src/config/aiModels/vertexai.ts

@@ -0,0 +1,200 @@
+import { AIChatModelCard } from '@/types/aiModel';
+
+// ref: https://ai.google.dev/gemini-api/docs/models/gemini
+const vertexaiChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+      vision: true,
+    },
+    contextWindowTokens: 2_097_152 + 8192,
+    description:
+      'Gemini 2.0 Pro Experimental 是 Google 最新的实验性多模态AI模型，与历史版本相比有一定的质量提升，特别是对于世界知识、代码和长上下文。',
+    displayName: 'Gemini 2.0 Pro Experimental 02-05',
+    enabled: true,
+    id: 'gemini-2.0-pro-exp-02-05',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0,
+      input: 0,
+      output: 0,
+    },
+    releasedAt: '2025-02-05',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      vision: true,
+    },
+    contextWindowTokens: 1_048_576 + 8192,
+    description:
+      'Gemini 2.0 Flash 提供下一代功能和改进，包括卓越的速度、原生工具使用、多模态生成和1M令牌上下文窗口。',
+    displayName: 'Gemini 2.0 Flash',
+    enabled: true,
+    id: 'gemini-2.0-flash',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.025,
+      input: 0.1,
+      output: 0.4,
+    },
+    releasedAt: '2025-02-05',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      vision: true,
+    },
+    contextWindowTokens: 1_048_576 + 8192,
+    description:
+      'Gemini 2.0 Flash 提供下一代功能和改进，包括卓越的速度、原生工具使用、多模态生成和1M令牌上下文窗口。',
+    displayName: 'Gemini 2.0 Flash 001',
+    id: 'gemini-2.0-flash-001',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.025,
+      input: 0.1,
+      output: 0.4,
+    },
+    releasedAt: '2025-02-05',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      vision: true,
+    },
+    contextWindowTokens: 1_048_576 + 8192,
+    description: '一个 Gemini 2.0 Flash 模型，针对成本效益和低延迟等目标进行了优化。',
+    displayName: 'Gemini 2.0 Flash-Lite Preview 02-05',
+    id: 'gemini-2.0-flash-lite-preview-02-05',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.018_75,
+      input: 0.075,
+      output: 0.3,
+    },
+    releasedAt: '2025-02-05',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      reasoning: true,
+      vision: true,
+    },
+    contextWindowTokens: 1_048_576 + 65_536,
+    description:
+      'Gemini 2.0 Flash Thinking Exp 是 Google 的实验性多模态推理AI模型，能对复杂问题进行推理，拥有新的思维能力。',
+    displayName: 'Gemini 2.0 Flash Thinking Experimental 01-21',
+    enabled: true,
+    id: 'gemini-2.0-flash-thinking-exp-01-21',
+    maxOutput: 65_536,
+    pricing: {
+      cachedInput: 0,
+      input: 0,
+      output: 0,
+    },
+    releasedAt: '2025-01-21',
+    type: 'chat',
+  },
+  {
+    abilities: { functionCall: true, vision: true },
+    contextWindowTokens: 1_000_000 + 8192,
+    description:
+      'Gemini 1.5 Flash 是Google最新的多模态AI模型，具备快速处理能力，支持文本、图像和视频输入，适用于多种任务的高效扩展。',
+    displayName: 'Gemini 1.5 Flash',
+    enabled: true,
+    id: 'gemini-1.5-flash',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.018_75,
+      input: 0.075,
+      output: 0.3,
+    },
+    type: 'chat',
+  },
+  {
+    abilities: { functionCall: true, vision: true },
+    contextWindowTokens: 1_000_000 + 8192,
+    description: 'Gemini 1.5 Flash 002 是一款高效的多模态模型，支持广泛应用的扩展。',
+    displayName: 'Gemini 1.5 Flash 002',
+    enabled: true,
+    id: 'gemini-1.5-flash-002',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.018_75,
+      input: 0.075,
+      output: 0.3,
+    },
+    releasedAt: '2024-09-25',
+    type: 'chat',
+  },
+  {
+    abilities: { functionCall: true, vision: true },
+    contextWindowTokens: 1_000_000 + 8192,
+    description: 'Gemini 1.5 Flash 001 是一款高效的多模态模型，支持广泛应用的扩展。',
+    displayName: 'Gemini 1.5 Flash 001',
+    id: 'gemini-1.5-flash-001',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.018_75,
+      input: 0.075,
+      output: 0.3,
+    },
+    type: 'chat',
+  },
+  {
+    abilities: { functionCall: true, vision: true },
+    contextWindowTokens: 2_000_000 + 8192,
+    description:
+      'Gemini 1.5 Pro 支持高达200万个tokens，是中型多模态模型的理想选择，适用于复杂任务的多方面支持。',
+    displayName: 'Gemini 1.5 Pro',
+    enabled: true,
+    id: 'gemini-1.5-pro-latest',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.875,
+      input: 3.5,
+      output: 10.5,
+    },
+    releasedAt: '2024-02-15',
+    type: 'chat',
+  },
+  {
+    abilities: { functionCall: true, vision: true },
+    contextWindowTokens: 2_000_000 + 8192,
+    description:
+      'Gemini 1.5 Pro 002 是最新的生产就绪模型，提供更高质量的输出，特别在数学、长上下文和视觉任务方面有显著提升。',
+    displayName: 'Gemini 1.5 Pro 002',
+    enabled: true,
+    id: 'gemini-1.5-pro-002',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.315,
+      input: 1.25,
+      output: 2.5,
+    },
+    releasedAt: '2024-09-24',
+    type: 'chat',
+  },
+  {
+    abilities: { functionCall: true, vision: true },
+    contextWindowTokens: 2_000_000 + 8192,
+    description: 'Gemini 1.5 Pro 001 是可扩展的多模态AI解决方案，支持广泛的复杂任务。',
+    displayName: 'Gemini 1.5 Pro 001',
+    id: 'gemini-1.5-pro-001',
+    maxOutput: 8192,
+    pricing: {
+      cachedInput: 0.875,
+      input: 3.5,
+      output: 10.5,
+    },
+    releasedAt: '2024-02-15',
+    type: 'chat',
+  },
+];
+
+export const allModels = [...vertexaiChatModels];
+
+export default allModels;

package/src/config/modelProviders/index.ts

@@ -40,6 +40,7 @@ import TaichuProvider from './taichu';
 import TencentcloudProvider from './tencentcloud';
 import TogetherAIProvider from './togetherai';
 import UpstageProvider from './upstage';
+import VertexAIProvider from './vertexai';
 import VLLMProvider from './vllm';
 import VolcengineProvider from './volcengine';
 import WenxinProvider from './wenxin';
@@ -102,6 +103,7 @@ export const DEFAULT_MODEL_PROVIDER_LIST = [
   AnthropicProvider,
   BedrockProvider,
   GoogleProvider,
+  VertexAIProvider,
   DeepSeekProvider,
   HuggingFaceProvider,
   OpenRouterProvider,
@@ -191,6 +193,7 @@ export { default as TaichuProviderCard } from './taichu';
 export { default as TencentCloudProviderCard } from './tencentcloud';
 export { default as TogetherAIProviderCard } from './togetherai';
 export { default as UpstageProviderCard } from './upstage';
+export { default as VertexAIProviderCard } from './vertexai';
 export { default as VLLMProviderCard } from './vllm';
 export { default as VolcengineProviderCard } from './volcengine';
 export { default as WenxinProviderCard } from './wenxin';

package/src/config/modelProviders/vertexai.ts

@@ -0,0 +1,22 @@
+import { ModelProviderCard } from '@/types/llm';
+
+// ref: https://ai.google.dev/gemini-api/docs/models/gemini
+const VertexAI: ModelProviderCard = {
+  chatModels: [],
+  checkModel: 'gemini-1.5-flash-001',
+  description:
+    'Google 的 Gemini 系列是其最先进、通用的 AI模型，由 Google DeepMind 打造，专为多模态设计，支持文本、代码、图像、音频和视频的无缝理解与处理。适用于从数据中心到移动设备的多种环境，极大提升了AI模型的效率与应用广泛性。',
+  id: 'vertexai',
+  modelsUrl: 'https://console.cloud.google.com/vertex-ai/model-garden',
+  name: 'VertexAI',
+  settings: {
+    disableBrowserRequest: true,
+    smoothing: {
+      speed: 2,
+      text: true,
+    },
+  },
+  url: 'https://cloud.google.com/vertex-ai',
+};
+
+export default VertexAI;

package/src/database/client/db.ts

@@ -201,7 +201,8 @@ export class DatabaseManager {
         const dbName = 'lobechat';
 
         // make db as web worker if worker is available
-        if (typeof Worker !== 'undefined') {
+        // https://github.com/lobehub/lobe-chat/issues/5785
+        if (typeof Worker !== 'undefined' && typeof navigator.locks !== 'undefined') {
           db = await initPgliteWorker({
             dbName,
             fsBundle: fsBundle as Blob,

package/src/libs/agent-runtime/error.ts

@@ -13,6 +13,7 @@ export const AgentRuntimeErrorType = {
   OllamaBizError: 'OllamaBizError',
 
   InvalidBedrockCredentials: 'InvalidBedrockCredentials',
+  InvalidVertexCredentials: 'InvalidVertexCredentials',
   StreamChunkError: 'StreamChunkError',
 
   InvalidGithubToken: 'InvalidGithubToken',

package/src/libs/agent-runtime/google/index.ts

@@ -1,3 +1,4 @@
+import type { VertexAI } from '@google-cloud/vertexai';
 import {
   Content,
   FunctionCallPart,
@@ -9,6 +10,7 @@ import {
 } from '@google/generative-ai';
 
 import type { ChatModelCard } from '@/types/llm';
+import { VertexAIStream } from '@/libs/agent-runtime/utils/streams/vertex-ai';
 import { imageUrlToBase64 } from '@/utils/imageToBase64';
 import { safeParseJSON } from '@/utils/safeParseJSON';
 
@@ -56,17 +58,27 @@ function getThreshold(model: string): HarmBlockThreshold {
 
 const DEFAULT_BASE_URL = 'https://generativelanguage.googleapis.com';
 
+interface LobeGoogleAIParams {
+  apiKey?: string;
+  baseURL?: string;
+  client?: GoogleGenerativeAI | VertexAI;
+  isVertexAi?: boolean;
+}
+
 export class LobeGoogleAI implements LobeRuntimeAI {
   private client: GoogleGenerativeAI;
+  private isVertexAi: boolean;
   baseURL?: string;
   apiKey?: string;
 
-  constructor({ apiKey, baseURL }: { apiKey?: string; baseURL?: string } = {}) {
+  constructor({ apiKey, baseURL, client, isVertexAi }: LobeGoogleAIParams = {}) {
     if (!apiKey) throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
 
     this.client = new GoogleGenerativeAI(apiKey);
-    this.baseURL = baseURL || DEFAULT_BASE_URL;
     this.apiKey = apiKey;
+    this.client = client ? (client as GoogleGenerativeAI) : new GoogleGenerativeAI(apiKey);
+    this.baseURL = client ? undefined : baseURL || DEFAULT_BASE_URL;
+    this.isVertexAi = isVertexAi || false;
   }
 
   async chat(rawPayload: ChatStreamPayload, options?: ChatCompetitionOptions) {
@@ -117,18 +129,24 @@ export class LobeGoogleAI implements LobeRuntimeAI {
       const googleStream = convertIterableToStream(geminiStreamResult.stream);
       const [prod, useForDebug] = googleStream.tee();
 
-      if (process.env.DEBUG_GOOGLE_CHAT_COMPLETION === '1') {
+      const key = this.isVertexAi
+        ? 'DEBUG_VERTEX_AI_CHAT_COMPLETION'
+        : 'DEBUG_GOOGLE_CHAT_COMPLETION';
+
+      if (process.env[key] === '1') {
         debugStream(useForDebug).catch();
       }
 
       // Convert the response into a friendly text-stream
-      const stream = GoogleGenerativeAIStream(prod, options?.callback);
+      const Stream = this.isVertexAi ? VertexAIStream : GoogleGenerativeAIStream;
+      const stream = Stream(prod, options?.callback);
 
       // Respond with the stream
       return StreamingResponse(stream, { headers: options?.headers });
     } catch (e) {
       const err = e as Error;
 
+      console.log(err);
       const { errorType, error } = this.parseErrorMessage(err.message);
 
       throw AgentRuntimeError.chat({ error, errorType, provider: ModelProvider.Google });

package/src/libs/agent-runtime/types/type.ts

@@ -66,6 +66,7 @@ export enum ModelProvider {
   TogetherAI = 'togetherai',
   Upstage = 'upstage',
   VLLM = 'vllm',
+  VertexAI = 'vertexai',
   Volcengine = 'volcengine',
   Wenxin = 'wenxin',
   XAI = 'xai',