feat(knowledge): support Voyage AI (#3810)

* feat(knowledge): support Voyage AI

* chore

src/main/embeddings/Embeddings.ts

@@ -0,0 +1,24 @@
+import type { BaseEmbeddings } from '@llm-tools/embedjs-interfaces'
+import { KnowledgeBaseParams } from '@types'
+
+import EmbeddingsFactory from './EmbeddingsFactory'
+
+export default class Embeddings {
+  private sdk: BaseEmbeddings
+  constructor({ model, apiKey, apiVersion, baseURL, dimensions }: KnowledgeBaseParams) {
+    this.sdk = EmbeddingsFactory.create({ model, apiKey, apiVersion, baseURL, dimensions } as KnowledgeBaseParams)
+  }
+  public async init(): Promise<void> {
+    return this.sdk.init()
+  }
+  public async getDimensions(): Promise<number> {
+    return this.sdk.getDimensions()
+  }
+  public async embedDocuments(texts: string[]): Promise<number[][]> {
+    return this.sdk.embedDocuments(texts)
+  }
+
+  public async embedQuery(text: string): Promise<number[]> {
+    return this.sdk.embedQuery(text)
+  }
+}

src/main/embeddings/EmbeddingsFactory.ts

@@ -0,0 +1,38 @@
+import type { BaseEmbeddings } from '@llm-tools/embedjs-interfaces'
+import { OpenAiEmbeddings } from '@llm-tools/embedjs-openai'
+import { AzureOpenAiEmbeddings } from '@llm-tools/embedjs-openai/src/azure-openai-embeddings'
+import { getInstanceName } from '@main/utils'
+import { KnowledgeBaseParams } from '@types'
+
+import VoyageEmbeddings from './VoyageEmbeddings'
+
+export default class EmbeddingsFactory {
+  static create({ model, apiKey, apiVersion, baseURL, dimensions }: KnowledgeBaseParams): BaseEmbeddings {
+    const batchSize = 10
+    if (model.includes('voyage')) {
+      return new VoyageEmbeddings({
+        modelName: model,
+        apiKey,
+        outputDimension: dimensions,
+        batchSize: 8
+      })
+    }
+    if (apiVersion !== undefined) {
+      return new AzureOpenAiEmbeddings({
+        azureOpenAIApiKey: apiKey,
+        azureOpenAIApiVersion: apiVersion,
+        azureOpenAIApiDeploymentName: model,
+        azureOpenAIApiInstanceName: getInstanceName(baseURL),
+        dimensions,
+        batchSize
+      })
+    }
+    return new OpenAiEmbeddings({
+      model,
+      apiKey,
+      dimensions,
+      batchSize,
+      configuration: { baseURL }
+    })
+  }
+}

src/main/embeddings/VoyageEmbeddings.ts

@@ -0,0 +1,30 @@
+import { VoyageEmbeddings as _VoyageEmbeddings } from '@langchain/community/embeddings/voyage'
+import { BaseEmbeddings } from '@llm-tools/embedjs-interfaces'
+
+export default class VoyageEmbeddings extends BaseEmbeddings {
+  private model: _VoyageEmbeddings
+  constructor(private readonly configuration?: ConstructorParameters<typeof _VoyageEmbeddings>[0]) {
+    super()
+    if (!this.configuration) this.configuration = {}
+    if (!this.configuration.modelName) this.configuration.modelName = 'voyage-3'
+
+    if (!this.configuration.outputDimension) {
+      throw new Error('You need to pass in the optional dimensions parameter for this model')
+    }
+    this.model = new _VoyageEmbeddings(this.configuration)
+  }
+  override async getDimensions(): Promise<number> {
+    if (!this.configuration?.outputDimension) {
+      throw new Error('You need to pass in the optional dimensions parameter for this model')
+    }
+    return this.configuration?.outputDimension
+  }
+
+  override async embedDocuments(texts: string[]): Promise<number[][]> {
+    return this.model.embedDocuments(texts)
+  }
+
+  override async embedQuery(text: string): Promise<number[]> {
+    return this.model.embedQuery(text)
+  }
+}

src/main/reranker/RerankerFactory.ts

@@ -4,6 +4,7 @@ import BaseReranker from './BaseReranker'
 import DefaultReranker from './DefaultReranker'
 import JinaReranker from './JinaReranker'
 import SiliconFlowReranker from './SiliconFlowReranker'
+import VoyageReranker from './VoyageReranker'
 
 export default class RerankerFactory {
   static create(base: KnowledgeBaseParams): BaseReranker {
@@ -11,6 +12,8 @@ export default class RerankerFactory {
       return new SiliconFlowReranker(base)
     } else if (base.rerankModelProvider === 'jina') {
       return new JinaReranker(base)
+    } else if (base.rerankModelProvider === 'voyageai') {
+      return new VoyageReranker(base)
     }
     return new DefaultReranker(base)
   }

src/main/reranker/VoyageReranker.ts

@@ -0,0 +1,60 @@
+import { ExtractChunkData } from '@llm-tools/embedjs-interfaces'
+import { KnowledgeBaseParams } from '@types'
+import axios from 'axios'
+
+import BaseReranker from './BaseReranker'
+
+export default class VoyageReranker extends BaseReranker {
+  constructor(base: KnowledgeBaseParams) {
+    super(base)
+  }
+
+  public rerank = async (query: string, searchResults: ExtractChunkData[]): Promise<ExtractChunkData[]> => {
+    let baseURL = this.base?.rerankBaseURL?.endsWith('/')
+      ? this.base.rerankBaseURL.slice(0, -1)
+      : this.base.rerankBaseURL
+
+    if (baseURL && !baseURL.endsWith('/v1')) {
+      baseURL = `${baseURL}/v1`
+    }
+
+    const url = `${baseURL}/rerank`
+
+    const requestBody = {
+      model: this.base.rerankModel,
+      query,
+      documents: searchResults.map((doc) => doc.pageContent),
+      top_k: this.base.topN,
+      return_documents: false,
+      truncation: true
+    }
+
+    try {
+      const { data } = await axios.post(url, requestBody, {
+        headers: {
+          ...this.defaultHeaders()
+        }
+      })
+
+      const rerankResults = data.data
+
+      const resultMap = new Map(rerankResults.map((result: any) => [result.index, result.relevance_score || 0]))
+
+      return searchResults
+        .map((doc: ExtractChunkData, index: number) => {
+          const score = resultMap.get(index)
+          if (score === undefined) return undefined
+
+          return {
+            ...doc,
+            score
+          }
+        })
+        .filter((doc): doc is ExtractChunkData => doc !== undefined)
+        .sort((a, b) => b.score - a.score)
+    } catch (error: any) {
+      console.error('Voyage Reranker API 错误:', error.message || error)
+      throw new Error(`${error} - BaseUrl: ${baseURL}`)
+    }
+  }
+}

src/main/services/KnowledgeService.ts

@@ -21,11 +21,10 @@ import type { ExtractChunkData } from '@llm-tools/embedjs-interfaces'
 import { LibSqlDb } from '@llm-tools/embedjs-libsql'
 import { SitemapLoader } from '@llm-tools/embedjs-loader-sitemap'
 import { WebLoader } from '@llm-tools/embedjs-loader-web'
-import { AzureOpenAiEmbeddings, OpenAiEmbeddings } from '@llm-tools/embedjs-openai'
+import Embeddings from '@main/embeddings/Embeddings'
 import { addFileLoader } from '@main/loader'
 import Reranker from '@main/reranker/Reranker'
 import { windowService } from '@main/services/WindowService'
-import { getInstanceName } from '@main/utils'
 import { getAllFiles } from '@main/utils/file'
 import type { LoaderReturn } from '@shared/config/types'
 import { FileType, KnowledgeBaseParams, KnowledgeItem } from '@types'
@@ -114,29 +113,20 @@ class KnowledgeService {
     baseURL,
     dimensions
   }: KnowledgeBaseParams): Promise<RAGApplication> => {
-    const batchSize = 10
-    return new RAGApplicationBuilder()
-      .setModel('NO_MODEL')
-      .setEmbeddingModel(
-        apiVersion
-          ? new AzureOpenAiEmbeddings({
-              azureOpenAIApiKey: apiKey,
-              azureOpenAIApiVersion: apiVersion,
-              azureOpenAIApiDeploymentName: model,
-              azureOpenAIApiInstanceName: getInstanceName(baseURL),
-              dimensions,
-              batchSize
-            })
-          : new OpenAiEmbeddings({
-              model,
-              apiKey,
-              dimensions,
-              batchSize,
-              configuration: { baseURL }
-            })
-      )
-      .setVectorDatabase(new LibSqlDb({ path: path.join(this.storageDir, id) }))
-      .build()
+    let ragApplication: RAGApplication
+    const embeddings = new Embeddings({ model, apiKey, apiVersion, baseURL, dimensions } as KnowledgeBaseParams)
+    try {
+      ragApplication = await new RAGApplicationBuilder()
+        .setModel('NO_MODEL')
+        .setEmbeddingModel(embeddings)
+        .setVectorDatabase(new LibSqlDb({ path: path.join(this.storageDir, id) }))
+        .build()
+    } catch (e) {
+      Logger.error(e)
+      throw new Error(`Failed to create RAGApplication: ${e}`)
+    }
+
+    return ragApplication
   }
 
   public create = async (_: Electron.IpcMainInvokeEvent, base: KnowledgeBaseParams): Promise<void> => {

src/renderer/src/config/embedings.ts

@@ -242,6 +242,58 @@ export const EMBEDDING_MODELS = [
   {
     id: 'mistral-embed',
     max_context: 8000
+  },
+  {
+    id: 'voyage-3-large',
+    max_context: 1024
+  },
+  {
+    id: 'voyage-3-large',
+    max_context: 256
+  },
+  {
+    id: 'voyage-3-large',
+    max_context: 512
+  },
+  {
+    id: 'voyage-3-large',
+    max_context: 2048
+  },
+  {
+    id: 'voyage-3',
+    max_context: 1024
+  },
+  {
+    id: 'voyage-3-lite',
+    max_context: 512
+  },
+  {
+    id: 'voyage-code-3',
+    max_context: 1024
+  },
+  {
+    id: 'voyage-code-3',
+    max_context: 256
+  },
+  {
+    id: 'voyage-code-3',
+    max_context: 512
+  },
+  {
+    id: 'voyage-code-3',
+    max_context: 2048
+  },
+  {
+    id: 'voyage-finance-2',
+    max_context: 1024
+  },
+  {
+    id: 'voyage-law-2',
+    max_context: 1024
+  },
+  {
+    id: 'voyage-code-2',
+    max_context: 1536
   }
 ]
 

src/renderer/src/config/models.ts

@@ -122,6 +122,7 @@ import UpstageModelLogo from '@renderer/assets/images/models/upstage.png'
 import UpstageModelLogoDark from '@renderer/assets/images/models/upstage_dark.png'
 import ViduModelLogo from '@renderer/assets/images/models/vidu.png'
 import ViduModelLogoDark from '@renderer/assets/images/models/vidu_dark.png'
+import VoyageModelLogo from '@renderer/assets/images/models/voyageai.png'
 import WenxinModelLogo from '@renderer/assets/images/models/wenxin.png'
 import WenxinModelLogoDark from '@renderer/assets/images/models/wenxin_dark.png'
 import XirangModelLogo from '@renderer/assets/images/models/xirang.png'
@@ -175,7 +176,8 @@ export const REASONING_REGEX =
   /^(o\d+(?:-[\w-]+)?|.*\b(?:reasoner|thinking)\b.*|.*-[rR]\d+.*|.*\bqwq(?:-[\w-]+)?\b.*|.*\bhunyuan-t1(?:-[\w-]+)?\b.*)$/i
 
 // Embedding models
-export const EMBEDDING_REGEX = /(?:^text-|embed|bge-|e5-|LLM2Vec|retrieval|uae-|gte-|jina-clip|jina-embeddings)/i
+export const EMBEDDING_REGEX =
+  /(?:^text-|embed|bge-|e5-|LLM2Vec|retrieval|uae-|gte-|jina-clip|jina-embeddings|voyage-)/i
 
 // Rerank models
 export const RERANKING_REGEX = /(?:rerank|re-rank|re-ranker|re-ranking|retrieval|retriever)/i
@@ -327,7 +329,8 @@ export function getModelLogo(modelId: string) {
     embedding: isLight ? EmbeddingModelLogo : EmbeddingModelLogoDark,
     perplexity: isLight ? PerplexityModelLogo : PerplexityModelLogoDark,
     sonar: isLight ? PerplexityModelLogo : PerplexityModelLogoDark,
-    'bge-': BgeModelLogo
+    'bge-': BgeModelLogo,
+    'voyage-': VoyageModelLogo
   }
 
   for (const key in logoMap) {
@@ -1801,7 +1804,63 @@ export const SYSTEM_MODELS: Record<string, Model[]> = {
       group: 'DeepSeek'
     }
   ],
-  gpustack: []
+  gpustack: [],
+  voyageai: [
+    {
+      id: 'voyage-3-large',
+      provider: 'voyageai',
+      name: 'voyage-3-large',
+      group: 'Voyage Embeddings V3'
+    },
+    {
+      id: 'voyage-3',
+      provider: 'voyageai',
+      name: 'voyage-3',
+      group: 'Voyage Embeddings V3'
+    },
+    {
+      id: 'voyage-3-lite',
+      provider: 'voyageai',
+      name: 'voyage-3-lite',
+      group: 'Voyage Embeddings V3'
+    },
+    {
+      id: 'voyage-code-3',
+      provider: 'voyageai',
+      name: 'voyage-code-3',
+      group: 'Voyage Embeddings V3'
+    },
+    {
+      id: 'voyage-finance-3',
+      provider: 'voyageai',
+      name: 'voyage-finance-3',
+      group: 'Voyage Embeddings V2'
+    },
+    {
+      id: 'voyage-law-2',
+      provider: 'voyageai',
+      name: 'voyage-law-2',
+      group: 'Voyage Embeddings V2'
+    },
+    {
+      id: 'voyage-code-2',
+      provider: 'voyageai',
+      name: 'voyage-code-2',
+      group: 'Voyage Embeddings V2'
+    },
+    {
+      id: 'rerank-2',
+      provider: 'voyageai',
+      name: 'rerank-2',
+      group: 'Voyage Rerank V2'
+    },
+    {
+      id: 'rerank-2-lite',
+      provider: 'voyageai',
+      name: 'rerank-2-lite',
+      group: 'Voyage Rerank V2'
+    }
+  ]
 }
 
 export const TEXT_TO_IMAGES_MODELS = [

src/renderer/src/config/providers.ts

@@ -38,6 +38,7 @@ import StepProviderLogo from '@renderer/assets/images/providers/step.png'
 import TencentCloudProviderLogo from '@renderer/assets/images/providers/tencent-cloud-ti.png'
 import TogetherProviderLogo from '@renderer/assets/images/providers/together.png'
 import BytedanceProviderLogo from '@renderer/assets/images/providers/volcengine.png'
+import VoyageAIProviderLogo from '@renderer/assets/images/providers/voyageai.png'
 import XirangProviderLogo from '@renderer/assets/images/providers/xirang.png'
 import ZeroOneProviderLogo from '@renderer/assets/images/providers/zero-one.png'
 import ZhipuProviderLogo from '@renderer/assets/images/providers/zhipu.png'
@@ -86,14 +87,15 @@ const PROVIDER_LOGO_MAP = {
   o3: O3ProviderLogo,
   'tencent-cloud-ti': TencentCloudProviderLogo,
   gpustack: GPUStackProviderLogo,
-  alayanew: AlayaNewProviderLogo
+  alayanew: AlayaNewProviderLogo,
+  voyageai: VoyageAIProviderLogo
 } as const
 
 export function getProviderLogo(providerId: string) {
   return PROVIDER_LOGO_MAP[providerId as keyof typeof PROVIDER_LOGO_MAP]
 }
 
-export const SUPPORTED_REANK_PROVIDERS = ['silicon', 'jina']
+export const SUPPORTED_REANK_PROVIDERS = ['silicon', 'jina', 'voyageai']
 
 export const PROVIDER_CONFIG = {
   openai: {
@@ -560,5 +562,16 @@ export const PROVIDER_CONFIG = {
       docs: 'https://docs.gpustack.ai/latest/',
       models: 'https://docs.gpustack.ai/latest/overview/#supported-models'
     }
+  },
+  voyageai: {
+    api: {
+      url: 'https://api.voyageai.com'
+    },
+    websites: {
+      official: 'https://www.voyageai.com/',
+      apiKey: 'https://dashboard.voyageai.com/organization/api-keys',
+      docs: 'https://docs.voyageai.com/docs',
+      models: 'https://docs.voyageai.com/docs'
+    }
   }
 }

src/renderer/src/i18n/locales/en-us.json

@@ -678,7 +678,8 @@
       "xirang": "State Cloud Xirang",
       "yi": "Yi",
       "zhinao": "360AI",
-      "zhipu": "ZHIPU AI"
+      "zhipu": "ZHIPU AI",
+      "voyageai": "Voyage AI"
     },
     "restore": {
       "confirm": "Are you sure you want to restore data?",

src/renderer/src/i18n/locales/ja-jp.json

@@ -678,7 +678,8 @@
       "xirang": "天翼クラウド 息壤",
       "yi": "零一万物",
       "zhinao": "360智脳",
-      "zhipu": "智譜AI"
+      "zhipu": "智譜AI",
+      "voyageai": "Voyage AI"
     },
     "restore": {
       "confirm": "データを復元しますか？",

src/renderer/src/i18n/locales/ru-ru.json

@@ -678,7 +678,8 @@
       "xirang": "State Cloud Xirang",
       "yi": "Yi",
       "zhinao": "360AI",
-      "zhipu": "ZHIPU AI"
+      "zhipu": "ZHIPU AI",
+      "voyageai": "Voyage AI"
     },
     "restore": {
       "confirm": "Вы уверены, что хотите восстановить данные?",

src/renderer/src/i18n/locales/zh-cn.json

@@ -678,7 +678,8 @@
       "xirang": "天翼云息壤",
       "yi": "零一万物",
       "zhinao": "360智脑",
-      "zhipu": "智谱AI"
+      "zhipu": "智谱AI",
+      "voyageai":"Voyage AI"
     },
     "restore": {
       "confirm": "确定要恢复数据吗？",

src/renderer/src/i18n/locales/zh-tw.json

@@ -678,7 +678,8 @@
       "xirang": "天翼雲息壤",
       "yi": "零一萬物",
       "zhinao": "360 智腦",
-      "zhipu": "智譜 AI"
+      "zhipu": "智譜 AI",
+      "voyageai": "Voyage AI"
     },
     "restore": {
       "confirm": "確定要復原資料嗎？",

src/renderer/src/pages/settings/ProviderSettings/ModelList.tsx

@@ -257,6 +257,7 @@ const ModelList: React.FC<ModelListProps> = ({ provider: _provider, modelStatuse
           {sortedModelGroups[group].map((model) => {
             const modelStatus = modelStatuses.find((status) => status.model.id === model.id)
             const isChecking = modelStatus?.checking === true
+            console.log('model', model.id, getModelLogo(model.id))
 
             return (
               <ModelListItem key={model.id}>

src/renderer/src/store/index.ts

@@ -40,7 +40,7 @@ const persistedReducer = persistReducer(
   {
     key: 'cherry-studio',
     storage,
-    version: 83,
+    version: 84,
     blacklist: ['runtime', 'messages'],
     migrate
   },

src/renderer/src/store/llm.ts

@@ -456,6 +456,16 @@ export const INITIAL_PROVIDERS: Provider[] = [
     models: SYSTEM_MODELS.gpustack,
     isSystem: true,
     enabled: false
+  },
+  {
+    id: 'voyageai',
+    name: 'VoyageAI',
+    type: 'openai',
+    apiKey: '',
+    apiHost: 'https://api.voyageai.com',
+    models: SYSTEM_MODELS.voyageai,
+    isSystem: true,
+    enabled: false
   }
 ]
 

src/renderer/src/store/migrate.ts

@@ -795,6 +795,10 @@ const migrateConfig = {
     state.settings.launchToTray = false
     state.settings.trayOnClose = true
     return state
+  },
+  '84': (state: RootState) => {
+    addProvider(state, 'voyageai')
+    return state
   }
 }