@lobehub/chat 1.45.17 → 1.46.0

package/src/libs/agent-runtime/lmstudio/index.test.ts

@@ -0,0 +1,255 @@
+// @vitest-environment node
+import OpenAI from 'openai';
+import { Mock, afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+import {
+  ChatStreamCallbacks,
+  LobeOpenAICompatibleRuntime,
+  ModelProvider,
+} from '@/libs/agent-runtime';
+
+import * as debugStreamModule from '../utils/debugStream';
+import { LobeLMStudioAI } from './index';
+
+const provider = ModelProvider.LMStudio;
+const defaultBaseURL = 'http://localhost:1234/v1';
+
+const bizErrorType = 'ProviderBizError';
+const invalidErrorType = 'InvalidProviderAPIKey';
+
+// Mock the console.error to avoid polluting test output
+vi.spyOn(console, 'error').mockImplementation(() => {});
+
+let instance: LobeOpenAICompatibleRuntime;
+
+beforeEach(() => {
+  instance = new LobeLMStudioAI({ apiKey: 'test' });
+
+  // 使用 vi.spyOn 来模拟 chat.completions.create 方法
+  vi.spyOn(instance['client'].chat.completions, 'create').mockResolvedValue(
+    new ReadableStream() as any,
+  );
+});
+
+afterEach(() => {
+  vi.clearAllMocks();
+});
+
+describe('LobeLMStudioAI', () => {
+  describe('init', () => {
+    it('should correctly initialize with an API key', async () => {
+      const instance = new LobeLMStudioAI({ apiKey: 'test_api_key' });
+      expect(instance).toBeInstanceOf(LobeLMStudioAI);
+      expect(instance.baseURL).toEqual(defaultBaseURL);
+    });
+  });
+
+  describe('chat', () => {
+    describe('Error', () => {
+      it('should return OpenAIBizError with an openai error response when OpenAI.APIError is thrown', async () => {
+        // Arrange
+        const apiError = new OpenAI.APIError(
+          400,
+          {
+            status: 400,
+            error: {
+              message: 'Bad Request',
+            },
+          },
+          'Error message',
+          {},
+        );
+
+        vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(apiError);
+
+        // Act
+        try {
+          await instance.chat({
+            messages: [{ content: 'Hello', role: 'user' }],
+            model: 'deepseek-chat',
+            temperature: 0,
+          });
+        } catch (e) {
+          expect(e).toEqual({
+            endpoint: defaultBaseURL,
+            error: {
+              error: { message: 'Bad Request' },
+              status: 400,
+            },
+            errorType: bizErrorType,
+            provider,
+          });
+        }
+      });
+
+      it('should throw AgentRuntimeError with NoOpenAIAPIKey if no apiKey is provided', async () => {
+        try {
+          new LobeLMStudioAI({});
+        } catch (e) {
+          expect(e).toEqual({ errorType: invalidErrorType });
+        }
+      });
+
+      it('should return OpenAIBizError with the cause when OpenAI.APIError is thrown with cause', async () => {
+        // Arrange
+        const errorInfo = {
+          stack: 'abc',
+          cause: {
+            message: 'api is undefined',
+          },
+        };
+        const apiError = new OpenAI.APIError(400, errorInfo, 'module error', {});
+
+        vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(apiError);
+
+        // Act
+        try {
+          await instance.chat({
+            messages: [{ content: 'Hello', role: 'user' }],
+            model: 'deepseek-chat',
+            temperature: 0,
+          });
+        } catch (e) {
+          expect(e).toEqual({
+            endpoint: defaultBaseURL,
+            error: {
+              cause: { message: 'api is undefined' },
+              stack: 'abc',
+            },
+            errorType: bizErrorType,
+            provider,
+          });
+        }
+      });
+
+      it('should return OpenAIBizError with an cause response with desensitize Url', async () => {
+        // Arrange
+        const errorInfo = {
+          stack: 'abc',
+          cause: { message: 'api is undefined' },
+        };
+        const apiError = new OpenAI.APIError(400, errorInfo, 'module error', {});
+
+        instance = new LobeLMStudioAI({
+          apiKey: 'test',
+
+          baseURL: 'https://api.abc.com/v1',
+        });
+
+        vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(apiError);
+
+        // Act
+        try {
+          await instance.chat({
+            messages: [{ content: 'Hello', role: 'user' }],
+            model: 'deepseek-chat',
+            temperature: 0,
+          });
+        } catch (e) {
+          expect(e).toEqual({
+            endpoint: 'https://api.***.com/v1',
+            error: {
+              cause: { message: 'api is undefined' },
+              stack: 'abc',
+            },
+            errorType: bizErrorType,
+            provider,
+          });
+        }
+      });
+
+      it('should throw an InvalidDeepSeekAPIKey error type on 401 status code', async () => {
+        // Mock the API call to simulate a 401 error
+        const error = new Error('Unauthorized') as any;
+        error.status = 401;
+        vi.mocked(instance['client'].chat.completions.create).mockRejectedValue(error);
+
+        try {
+          await instance.chat({
+            messages: [{ content: 'Hello', role: 'user' }],
+            model: 'deepseek-chat',
+            temperature: 0,
+          });
+        } catch (e) {
+          // Expect the chat method to throw an error with InvalidDeepSeekAPIKey
+          expect(e).toEqual({
+            endpoint: defaultBaseURL,
+            error: new Error('Unauthorized'),
+            errorType: invalidErrorType,
+            provider,
+          });
+        }
+      });
+
+      it('should return AgentRuntimeError for non-OpenAI errors', async () => {
+        // Arrange
+        const genericError = new Error('Generic Error');
+
+        vi.spyOn(instance['client'].chat.completions, 'create').mockRejectedValue(genericError);
+
+        // Act
+        try {
+          await instance.chat({
+            messages: [{ content: 'Hello', role: 'user' }],
+            model: 'deepseek-chat',
+            temperature: 0,
+          });
+        } catch (e) {
+          expect(e).toEqual({
+            endpoint: defaultBaseURL,
+            errorType: 'AgentRuntimeError',
+            provider,
+            error: {
+              name: genericError.name,
+              cause: genericError.cause,
+              message: genericError.message,
+              stack: genericError.stack,
+            },
+          });
+        }
+      });
+    });
+
+    describe('DEBUG', () => {
+      it('should call debugStream and return StreamingTextResponse when DEBUG_LMSTUDIO_CHAT_COMPLETION is 1', async () => {
+        // Arrange
+        const mockProdStream = new ReadableStream() as any; // 模拟的 prod 流
+        const mockDebugStream = new ReadableStream({
+          start(controller) {
+            controller.enqueue('Debug stream content');
+            controller.close();
+          },
+        }) as any;
+        mockDebugStream.toReadableStream = () => mockDebugStream; // 添加 toReadableStream 方法
+
+        // 模拟 chat.completions.create 返回值，包括模拟的 tee 方法
+        (instance['client'].chat.completions.create as Mock).mockResolvedValue({
+          tee: () => [mockProdStream, { toReadableStream: () => mockDebugStream }],
+        });
+
+        // 保存原始环境变量值
+        const originalDebugValue = process.env.DEBUG_LMSTUDIO_CHAT_COMPLETION;
+
+        // 模拟环境变量
+        process.env.DEBUG_LMSTUDIO_CHAT_COMPLETION = '1';
+        vi.spyOn(debugStreamModule, 'debugStream').mockImplementation(() => Promise.resolve());
+
+        // 执行测试
+        // 运行你的测试函数，确保它会在条件满足时调用 debugStream
+        // 假设的测试函数调用，你可能需要根据实际情况调整
+        await instance.chat({
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'deepseek-chat',
+          stream: true,
+          temperature: 0,
+        });
+
+        // 验证 debugStream 被调用
+        expect(debugStreamModule.debugStream).toHaveBeenCalled();
+
+        // 恢复原始环境变量值
+        process.env.DEBUG_LMSTUDIO_CHAT_COMPLETION = originalDebugValue;
+      });
+    });
+  });
+});

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

@@ -0,0 +1,11 @@
+import { ModelProvider } from '../types';
+import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
+
+export const LobeLMStudioAI = LobeOpenAICompatibleFactory({
+  apiKey: 'placeholder-to-avoid-error',
+  baseURL: 'http://localhost:1234/v1',
+  debug: {
+    chatCompletion: () => process.env.DEBUG_LMSTUDIO_CHAT_COMPLETION === '1',
+  },
+  provider: ModelProvider.LMStudio,
+});

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

@@ -6,7 +6,13 @@ import { ChatModelCard } from '@/types/llm';
 
 import { LobeRuntimeAI } from '../BaseAI';
 import { AgentRuntimeErrorType } from '../error';
-import { ChatCompetitionOptions, ChatStreamPayload, ModelProvider } from '../types';
+import {
+  ChatCompetitionOptions,
+  ChatStreamPayload,
+  Embeddings,
+  EmbeddingsPayload,
+  ModelProvider,
+} from '../types';
 import { AgentRuntimeError } from '../utils/createError';
 import { debugStream } from '../utils/debugStream';
 import { StreamingResponse } from '../utils/response';
@@ -84,6 +90,18 @@ export class LobeOllamaAI implements LobeRuntimeAI {
     }
   }
 
+  async embeddings(payload: EmbeddingsPayload): Promise<Embeddings[]> {
+    const input = Array.isArray(payload.input) ? payload.input : [payload.input];
+    const promises = input.map((inputText: string) =>
+      this.invokeEmbeddingModel({
+        dimensions: payload.dimensions,
+        input: inputText,
+        model: payload.model,
+      }),
+    );
+    return await Promise.all(promises);
+  }
+
   async models(): Promise<ChatModelCard[]> {
     const list = await this.client.list();
     return list.models.map((model) => ({
@@ -91,6 +109,24 @@ export class LobeOllamaAI implements LobeRuntimeAI {
     }));
   }
 
+  private invokeEmbeddingModel = async (payload: EmbeddingsPayload): Promise<Embeddings> => {
+    try {
+      const responseBody = await this.client.embeddings({
+        model: payload.model,
+        prompt: payload.input as string,
+      });
+      return responseBody.embedding;
+    } catch (error) {
+      const e = error as { message: string; name: string; status_code: number };
+
+      throw AgentRuntimeError.chat({
+        error: { message: e.message, name: e.name, status_code: e.status_code },
+        errorType: AgentRuntimeErrorType.OllamaBizError,
+        provider: ModelProvider.Ollama,
+      });
+    }
+  };
+
   private buildOllamaMessages(messages: OpenAIChatMessage[]) {
     return messages.map((message) => this.convertContentToOllamaMessage(message));
   }

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

@@ -39,6 +39,7 @@ export enum ModelProvider {
   HuggingFace = 'huggingface',
   Hunyuan = 'hunyuan',
   InternLM = 'internlm',
+  LMStudio = 'lmstudio',
   Minimax = 'minimax',
   Mistral = 'mistral',
   Moonshot = 'moonshot',

package/src/server/globalConfig/index.ts

@@ -1,6 +1,7 @@
 import { appEnv, getAppConfig } from '@/config/app';
 import { authEnv } from '@/config/auth';
 import { fileEnv } from '@/config/file';
+import { knowledgeEnv } from '@/config/knowledge';
 import { langfuseEnv } from '@/config/langfuse';
 import { enableNextAuth } from '@/const/auth';
 import { parseSystemAgent } from '@/server/globalConfig/parseSystemAgent';
@@ -9,6 +10,7 @@ import { GlobalServerConfig } from '@/types/serverConfig';
 import { genServerLLMConfig } from './_deprecated';
 import { genServerAiProvidersConfig } from './genServerAiProviderConfig';
 import { parseAgentConfig } from './parseDefaultAgent';
+import { parseFilesConfig } from './parseFilesConfig';
 
 export const getServerGlobalConfig = () => {
   const { ACCESS_CODES, DEFAULT_AGENT_CONFIG } = getAppConfig();
@@ -73,3 +75,7 @@ export const getServerDefaultAgentConfig = () => {
 
   return parseAgentConfig(DEFAULT_AGENT_CONFIG) || {};
 };
+
+export const getServerDefaultFilesConfig = () => {
+  return parseFilesConfig(knowledgeEnv.DEFAULT_FILES_CONFIG);
+};

package/src/server/globalConfig/parseFilesConfig.test.ts

@@ -0,0 +1,17 @@
+import { describe, expect, it } from 'vitest';
+
+import { parseFilesConfig } from './parseFilesConfig';
+
+describe('parseFilesConfig', () => {
+  // 测试embeddings配置是否被正确解析
+  it('parses embeddings configuration correctly', () => {
+    const envStr =
+      'embedding_model=openai/embedding-text-3-large,reranker_model=cohere/rerank-english-v3.0,query_model=full_text';
+    const expected = {
+      embeddingModel: { provider: 'openai', model: 'embedding-text-3-large' },
+      rerankerModel: { provider: 'cohere', model: 'rerank-english-v3.0' },
+      queryModel: 'full_text',
+    };
+    expect(parseFilesConfig(envStr)).toEqual(expected);
+  });
+});

package/src/server/globalConfig/parseFilesConfig.ts

@@ -0,0 +1,57 @@
+import { DEFAULT_FILES_CONFIG } from '@/const/settings/knowledge';
+import { SystemEmbeddingConfig } from '@/types/knowledgeBase';
+import { FilesConfig } from '@/types/user/settings/filesConfig';
+
+const protectedKeys = Object.keys({
+  embedding_model: null,
+  query_model: null,
+  reranker_model: null,
+});
+
+export const parseFilesConfig = (envString: string = ''): SystemEmbeddingConfig => {
+  if (!envString) return DEFAULT_FILES_CONFIG;
+  const config: FilesConfig = {} as any;
+
+  // 处理全角逗号和多余空格
+  let envValue = envString.replaceAll('，', ',').trim();
+
+  const pairs = envValue.split(',');
+
+  for (const pair of pairs) {
+    const [key, value] = pair.split('=').map((s) => s.trim());
+
+    if (key && value) {
+      const [provider, ...modelParts] = value.split('/');
+      const model = modelParts.join('/');
+
+      if ((!provider || !model) && key !== 'query_model') {
+        throw new Error('Missing model or provider value');
+      }
+
+      if (key === 'query_model' && value === '') {
+        throw new Error('Missing query mode value');
+      }
+
+      if (protectedKeys.includes(key)) {
+        switch (key) {
+          case 'embedding_model': {
+            config.embeddingModel = { model: model.trim(), provider: provider.trim() };
+            break;
+          }
+          case 'reranker_model': {
+            config.rerankerModel = { model: model.trim(), provider: provider.trim() };
+            break;
+          }
+          case 'query_model': {
+            config.queryModel = value;
+            break;
+          }
+        }
+      }
+    } else {
+      throw new Error('Invalid environment variable format');
+    }
+  }
+
+  return config;
+};

package/src/server/routers/async/file.ts

@@ -5,15 +5,15 @@ import { z } from 'zod';
 
 import { serverDBEnv } from '@/config/db';
 import { fileEnv } from '@/config/file';
-import { DEFAULT_EMBEDDING_MODEL } from '@/const/settings';
+import { DEFAULT_FILE_EMBEDDING_MODEL_ITEM } from '@/const/settings/knowledge';
 import { NewChunkItem, NewEmbeddingsItem } from '@/database/schemas';
 import { serverDB } from '@/database/server';
 import { ASYNC_TASK_TIMEOUT, AsyncTaskModel } from '@/database/server/models/asyncTask';
 import { ChunkModel } from '@/database/server/models/chunk';
 import { EmbeddingModel } from '@/database/server/models/embedding';
 import { FileModel } from '@/database/server/models/file';
-import { ModelProvider } from '@/libs/agent-runtime';
 import { asyncAuthedProcedure, asyncRouter as router } from '@/libs/trpc/async';
+import { getServerDefaultFilesConfig } from '@/server/globalConfig';
 import { initAgentRuntimeWithUserPayload } from '@/server/modules/AgentRuntime';
 import { S3 } from '@/server/modules/S3';
 import { ChunkService } from '@/server/services/chunk';
@@ -44,7 +44,6 @@ export const fileRouter = router({
     .input(
       z.object({
         fileId: z.string(),
-        model: z.string().default(DEFAULT_EMBEDDING_MODEL),
         taskId: z.string(),
       }),
     )
@@ -57,6 +56,9 @@ export const fileRouter = router({
 
       const asyncTask = await ctx.asyncTaskModel.findById(input.taskId);
 
+      const { model, provider } =
+        getServerDefaultFilesConfig().embeddingModel || DEFAULT_FILE_EMBEDDING_MODEL_ITEM;
+
       if (!asyncTask) throw new TRPCError({ code: 'BAD_REQUEST', message: 'Async Task not found' });
 
       try {
@@ -84,13 +86,12 @@ export const fileRouter = router({
 
           const chunks = await ctx.chunkModel.getChunksTextByFileId(input.fileId);
           const requestArray = chunk(chunks, CHUNK_SIZE);
-
           try {
             await pMap(
               requestArray,
               async (chunks, index) => {
                 const agentRuntime = await initAgentRuntimeWithUserPayload(
-                  ModelProvider.OpenAI,
+                  provider,
                   ctx.jwtPayload,
                 );
 
@@ -98,11 +99,10 @@ export const fileRouter = router({
                 console.log(`执行第 ${number} 个任务`);
 
                 console.time(`任务[${number}]: embeddings`);
-
                 const embeddings = await agentRuntime.embeddings({
                   dimensions: 1024,
                   input: chunks.map((c) => c.text),
-                  model: input.model,
+                  model,
                 });
                 console.timeEnd(`任务[${number}]: embeddings`);
 
@@ -111,7 +111,7 @@ export const fileRouter = router({
                     chunkId: chunks[idx].id,
                     embeddings: e,
                     fileId: input.fileId,
-                    model: input.model,
+                    model,
                   })) || [];
 
                 console.time(`任务[${number}]: insert db`);

package/src/server/routers/lambda/chunk.ts

@@ -1,7 +1,7 @@
 import { inArray } from 'drizzle-orm/expressions';
 import { z } from 'zod';
 
-import { DEFAULT_EMBEDDING_MODEL } from '@/const/settings';
+import { DEFAULT_FILE_EMBEDDING_MODEL_ITEM } from '@/const/settings/knowledge';
 import { knowledgeBaseFiles } from '@/database/schemas';
 import { serverDB } from '@/database/server';
 import { AsyncTaskModel } from '@/database/server/models/asyncTask';
@@ -9,9 +9,9 @@ import { ChunkModel } from '@/database/server/models/chunk';
 import { EmbeddingModel } from '@/database/server/models/embedding';
 import { FileModel } from '@/database/server/models/file';
 import { MessageModel } from '@/database/server/models/message';
-import { ModelProvider } from '@/libs/agent-runtime';
 import { authedProcedure, router } from '@/libs/trpc';
 import { keyVaults } from '@/libs/trpc/middleware/keyVaults';
+import { getServerDefaultFilesConfig } from '@/server/globalConfig';
 import { initAgentRuntimeWithUserPayload } from '@/server/modules/AgentRuntime';
 import { ChunkService } from '@/server/services/chunk';
 import { SemanticSearchSchema } from '@/types/rag';
@@ -101,21 +101,18 @@ export const chunkRouter = router({
     .input(
       z.object({
         fileIds: z.array(z.string()).optional(),
-        model: z.string().default(DEFAULT_EMBEDDING_MODEL),
         query: z.string(),
       }),
     )
     .mutation(async ({ ctx, input }) => {
-      console.time('embedding');
-      const agentRuntime = await initAgentRuntimeWithUserPayload(
-        ModelProvider.OpenAI,
-        ctx.jwtPayload,
-      );
+      const { model, provider } =
+        getServerDefaultFilesConfig().embeddingModel || DEFAULT_FILE_EMBEDDING_MODEL_ITEM;
+      const agentRuntime = await initAgentRuntimeWithUserPayload(provider, ctx.jwtPayload);
 
       const embeddings = await agentRuntime.embeddings({
         dimensions: 1024,
         input: input.query,
-        model: input.model,
+        model,
       });
       console.timeEnd('embedding');
 
@@ -130,27 +127,25 @@ export const chunkRouter = router({
     .input(SemanticSearchSchema)
     .mutation(async ({ ctx, input }) => {
       const item = await ctx.messageModel.findMessageQueriesById(input.messageId);
+      const { model, provider } =
+        getServerDefaultFilesConfig().embeddingModel || DEFAULT_FILE_EMBEDDING_MODEL_ITEM;
       let embedding: number[];
       let ragQueryId: string;
-
       // if there is no message rag or it's embeddings, then we need to create one
       if (!item || !item.embeddings) {
         // TODO: need to support customize
-        const agentRuntime = await initAgentRuntimeWithUserPayload(
-          ModelProvider.OpenAI,
-          ctx.jwtPayload,
-        );
+        const agentRuntime = await initAgentRuntimeWithUserPayload(provider, ctx.jwtPayload);
 
         const embeddings = await agentRuntime.embeddings({
           dimensions: 1024,
           input: input.rewriteQuery,
-          model: input.model || DEFAULT_EMBEDDING_MODEL,
+          model,
         });
 
         embedding = embeddings![0];
         const embeddingsId = await ctx.embeddingModel.create({
           embeddings: embedding,
-          model: input.model,
+          model,
         });
 
         const result = await ctx.messageModel.createMessageQuery({
@@ -182,6 +177,7 @@ export const chunkRouter = router({
         fileIds: finalFileIds,
         query: input.rewriteQuery,
       });
+      // TODO: need to rerank the chunks
       console.timeEnd('semanticSearch');
 
       return { chunks, queryId: ragQueryId };

package/src/types/knowledgeBase/index.ts

@@ -1,3 +1,5 @@
+import { FilesConfigItem } from '../user/settings/filesConfig';
+
 export enum KnowledgeBaseTabs {
   Files = 'files',
   Settings = 'Settings',
@@ -43,3 +45,9 @@ export interface KnowledgeItem {
   name: string;
   type: KnowledgeType;
 }
+
+export interface SystemEmbeddingConfig {
+  embeddingModel: FilesConfigItem;
+  queryModel: string;
+  rerankerModel: FilesConfigItem;
+}

package/src/types/user/settings/filesConfig.ts

@@ -0,0 +1,9 @@
+export interface FilesConfigItem {
+  model: string;
+  provider: string;
+}
+export interface FilesConfig {
+  embeddingModel: FilesConfigItem;
+  queryModel: string;
+  rerankerModel: FilesConfigItem;
+}

package/src/types/user/settings/keyVaults.ts

@@ -44,6 +44,7 @@ export interface UserKeyVaults {
   huggingface?: OpenAICompatibleKeyVault;
   hunyuan?: OpenAICompatibleKeyVault;
   internlm?: OpenAICompatibleKeyVault;
+  lmstudio?: OpenAICompatibleKeyVault;
   lobehub?: any;
   minimax?: OpenAICompatibleKeyVault;
   mistral?: OpenAICompatibleKeyVault;