@lobehub/chat 1.71.0 → 1.71.2

package/src/database/server/models/__tests__/chunk.test.ts

@@ -495,13 +495,13 @@ content in Table html is below:
     });
 
     // 测试结果限制
-    it('should limit results to 5 items', async () => {
+    it('should limit results to 15 items', async () => {
       const fileId = '1';
-      // Create 6 chunks
+      // Create 24 chunks
       const chunkResult = await serverDB
         .insert(chunks)
         .values(
-          Array(6)
+          Array(24)
             .fill(0)
             .map((_, i) => ({ text: `Test Chunk ${i}`, userId })),
         )
@@ -528,7 +528,7 @@ content in Table html is below:
         query: 'test',
       });
 
-      expect(result).toHaveLength(5);
+      expect(result).toHaveLength(15);
     });
   });
 });

package/src/database/server/models/chunk.ts

@@ -207,7 +207,8 @@ export class ChunkModel {
       .leftJoin(files, eq(files.id, fileChunks.fileId))
       .where(inArray(fileChunks.fileId, fileIds))
       .orderBy((t) => desc(t.similarity))
-      .limit(5);
+      // 先放宽到 15
+      .limit(15);
 
     return result.map((item) => {
       return {

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

@@ -38,6 +38,10 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
   apiKey?: string;
   private id: string;
 
+  private isDebug() {
+    return process.env.DEBUG_ANTHROPIC_CHAT_COMPLETION === '1';
+  }
+
   constructor({ apiKey, baseURL = DEFAULT_BASE_URL, id, ...res }: AnthropicAIParams = {}) {
     if (!apiKey) throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
 
@@ -51,6 +55,11 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
     try {
       const anthropicPayload = await this.buildAnthropicPayload(payload);
 
+      if (this.isDebug()) {
+        console.log('[requestPayload]');
+        console.log(JSON.stringify(anthropicPayload), '\n');
+      }
+
       const response = await this.client.messages.create(
         { ...anthropicPayload, stream: true },
         {
@@ -60,7 +69,7 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
 
       const [prod, debug] = response.tee();
 
-      if (process.env.DEBUG_ANTHROPIC_CHAT_COMPLETION === '1') {
+      if (this.isDebug()) {
         debugStream(debug.toReadableStream()).catch(console.error);
       }
 

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

@@ -10,7 +10,6 @@ import {
   SchemaType,
 } from '@google/generative-ai';
 
-import { VertexAIStream } from '@/libs/agent-runtime/utils/streams/vertex-ai';
 import type { ChatModelCard } from '@/types/llm';
 import { imageUrlToBase64 } from '@/utils/imageToBase64';
 import { safeParseJSON } from '@/utils/safeParseJSON';
@@ -24,15 +23,27 @@ import {
   OpenAIChatMessage,
   UserMessageContentPart,
 } from '../types';
-import { ModelProvider } from '../types/type';
 import { AgentRuntimeError } from '../utils/createError';
 import { debugStream } from '../utils/debugStream';
 import { StreamingResponse } from '../utils/response';
-import { GoogleGenerativeAIStream, convertIterableToStream } from '../utils/streams';
+import {
+  GoogleGenerativeAIStream,
+  VertexAIStream,
+  convertIterableToStream,
+} from '../utils/streams';
 import { parseDataUri } from '../utils/uriParser';
 
 const modelsOffSafetySettings = new Set(['gemini-2.0-flash-exp']);
-const modelsWithModalities = new Set(['gemini-2.0-flash-exp']);
+
+const modelsWithModalities = new Set([
+  'gemini-2.0-flash-exp',
+  'gemini-2.0-flash-exp-image-generation',
+]);
+
+const modelsDisableInstuction = new Set([
+  'gemini-2.0-flash-exp',
+  'gemini-2.0-flash-exp-image-generation',
+]);
 
 export interface GoogleModelCard {
   displayName: string;
@@ -65,6 +76,7 @@ interface LobeGoogleAIParams {
   apiKey?: string;
   baseURL?: string;
   client?: GoogleGenerativeAI | VertexAI;
+  id?: string;
   isVertexAi?: boolean;
 }
 
@@ -73,8 +85,9 @@ export class LobeGoogleAI implements LobeRuntimeAI {
   private isVertexAi: boolean;
   baseURL?: string;
   apiKey?: string;
+  provider: string;
 
-  constructor({ apiKey, baseURL, client, isVertexAi }: LobeGoogleAIParams = {}) {
+  constructor({ apiKey, baseURL, client, isVertexAi, id }: LobeGoogleAIParams = {}) {
     if (!apiKey) throw AgentRuntimeError.createError(AgentRuntimeErrorType.InvalidProviderAPIKey);
 
     this.client = new GoogleGenerativeAI(apiKey);
@@ -82,6 +95,8 @@ export class LobeGoogleAI implements LobeRuntimeAI {
     this.client = client ? (client as GoogleGenerativeAI) : new GoogleGenerativeAI(apiKey);
     this.baseURL = client ? undefined : baseURL || DEFAULT_BASE_URL;
     this.isVertexAi = isVertexAi || false;
+
+    this.provider = id || (isVertexAi ? 'vertexai' : 'google');
   }
 
   async chat(rawPayload: ChatStreamPayload, options?: ChatCompetitionOptions) {
@@ -97,9 +112,7 @@ export class LobeGoogleAI implements LobeRuntimeAI {
             generationConfig: {
               maxOutputTokens: payload.max_tokens,
               // @ts-expect-error - Google SDK 0.24.0 doesn't have this property for now with
-              response_modalities: modelsWithModalities.has(model)
-                ? ['Text', 'Image']
-                : undefined,
+              response_modalities: modelsWithModalities.has(model) ? ['Text', 'Image'] : undefined,
               temperature: payload.temperature,
               topP: payload.top_p,
             },
@@ -129,7 +142,9 @@ export class LobeGoogleAI implements LobeRuntimeAI {
         )
         .generateContentStream({
           contents,
-          systemInstruction: payload.system as string,
+          systemInstruction: modelsDisableInstuction.has(model)
+            ? undefined
+            : (payload.system as string),
           tools: this.buildGoogleTools(payload.tools, payload),
         });
 
@@ -156,7 +171,7 @@ export class LobeGoogleAI implements LobeRuntimeAI {
       console.log(err);
       const { errorType, error } = this.parseErrorMessage(err.message);
 
-      throw AgentRuntimeError.chat({ error, errorType, provider: ModelProvider.Google });
+      throw AgentRuntimeError.chat({ error, errorType, provider: this.provider });
     }
   }
 

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

@@ -37,7 +37,6 @@ import { LobeStepfunAI } from './stepfun';
 import { LobeTaichuAI } from './taichu';
 import { LobeTencentCloudAI } from './tencentcloud';
 import { LobeTogetherAI } from './togetherai';
-import { ModelProvider } from './types';
 import { LobeUpstageAI } from './upstage';
 import { LobeVLLMAI } from './vllm';
 import { LobeVolcengineAI } from './volcengine';
@@ -47,51 +46,51 @@ import { LobeZeroOneAI } from './zeroone';
 import { LobeZhipuAI } from './zhipu';
 
 export const providerRuntimeMap = {
-  [ModelProvider.OpenAI]: LobeOpenAI,
-  [ModelProvider.Azure]: LobeAzureOpenAI,
-  [ModelProvider.AzureAI]: LobeAzureAI,
-  [ModelProvider.ZhiPu]: LobeZhipuAI,
-  [ModelProvider.Google]: LobeGoogleAI,
-  [ModelProvider.Moonshot]: LobeMoonshotAI,
-  [ModelProvider.Bedrock]: LobeBedrockAI,
-  [ModelProvider.LMStudio]: LobeLMStudioAI,
-  [ModelProvider.Ollama]: LobeOllamaAI,
-  [ModelProvider.VLLM]: LobeVLLMAI,
-  [ModelProvider.Perplexity]: LobePerplexityAI,
-  [ModelProvider.Anthropic]: LobeAnthropicAI,
-  [ModelProvider.DeepSeek]: LobeDeepSeekAI,
-  [ModelProvider.HuggingFace]: LobeHuggingFaceAI,
-  [ModelProvider.Minimax]: LobeMinimaxAI,
-  [ModelProvider.Mistral]: LobeMistralAI,
-  [ModelProvider.Groq]: LobeGroq,
-  [ModelProvider.Github]: LobeGithubAI,
-  [ModelProvider.OpenRouter]: LobeOpenRouterAI,
-  [ModelProvider.TogetherAI]: LobeTogetherAI,
-  [ModelProvider.FireworksAI]: LobeFireworksAI,
-  [ModelProvider.ZeroOne]: LobeZeroOneAI,
-  [ModelProvider.Stepfun]: LobeStepfunAI,
-  [ModelProvider.Qwen]: LobeQwenAI,
-  [ModelProvider.Novita]: LobeNovitaAI,
-  [ModelProvider.Nvidia]: LobeNvidiaAI,
-  [ModelProvider.Taichu]: LobeTaichuAI,
-  [ModelProvider.Baichuan]: LobeBaichuanAI,
-  [ModelProvider.Ai360]: LobeAi360AI,
-  [ModelProvider.SiliconCloud]: LobeSiliconCloudAI,
-  [ModelProvider.GiteeAI]: LobeGiteeAI,
-  [ModelProvider.Upstage]: LobeUpstageAI,
-  [ModelProvider.Spark]: LobeSparkAI,
-  [ModelProvider.Ai21]: LobeAi21AI,
-  [ModelProvider.Hunyuan]: LobeHunyuanAI,
-  [ModelProvider.SenseNova]: LobeSenseNovaAI,
-  [ModelProvider.XAI]: LobeXAI,
-  [ModelProvider.Jina]: LobeJinaAI,
-  [ModelProvider.SambaNova]: LobeSambaNovaAI,
-  [ModelProvider.Cloudflare]: LobeCloudflareAI,
-  [ModelProvider.InternLM]: LobeInternLMAI,
-  [ModelProvider.Higress]: LobeHigressAI,
-  [ModelProvider.TencentCloud]: LobeTencentCloudAI,
-  [ModelProvider.Volcengine]: LobeVolcengineAI,
-  [ModelProvider.PPIO]: LobePPIOAI,
-  [ModelProvider.Doubao]: LobeVolcengineAI,
-  [ModelProvider.Wenxin]: LobeWenxinAI,
+  ai21: LobeAi21AI,
+  ai360: LobeAi360AI,
+  anthropic: LobeAnthropicAI,
+  azure: LobeAzureOpenAI,
+  azureai: LobeAzureAI,
+  baichuan: LobeBaichuanAI,
+  bedrock: LobeBedrockAI,
+  cloudflare: LobeCloudflareAI,
+  deepseek: LobeDeepSeekAI,
+  doubao: LobeVolcengineAI,
+  fireworksai: LobeFireworksAI,
+  giteeai: LobeGiteeAI,
+  github: LobeGithubAI,
+  google: LobeGoogleAI,
+  groq: LobeGroq,
+  higress: LobeHigressAI,
+  huggingface: LobeHuggingFaceAI,
+  hunyuan: LobeHunyuanAI,
+  internlm: LobeInternLMAI,
+  jina: LobeJinaAI,
+  lmstudio: LobeLMStudioAI,
+  minimax: LobeMinimaxAI,
+  mistral: LobeMistralAI,
+  moonshot: LobeMoonshotAI,
+  novita: LobeNovitaAI,
+  nvidia: LobeNvidiaAI,
+  ollama: LobeOllamaAI,
+  openai: LobeOpenAI,
+  openrouter: LobeOpenRouterAI,
+  perplexity: LobePerplexityAI,
+  ppio: LobePPIOAI,
+  qwen: LobeQwenAI,
+  sambanova: LobeSambaNovaAI,
+  sensenova: LobeSenseNovaAI,
+  siliconcloud: LobeSiliconCloudAI,
+  spark: LobeSparkAI,
+  stepfun: LobeStepfunAI,
+  taichu: LobeTaichuAI,
+  tencentcloud: LobeTencentCloudAI,
+  togetherai: LobeTogetherAI,
+  upstage: LobeUpstageAI,
+  vllm: LobeVLLMAI,
+  volcengine: LobeVolcengineAI,
+  wenxin: LobeWenxinAI,
+  xai: LobeXAI,
+  zeroone: LobeZeroOneAI,
+  zhipu: LobeZhipuAI,
 };

package/src/libs/agent-runtime/utils/streams/google-ai.test.ts

@@ -193,4 +193,94 @@ describe('GoogleGenerativeAIStream', () => {
       `data: {"inputImageTokens":258,"inputTextTokens":8,"totalInputTokens":266,"totalTokens":266}\n\n`,
     ]);
   });
+
+  it('should handle stop with content', async () => {
+    vi.spyOn(uuidModule, 'nanoid').mockReturnValueOnce('1');
+
+    const data = [
+      {
+        candidates: [
+          {
+            content: { parts: [{ text: '234' }], role: 'model' },
+            safetyRatings: [
+              { category: 'HARM_CATEGORY_HATE_SPEECH', probability: 'NEGLIGIBLE' },
+              { category: 'HARM_CATEGORY_DANGEROUS_CONTENT', probability: 'NEGLIGIBLE' },
+              { category: 'HARM_CATEGORY_HARASSMENT', probability: 'NEGLIGIBLE' },
+              { category: 'HARM_CATEGORY_SEXUALLY_EXPLICIT', probability: 'NEGLIGIBLE' },
+            ],
+          },
+        ],
+        text: () => '234',
+        usageMetadata: {
+          promptTokenCount: 20,
+          totalTokenCount: 20,
+          promptTokensDetails: [{ modality: 'TEXT', tokenCount: 20 }],
+        },
+        modelVersion: 'gemini-2.0-flash-exp-image-generation',
+      },
+      {
+        text: () => '567890\n',
+        candidates: [
+          {
+            content: { parts: [{ text: '567890\n' }], role: 'model' },
+            finishReason: 'STOP',
+            safetyRatings: [
+              { category: 'HARM_CATEGORY_HATE_SPEECH', probability: 'NEGLIGIBLE' },
+              { category: 'HARM_CATEGORY_DANGEROUS_CONTENT', probability: 'NEGLIGIBLE' },
+              { category: 'HARM_CATEGORY_HARASSMENT', probability: 'NEGLIGIBLE' },
+              { category: 'HARM_CATEGORY_SEXUALLY_EXPLICIT', probability: 'NEGLIGIBLE' },
+            ],
+          },
+        ],
+        usageMetadata: {
+          promptTokenCount: 19,
+          candidatesTokenCount: 11,
+          totalTokenCount: 30,
+          promptTokensDetails: [{ modality: 'TEXT', tokenCount: 19 }],
+          candidatesTokensDetails: [{ modality: 'TEXT', tokenCount: 11 }],
+        },
+        modelVersion: 'gemini-2.0-flash-exp-image-generation',
+      },
+    ];
+
+    const mockGoogleStream = new ReadableStream({
+      start(controller) {
+        data.forEach((item) => {
+          controller.enqueue(item);
+        });
+
+        controller.close();
+      },
+    });
+
+    const protocolStream = GoogleGenerativeAIStream(mockGoogleStream);
+
+    const decoder = new TextDecoder();
+    const chunks = [];
+
+    // @ts-ignore
+    for await (const chunk of protocolStream) {
+      chunks.push(decoder.decode(chunk, { stream: true }));
+    }
+
+    expect(chunks).toEqual(
+      [
+        'id: chat_1',
+        'event: text',
+        'data: "234"\n',
+
+        'id: chat_1',
+        'event: text',
+        `data: "567890\\n"\n`,
+        // stop
+        'id: chat_1',
+        'event: stop',
+        `data: "STOP"\n`,
+        // usage
+        'id: chat_1',
+        'event: usage',
+        `data: {"inputTextTokens":19,"totalInputTokens":19,"totalOutputTokens":11,"totalTokens":30}\n`,
+      ].map((i) => i + '\n'),
+    );
+  });
 });

package/src/libs/agent-runtime/utils/streams/google-ai.ts

@@ -71,6 +71,7 @@ const transformGoogleGenerativeAIStream = (
       if (chunk.usageMetadata) {
         const usage = chunk.usageMetadata;
         return [
+          !!text ? { data: text, id: context?.id, type: 'text' } : undefined,
           { data: candidate.finishReason, id: context?.id, type: 'stop' },
           {
             data: {
@@ -88,7 +89,7 @@ const transformGoogleGenerativeAIStream = (
             id: context?.id,
             type: 'usage',
           },
-        ];
+        ].filter(Boolean) as StreamProtocolChunk[];
       }
       return { data: candidate.finishReason, id: context?.id, type: 'stop' };
     }

package/src/libs/agent-runtime/utils/streams/index.ts

@@ -6,3 +6,4 @@ export * from './openai';
 export * from './protocol';
 export * from './qwen';
 export * from './spark';
+export * from './vertex-ai';

package/src/locales/default/components.ts

@@ -78,6 +78,7 @@ export default {
       custom: '自定义模型，默认设定同时支持函数调用与视觉识别，请根据实际情况验证上述能力的可用性',
       file: '该模型支持上传文件读取与识别',
       functionCall: '该模型支持函数调用（Function Call）',
+      imageOutput: '该模型支持生成图片',
       reasoning: '该模型支持深度思考',
       search: '该模型支持联网搜索',
       tokens: '该模型单个会话最多支持 {{tokens}} Tokens',

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

@@ -122,13 +122,16 @@ export const agentRouter = router({
       const knowledge = await ctx.agentModel.getAgentAssignedKnowledge(input.agentId);
 
       return [
-        ...files.map((file) => ({
-          enabled: knowledge.files.some((item) => item.id === file.id),
-          fileType: file.fileType,
-          id: file.id,
-          name: file.name,
-          type: KnowledgeType.File,
-        })),
+        ...files
+          // 过滤掉所有图片
+          .filter((file) => !file.fileType.startsWith('image'))
+          .map((file) => ({
+            enabled: knowledge.files.some((item) => item.id === file.id),
+            fileType: file.fileType,
+            id: file.id,
+            name: file.name,
+            type: KnowledgeType.File,
+          })),
         ...knowledgeBases.map((knowledgeBase) => ({
           avatar: knowledgeBase.avatar,
           description: knowledgeBase.description,

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

@@ -1,3 +1,4 @@
+import { TRPCError } from '@trpc/server';
 import { inArray } from 'drizzle-orm/expressions';
 import { z } from 'zod';
 
@@ -126,60 +127,75 @@ export const chunkRouter = router({
   semanticSearchForChat: chunkProcedure
     .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(provider, ctx.jwtPayload);
-
-        const embeddings = await agentRuntime.embeddings({
-          dimensions: 1024,
-          input: input.rewriteQuery,
-          model,
-        });
-
-        embedding = embeddings![0];
-        const embeddingsId = await ctx.embeddingModel.create({
-          embeddings: embedding,
-          model,
+      try {
+        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(provider, ctx.jwtPayload);
+
+          // slice content to make sure in the context window limit
+          const query =
+            input.rewriteQuery.length > 8000
+              ? input.rewriteQuery.slice(0, 8000)
+              : input.rewriteQuery;
+
+          const embeddings = await agentRuntime.embeddings({
+            dimensions: 1024,
+            input: query,
+            model,
+          });
+
+          embedding = embeddings![0];
+          const embeddingsId = await ctx.embeddingModel.create({
+            embeddings: embedding,
+            model,
+          });
+
+          const result = await ctx.messageModel.createMessageQuery({
+            embeddingsId,
+            messageId: input.messageId,
+            rewriteQuery: input.rewriteQuery,
+            userQuery: input.userQuery,
+          });
+
+          ragQueryId = result.id;
+        } else {
+          embedding = item.embeddings;
+          ragQueryId = item.id;
+        }
+
+        let finalFileIds = input.fileIds ?? [];
+
+        if (input.knowledgeIds && input.knowledgeIds.length > 0) {
+          const knowledgeFiles = await serverDB.query.knowledgeBaseFiles.findMany({
+            where: inArray(knowledgeBaseFiles.knowledgeBaseId, input.knowledgeIds),
+          });
+
+          finalFileIds = knowledgeFiles.map((f) => f.fileId).concat(finalFileIds);
+        }
+
+        const chunks = await ctx.chunkModel.semanticSearchForChat({
+          embedding,
+          fileIds: finalFileIds,
+          query: input.rewriteQuery,
         });
 
-        const result = await ctx.messageModel.createMessageQuery({
-          embeddingsId,
-          messageId: input.messageId,
-          rewriteQuery: input.rewriteQuery,
-          userQuery: input.userQuery,
-        });
+        // TODO: need to rerank the chunks
 
-        ragQueryId = result.id;
-      } else {
-        embedding = item.embeddings;
-        ragQueryId = item.id;
-      }
+        return { chunks, queryId: ragQueryId };
+      } catch (e) {
+        console.error(e);
 
-      console.time('semanticSearch');
-      let finalFileIds = input.fileIds ?? [];
-
-      if (input.knowledgeIds && input.knowledgeIds.length > 0) {
-        const knowledgeFiles = await serverDB.query.knowledgeBaseFiles.findMany({
-          where: inArray(knowledgeBaseFiles.knowledgeBaseId, input.knowledgeIds),
+        throw new TRPCError({
+          code: 'INTERNAL_SERVER_ERROR',
+          message: (e as any).errorType || JSON.stringify(e),
         });
-
-        finalFileIds = knowledgeFiles.map((f) => f.fileId).concat(finalFileIds);
       }
-
-      const chunks = await ctx.chunkModel.semanticSearchForChat({
-        embedding,
-        fileIds: finalFileIds,
-        query: input.rewriteQuery,
-      });
-      // TODO: need to rerank the chunks
-      console.timeEnd('semanticSearch');
-
-      return { chunks, queryId: ragQueryId };
     }),
 });

package/src/store/chat/slices/aiChat/actions/rag.ts

@@ -21,7 +21,7 @@ export interface ChatRAGAction {
     id: string,
     userQuery: string,
     messages: string[],
-  ) => Promise<{ chunks: ChatSemanticSearchChunk[]; queryId: string; rewriteQuery?: string }>;
+  ) => Promise<{ chunks: ChatSemanticSearchChunk[]; queryId?: string; rewriteQuery?: string }>;
   /**
    * Rewrite user content to better RAG query
    */
@@ -74,17 +74,23 @@ export const chatRag: StateCreator<ChatStore, [['zustand/devtools', never]], [],
 
     // 2. retrieve chunks from semantic search
     const files = chatSelectors.currentUserFiles(get()).map((f) => f.id);
-    const { chunks, queryId } = await ragService.semanticSearchForChat({
-      fileIds: knowledgeIds().fileIds.concat(files),
-      knowledgeIds: knowledgeIds().knowledgeBaseIds,
-      messageId: id,
-      rewriteQuery: rewriteQuery || userQuery,
-      userQuery,
-    });
-
-    get().internal_toggleMessageRAGLoading(false, id);
-
-    return { chunks, queryId, rewriteQuery };
+    try {
+      const { chunks, queryId } = await ragService.semanticSearchForChat({
+        fileIds: knowledgeIds().fileIds.concat(files),
+        knowledgeIds: knowledgeIds().knowledgeBaseIds,
+        messageId: id,
+        rewriteQuery: rewriteQuery || userQuery,
+        userQuery,
+      });
+
+      get().internal_toggleMessageRAGLoading(false, id);
+
+      return { chunks, queryId, rewriteQuery };
+    } catch {
+      get().internal_toggleMessageRAGLoading(false, id);
+
+      return { chunks: [] };
+    }
   },
   internal_rewriteQuery: async (id, content, messages) => {
     let rewriteQuery = content;

package/src/types/aiModel.ts

@@ -28,6 +28,10 @@ export interface ModelAbilities {
    * whether model supports function call
    */
   functionCall?: boolean;
+  /**
+   * whether model supports image output
+   */
+  imageOutput?: boolean;
   /**
    * whether model supports reasoning
    */
@@ -36,7 +40,6 @@ export interface ModelAbilities {
    * whether model supports search web
    */
   search?: boolean;
-
   /**
    *  whether model supports vision
    */