feat(agent): Refactor search agents and implement SpeedSearchAgent

- Updated FileSearchAgent to improve code readability and formatting.
- Refactored SynthesizerAgent for better prompt handling and document processing.
- Enhanced TaskManagerAgent with clearer file context handling.
- Modified AgentSearch to maintain consistent parameter formatting.
- Introduced SpeedSearchAgent for optimized search functionality.
- Updated metaSearchAgent to support new SpeedSearchAgent.
- Improved file processing utilities for better document handling.
- Added test attachments for sporting events queries.

.github/copilot-instructions.md

@@ -15,6 +15,7 @@ The system works through these main steps:
 ## Architecture Details
 
 ### Technology Stack
+
 - **Frontend**: React, Next.js, Tailwind CSS
 - **Backend**: Node.js
 - **Database**: SQLite with Drizzle ORM
@@ -23,29 +24,31 @@ The system works through these main steps:
 - **Content Processing**: Mozilla Readability, Cheerio, Playwright
 
 ### Database (SQLite + Drizzle ORM)
+
 - Schema: `src/lib/db/schema.ts`
 - Tables: `messages`, `chats`, `systemPrompts`
 - Configuration: `drizzle.config.ts`
 - Local file: `data/db.sqlite`
 
 ### AI/ML Stack
+
 - **LLM Providers**: OpenAI, Anthropic, Groq, Ollama, Gemini, DeepSeek, LM Studio
 - **Embeddings**: Xenova Transformers, similarity search (cosine/dot product)
 - **Agents**: `webSearchAgent`, `analyzerAgent`, `synthesizerAgent`, `taskManagerAgent`
 
 ### External Services
+
 - **Search Engine**: SearXNG integration (`src/lib/searxng.ts`)
 - **Configuration**: TOML-based config file
 
 ### Data Flow
+
 1. User query → Task Manager Agent
 2. Web Search Agent → SearXNG → Content extraction
 3. Analyzer Agent → Content processing + embedding
 4. Synthesizer Agent → LLM response generation
 5. Response with cited sources
 
-
-
 ## Project Structure
 
 - `/src/app`: Next.js app directory with page components and API routes
@@ -117,22 +120,26 @@ When working on this codebase, you might need to:
 ## Code Style & Standards
 
 ### TypeScript Configuration
+
 - Strict mode enabled
 - ES2017 target
 - Path aliases: `@/*` → `src/*`
 - No test files (testing not implemented)
 
 ### Formatting & Linting
+
 - ESLint: Next.js core web vitals rules
 - Prettier: Use `npm run format:write` before commits
 - Import style: Use `@/` prefix for internal imports
 
 ### File Organization
+
 - Components: React functional components with TypeScript
 - API routes: Next.js App Router (`src/app/api/`)
 - Utilities: Grouped by domain (`src/lib/`)
 - Naming: camelCase for functions/variables, PascalCase for components
 
 ### Error Handling
+
 - Use try/catch blocks for async operations
 - Return structured error responses from API routes

docs/API/SEARCH.md

@@ -60,7 +60,6 @@ The API accepts a JSON object in the request body, where you define the focus mo
 - **`optimizationMode`** (string, optional): Specifies the optimization mode to control the balance between performance and quality. Available modes:
 
   - `speed`: Prioritize speed and get the quickest possible answer. Minimum effort retrieving web content. - Only uses SearXNG result previews.
-  - `balanced`: Find the right balance between speed and accuracy. Medium effort retrieving web content. - Uses web scraping technologies to retrieve partial content from full web pages.
   - `agent`: Use an agentic workflow to answer complex multi-part questions. This mode requires a model that is trained for tool use.
 
 - **`query`** (string, required): The search query or question.

docs/architecture/README.md

@@ -11,10 +11,8 @@ Perplexica's architecture consists of the following key components:
    - In Agent mode, the application uses an agentic workflow to answer complex multi-part questions
      - The agent can use reasoning steps to provide comprehensive answers to complex questions
      - Agent mode is experimental and may consume lots of tokens and take a long time to produce responses
-   - In Balanced mode, the application retrieves web content using Playwright and Mozilla Readability to extract relevant segments of web content
-     - Because it only uses segments of web content, it can be less accurate than Agent mode
    - In Speed mode, the application only uses the preview content returned by SearXNG
      - This content is provided by the search engines and contains minimal context from the actual web page
-     - This mode is the least accurate and is often prone to hallucination
+     - This mode prioritizes quick responses over accuracy
 
 For a more detailed explanation of how these components work together, see [WORKING.md](https://github.com/ItzCrazyKns/Perplexica/tree/master/docs/architecture/WORKING.md).

src/app/api/chat/route.ts

@@ -43,7 +43,7 @@ type EmbeddingModel = {
 
 type Body = {
   message: Message;
-  optimizationMode: 'speed' | 'balanced' | 'agent';
+  optimizationMode: 'speed' | 'agent';
   focusMode: string;
   history: Array<[string, string]>;
   files: Array<string>;

src/app/api/search/route.ts

@@ -30,7 +30,7 @@ interface embeddingModel {
 }
 
 interface ChatRequestBody {
-  optimizationMode: 'speed' | 'balanced' | 'agent';
+  optimizationMode: 'speed' | 'agent';
   focusMode: string;
   chatModel?: chatModel;
   embeddingModel?: embeddingModel;
@@ -52,7 +52,7 @@ export const POST = async (req: Request) => {
     }
 
     body.history = body.history || [];
-    body.optimizationMode = body.optimizationMode || 'balanced';
+    body.optimizationMode = body.optimizationMode || 'speed';
     body.stream = body.stream || false;
 
     const history: BaseMessage[] = body.history.map((msg) => {

src/components/MessageInput.tsx

@@ -158,6 +158,7 @@ const MessageInput = ({
               setFileIds={setFileIds}
               files={files}
               setFiles={setFiles}
+              optimizationMode={optimizationMode}
             />
           </div>
           <div className="flex flex-row items-center space-x-2">

src/components/MessageInputActions/Attach.tsx

@@ -14,16 +14,23 @@ const Attach = ({
   setFileIds,
   files,
   setFiles,
+  optimizationMode,
 }: {
   fileIds: string[];
   setFileIds: (fileIds: string[]) => void;
   files: FileType[];
   setFiles: (files: FileType[]) => void;
+  optimizationMode: string;
 }) => {
   const [loading, setLoading] = useState(false);
   const fileInputRef = useRef<any>();
 
+  const isSpeedMode = optimizationMode === 'speed';
+  const isDisabled = isSpeedMode;
+
   const handleChange = async (e: React.ChangeEvent<HTMLInputElement>) => {
+    if (isDisabled) return;
+
     setLoading(true);
     const data = new FormData();
 
@@ -37,7 +44,8 @@ const Attach = ({
     const embeddingModel = localStorage.getItem('embeddingModel');
     const chatModelProvider = localStorage.getItem('chatModelProvider');
     const chatModel = localStorage.getItem('chatModel');
-    const ollamaContextWindow = localStorage.getItem('ollamaContextWindow') || '2048';
+    const ollamaContextWindow =
+      localStorage.getItem('ollamaContextWindow') || '2048';
 
     data.append('embedding_model_provider', embeddingModelProvider!);
     data.append('embedding_model', embeddingModel!);
@@ -67,18 +75,23 @@ const Attach = ({
       </p>
     </div>
   ) : files.length > 0 ? (
+    <div className="relative group">
       <Popover className="relative w-full max-w-[15rem] md:max-w-md lg:max-w-lg">
         <PopoverButton
           type="button"
+          disabled={isDisabled}
           className={cn(
-          'flex flex-row items-center justify-between space-x-1 p-2 text-black/50 dark:text-white/50 rounded-xl hover:bg-light-secondary dark:hover:bg-dark-secondary active:scale-95 transition duration-200 hover:text-black dark:hover:text-white',
+            'flex flex-row items-center justify-between space-x-1 p-2 rounded-xl transition duration-200',
             files.length > 0 ? '-ml-2 lg:-ml-3' : '',
+            isDisabled
+              ? 'text-black/20 dark:text-white/20 cursor-not-allowed'
+              : 'text-black/50 dark:text-white/50 hover:bg-light-secondary dark:hover:bg-dark-secondary hover:text-black dark:hover:text-white',
           )}
         >
           {files.length > 1 && (
             <>
-            <File size={19} className="text-sky-400" />
+              <File size={19} className={isDisabled ? 'text-sky-900' : 'text-sky-400'} />
-            <p className="text-sky-400 inline whitespace-nowrap text-xs font-medium">
+              <p className={cn("inline whitespace-nowrap text-xs font-medium", isDisabled ? 'text-sky-900' : 'text-sky-400')}>
                 {files.length} files
               </p>
             </>
@@ -86,8 +99,8 @@ const Attach = ({
 
           {files.length === 1 && (
             <>
-            <File size={18} className="text-sky-400" />
+              <File size={18} className={isDisabled ? 'text-sky-900' : 'text-sky-400'} />
-            <p className="text-sky-400 text-xs font-medium">
+              <p className={cn("text-xs font-medium", isDisabled ? 'text-sky-900' : 'text-sky-400')}>
                 {files[0].fileName.length > 10
                   ? files[0].fileName.replace(/\.\w+$/, '').substring(0, 3) +
                     '...' +
@@ -115,8 +128,14 @@ const Attach = ({
                 <div className="flex flex-row items-center space-x-4">
                   <button
                     type="button"
-                  onClick={() => fileInputRef.current.click()}
+                    onClick={() => !isDisabled && fileInputRef.current.click()}
-                  className="flex flex-row items-center space-x-1 text-black/70 dark:text-white/70 hover:text-black hover:dark:text-white transition duration-200"
+                    disabled={isDisabled}
+                    className={cn(
+                      'flex flex-row items-center space-x-1 transition duration-200',
+                      isDisabled
+                        ? 'text-black/20 dark:text-white/20 cursor-not-allowed'
+                        : 'text-black/70 dark:text-white/70 hover:text-black hover:dark:text-white',
+                    )}
                   >
                     <input
                       type="file"
@@ -125,16 +144,25 @@ const Attach = ({
                       accept=".pdf,.docx,.txt"
                       multiple
                       hidden
+                      disabled={isDisabled}
                     />
                     <Plus size={18} />
                     <p className="text-xs">Add</p>
                   </button>
                   <button
                     onClick={() => {
+                      if (!isDisabled) {
                         setFiles([]);
                         setFileIds([]);
+                      }
                     }}
-                  className="flex flex-row items-center space-x-1 text-black/70 dark:text-white/70 hover:text-black hover:dark:text-white transition duration-200"
+                    disabled={isDisabled}
+                    className={cn(
+                      'flex flex-row items-center space-x-1 transition duration-200',
+                      isDisabled
+                        ? 'text-black/20 dark:text-white/20 cursor-not-allowed'
+                        : 'text-black/70 dark:text-white/70 hover:text-black hover:dark:text-white',
+                    )}
                   >
                     <Trash size={14} />
                     <p className="text-xs">Clear</p>
@@ -165,12 +193,26 @@ const Attach = ({
           </PopoverPanel>
         </Transition>
       </Popover>
+      {isSpeedMode && (
+        <div className="absolute bottom-full mb-2 left-1/2 transform -translate-x-1/2 opacity-0 group-hover:opacity-100 transition-opacity duration-200 pointer-events-none">
+          <div className="bg-black dark:bg-white text-white dark:text-black text-xs px-2 py-1 rounded whitespace-nowrap">
+            File attachments are disabled in Speed mode
+            <div className="absolute top-full left-1/2 transform -translate-x-1/2 border-4 border-transparent border-t-black dark:border-t-white"></div>
+          </div>
+        </div>
+      )}
+    </div>
   ) : (
+    <div className="relative group">
       <button
         type="button"
-      onClick={() => fileInputRef.current.click()}
+        onClick={() => !isDisabled && fileInputRef.current.click()}
+        disabled={isDisabled}
         className={cn(
-        'flex flex-row items-center space-x-1 text-black/50 dark:text-white/50 rounded-xl hover:bg-light-secondary dark:hover:bg-dark-secondary transition duration-200 hover:text-black dark:hover:text-white p-2',
+          'flex flex-row items-center space-x-1 rounded-xl transition duration-200 p-2',
+          isDisabled
+            ? 'text-black/20 dark:text-white/20 cursor-not-allowed'
+            : 'text-black/50 dark:text-white/50 hover:bg-light-secondary dark:hover:bg-dark-secondary hover:text-black dark:hover:text-white',
         )}
       >
         <input
@@ -180,9 +222,19 @@ const Attach = ({
           accept=".pdf,.docx,.txt"
           multiple
           hidden
+          disabled={isDisabled}
         />
         <Paperclip size="18" />
       </button>
+      {isSpeedMode && (
+        <div className="absolute bottom-full mb-2 left-1/2 transform -translate-x-1/2 opacity-0 group-hover:opacity-100 transition-opacity duration-200 pointer-events-none">
+          <div className="bg-black dark:bg-white text-white dark:text-black text-xs px-2 py-1 rounded whitespace-nowrap">
+            File attachments are disabled in Speed mode
+            <div className="absolute top-full left-1/2 transform -translate-x-1/2 border-4 border-transparent border-t-black dark:border-t-white"></div>
+          </div>
+        </div>
+      )}
+    </div>
   );
 };
 

src/components/MessageInputActions/AttachSmall.tsx

@@ -1,161 +0,0 @@
-import { cn } from '@/lib/utils';
-import {
-  Popover,
-  PopoverButton,
-  PopoverPanel,
-  Transition,
-} from '@headlessui/react';
-import { CopyPlus, File, LoaderCircle, Plus, Trash } from 'lucide-react';
-import { Fragment, useRef, useState } from 'react';
-import { File as FileType } from '../ChatWindow';
-
-const AttachSmall = ({
-  fileIds,
-  setFileIds,
-  files,
-  setFiles,
-}: {
-  fileIds: string[];
-  setFileIds: (fileIds: string[]) => void;
-  files: FileType[];
-  setFiles: (files: FileType[]) => void;
-}) => {
-  const [loading, setLoading] = useState(false);
-  const fileInputRef = useRef<any>();
-
-  const handleChange = async (e: React.ChangeEvent<HTMLInputElement>) => {
-    setLoading(true);
-    const data = new FormData();
-
-    for (let i = 0; i < e.target.files!.length; i++) {
-      data.append('files', e.target.files![i]);
-    }
-
-    const embeddingModelProvider = localStorage.getItem(
-      'embeddingModelProvider',
-    );
-    const embeddingModel = localStorage.getItem('embeddingModel');
-    const chatModelProvider = localStorage.getItem('chatModelProvider');
-    const chatModel = localStorage.getItem('chatModel');
-    const ollamaContextWindow = localStorage.getItem('ollamaContextWindow') || '2048';
-
-    data.append('embedding_model_provider', embeddingModelProvider!);
-    data.append('embedding_model', embeddingModel!);
-    data.append('chat_model_provider', chatModelProvider!);
-    data.append('chat_model', chatModel!);
-    if (chatModelProvider === 'ollama') {
-      data.append('ollama_context_window', ollamaContextWindow);
-    }
-
-    const res = await fetch(`/api/uploads`, {
-      method: 'POST',
-      body: data,
-    });
-
-    const resData = await res.json();
-
-    setFiles([...files, ...resData.files]);
-    setFileIds([...fileIds, ...resData.files.map((file: any) => file.fileId)]);
-    setLoading(false);
-  };
-
-  return loading ? (
-    <div className="flex flex-row items-center justify-between space-x-1 p-1">
-      <LoaderCircle size={20} className="text-sky-400 animate-spin" />
-    </div>
-  ) : files.length > 0 ? (
-    <Popover className="max-w-[15rem] md:max-w-md lg:max-w-lg">
-      <PopoverButton
-        type="button"
-        className="flex flex-row items-center justify-between space-x-1 p-1 text-black/50 dark:text-white/50 rounded-xl hover:bg-light-secondary dark:hover:bg-dark-secondary active:scale-95 transition duration-200 hover:text-black dark:hover:text-white"
-      >
-        <File size={20} className="text-sky-400" />
-      </PopoverButton>
-      <Transition
-        as={Fragment}
-        enter="transition ease-out duration-150"
-        enterFrom="opacity-0 translate-y-1"
-        enterTo="opacity-100 translate-y-0"
-        leave="transition ease-in duration-150"
-        leaveFrom="opacity-100 translate-y-0"
-        leaveTo="opacity-0 translate-y-1"
-      >
-        <PopoverPanel className="absolute z-10 w-64 md:w-[350px] bottom-14 -ml-3">
-          <div className="bg-light-primary dark:bg-dark-primary border rounded-md border-light-200 dark:border-dark-200 w-full max-h-[200px] md:max-h-none overflow-y-auto flex flex-col">
-            <div className="flex flex-row items-center justify-between px-3 py-2">
-              <h4 className="text-black dark:text-white font-medium text-sm">
-                Attached files
-              </h4>
-              <div className="flex flex-row items-center space-x-4">
-                <button
-                  type="button"
-                  onClick={() => fileInputRef.current.click()}
-                  className="flex flex-row items-center space-x-1 text-black/70 dark:text-white/70 hover:text-black hover:dark:text-white transition duration-200"
-                >
-                  <input
-                    type="file"
-                    onChange={handleChange}
-                    ref={fileInputRef}
-                    accept=".pdf,.docx,.txt"
-                    multiple
-                    hidden
-                  />
-                  <Plus size={18} />
-                  <p className="text-xs">Add</p>
-                </button>
-                <button
-                  onClick={() => {
-                    setFiles([]);
-                    setFileIds([]);
-                  }}
-                  className="flex flex-row items-center space-x-1 text-black/70 dark:text-white/70 hover:text-black hover:dark:text-white transition duration-200"
-                >
-                  <Trash size={14} />
-                  <p className="text-xs">Clear</p>
-                </button>
-              </div>
-            </div>
-            <div className="h-[0.5px] mx-2 bg-white/10" />
-            <div className="flex flex-col items-center">
-              {files.map((file, i) => (
-                <div
-                  key={i}
-                  className="flex flex-row items-center justify-start w-full space-x-3 p-3"
-                >
-                  <div className="bg-dark-100 flex items-center justify-center w-10 h-10 rounded-md">
-                    <File size={16} className="text-white/70" />
-                  </div>
-                  <p className="text-black/70 dark:text-white/70 text-sm">
-                    {file.fileName.length > 25
-                      ? file.fileName.replace(/\.\w+$/, '').substring(0, 25) +
-                        '...' +
-                        file.fileExtension
-                      : file.fileName}
-                  </p>
-                </div>
-              ))}
-            </div>
-          </div>
-        </PopoverPanel>
-      </Transition>
-    </Popover>
-  ) : (
-    <button
-      type="button"
-      onClick={() => fileInputRef.current.click()}
-      className="flex flex-row items-center space-x-1 text-black/50 dark:text-white/50 rounded-xl hover:bg-light-secondary dark:hover:bg-dark-secondary transition duration-200 hover:text-black dark:hover:text-white p-1"
-    >
-      <input
-        type="file"
-        onChange={handleChange}
-        ref={fileInputRef}
-        accept=".pdf,.docx,.txt"
-        multiple
-        hidden
-      />
-      <CopyPlus size={20} />
-    </button>
-  );
-};
-
-export default AttachSmall;

src/components/MessageInputActions/Optimization.tsx

@@ -12,7 +12,7 @@ const OptimizationModes = [
     key: 'speed',
     title: 'Speed',
     description:
-      'Prioritize speed and get the quickest possible answer. Minimum effort retrieving web content.',
+      'Prioritize speed and get the quickest possible answer. Uses only web search results - attached files will not be processed.',
     icon: <Zap size={20} className="text-[#FF9800]" />,
   },
   // {

src/lib/agents/contentRouterAgent.ts

@@ -15,9 +15,7 @@ const RouterDecisionSchema = z.object({
   decision: z
     .enum(['file_search', 'web_search', 'analyzer'])
     .describe('The next step to take in the workflow'),
-  reasoning: z
+  reasoning: z.string().describe('Explanation of why this decision was made'),
-    .string()
-    .describe('Explanation of why this decision was made'),
 });
 
 type RouterDecision = z.infer<typeof RouterDecisionSchema>;
@@ -63,7 +61,9 @@ export class ContentRouterAgent {
       const searchHistory = state.searchInstructionHistory.join(', ') || 'None';
 
       // Extract file topics if files are available
-      const fileTopics = hasFiles ? await this.extractFileTopics(state.fileIds!) : 'None';
+      const fileTopics = hasFiles
+        ? await this.extractFileTopics(state.fileIds!)
+        : 'None';
 
       // Emit routing decision event
       this.emitter.emit('agent_action', {
@@ -97,9 +97,12 @@ export class ContentRouterAgent {
       });
 
       // Use structured output for routing decision
-      const structuredLlm = this.llm.withStructuredOutput(RouterDecisionSchema, {
+      const structuredLlm = this.llm.withStructuredOutput(
+        RouterDecisionSchema,
+        {
           name: 'route_content',
-      });
+        },
+      );
 
       const routerDecision = await structuredLlm.invoke(
         [...removeThinkingBlocksFromMessages(state.messages), prompt],
@@ -112,7 +115,11 @@ export class ContentRouterAgent {
       console.log(`Focus mode: ${focusMode}`);
 
       // Validate decision based on focus mode restrictions
-      const validatedDecision = this.validateDecision(routerDecision, focusMode, hasFiles);
+      const validatedDecision = this.validateDecision(
+        routerDecision,
+        focusMode,
+        hasFiles,
+      );
 
       // Emit routing result event
       this.emitter.emit('agent_action', {
@@ -163,7 +170,7 @@ export class ContentRouterAgent {
    */
   private async extractFileTopics(fileIds: string[]): Promise<string> {
     try {
-      const topics = fileIds.map(fileId => {
+      const topics = fileIds.map((fileId) => {
         try {
           const filePath = path.join(process.cwd(), 'uploads', fileId);
           const contentPath = filePath + '-extracted.json';
@@ -199,14 +206,15 @@ export class ContentRouterAgent {
     hasFiles: boolean,
   ): RouterDecision {
     // Enforce focus mode restrictions for chat and localResearch modes
-    if ((focusMode === 'chat' || focusMode === 'localResearch') && 
+    if (
-        decision.decision === 'web_search') {
+      (focusMode === 'chat' || focusMode === 'localResearch') &&
-      
+      decision.decision === 'web_search'
+    ) {
       // Override to file_search if files are available, otherwise analyzer
       const fallbackDecision = hasFiles ? 'file_search' : 'analyzer';
 
       console.log(
-        `Overriding web_search decision to ${fallbackDecision} due to focus mode restriction: ${focusMode}`
+        `Overriding web_search decision to ${fallbackDecision} due to focus mode restriction: ${focusMode}`,
       );
 
       return {

src/lib/agents/fileSearchAgent.ts

@@ -5,7 +5,10 @@ import { EventEmitter } from 'events';
 import { Document } from 'langchain/document';
 import { AgentState } from './agentState';
 import { Embeddings } from '@langchain/core/embeddings';
-import { processFilesToDocuments, getRankedDocs } from '../utils/fileProcessing';
+import {
+  processFilesToDocuments,
+  getRankedDocs,
+} from '../utils/fileProcessing';
 
 export class FileSearchAgent {
   private llm: BaseChatModel;
@@ -79,12 +82,16 @@ export class FileSearchAgent {
         return new Command({
           goto: 'analyzer',
           update: {
-            messages: [new AIMessage('No searchable content found in attached files.')],
+            messages: [
+              new AIMessage('No searchable content found in attached files.'),
+            ],
           },
         });
       }
 
-      console.log(`Processed ${fileDocuments.length} file documents for search`);
+      console.log(
+        `Processed ${fileDocuments.length} file documents for search`,
+      );
 
       // Emit searching file content event
       this.emitter.emit('agent_action', {
@@ -139,7 +146,11 @@ export class FileSearchAgent {
         return new Command({
           goto: 'analyzer',
           update: {
-            messages: [new AIMessage('No relevant content found in attached files for the current task.')],
+            messages: [
+              new AIMessage(
+                'No relevant content found in attached files for the current task.',
+              ),
+            ],
           },
         });
       }
@@ -157,7 +168,8 @@ export class FileSearchAgent {
             totalTasks: state.tasks?.length || 1,
             relevantSections: rankedDocuments.length,
             searchedDocuments: fileDocuments.length,
-            documentCount: state.relevantDocuments.length + rankedDocuments.length,
+            documentCount:
+              state.relevantDocuments.length + rankedDocuments.length,
           },
         },
       });

src/lib/agents/synthesizerAgent.ts

@@ -34,13 +34,13 @@ export class SynthesizerAgent {
       // Format the prompt using the external template
       const template = PromptTemplate.fromTemplate(synthesizerPrompt);
 
-      const conversationHistory = removeThinkingBlocksFromMessages(state.messages)
+      const conversationHistory =
+        removeThinkingBlocksFromMessages(state.messages)
           .map((msg) => `<${msg.getType()}>${msg.content}</${msg.getType()}>`)
           .join('\n') || 'No previous conversation context';
 
       const relevantDocuments = state.relevantDocuments
-        .map(
+        .map((doc, index) => {
-          (doc, index) => {
           const isFile = doc.metadata?.url?.toLowerCase().includes('file');
           return `<${index + 1}>\n
     <title>${doc.metadata.title}</title>\n
@@ -48,8 +48,7 @@ export class SynthesizerAgent {
     ${isFile ? '' : '\n<url>' + doc.metadata.url + '</url>\n'}
     <content>\n${doc.pageContent}\n</content>\n
     </${index + 1}>`;
-          }
+        })
-        )
         .join('\n');
 
       const formattedPrompt = await template.format({

src/lib/agents/taskManagerAgent.ts

@@ -129,7 +129,8 @@ export class TaskManagerAgent {
       const template = PromptTemplate.fromTemplate(taskBreakdownPrompt);
 
       // Create file context information
-      const fileContext = state.fileIds && state.fileIds.length > 0
+      const fileContext =
+        state.fileIds && state.fileIds.length > 0
           ? `Files attached: ${state.fileIds.length} file(s) are available for analysis. Consider creating tasks that can leverage these attached files when appropriate.`
           : 'No files attached: Focus on tasks that can be answered through web research or general knowledge.';
 

src/lib/search/agentSearch.ts

@@ -155,7 +155,7 @@ export class AgentSearch {
   async searchAndAnswer(
     query: string,
     history: BaseMessage[] = [],
-    fileIds: string[] = []
+    fileIds: string[] = [],
   ) {
     const workflow = this.createWorkflow();
 

src/lib/search/index.ts

@@ -1,6 +1,9 @@
 import MetaSearchAgent from '@/lib/search/metaSearchAgent';
+import SpeedSearchAgent from '@/lib/search/speedSearch';
 import prompts from '../prompts';
 
+export { default as SpeedSearchAgent } from './speedSearch';
+
 export const searchHandlers: Record<string, MetaSearchAgent> = {
   webSearch: new MetaSearchAgent({
     activeEngines: [],

src/lib/search/metaSearchAgent.ts

@@ -1,32 +1,9 @@
 import type { Embeddings } from '@langchain/core/embeddings';
 import type { BaseChatModel } from '@langchain/core/language_models/chat_models';
 import { BaseMessage } from '@langchain/core/messages';
-import { StringOutputParser } from '@langchain/core/output_parsers';
-import {
-  ChatPromptTemplate,
-  MessagesPlaceholder,
-  PromptTemplate,
-} from '@langchain/core/prompts';
-import {
-  RunnableLambda,
-  RunnableMap,
-  RunnableSequence,
-} from '@langchain/core/runnables';
-import { StreamEvent } from '@langchain/core/tracers/log_stream';
-import { ChatOpenAI } from '@langchain/openai';
 import eventEmitter from 'events';
-import { Document } from 'langchain/document';
-import fs from 'node:fs';
-import path from 'node:path';
-import LineOutputParser from '../outputParsers/lineOutputParser';
-import LineListOutputParser from '../outputParsers/listLineOutputParser';
-import { searchSearxng } from '../searxng';
-import { formatDateForLLM } from '../utils';
-import computeSimilarity from '../utils/computeSimilarity';
-import { getDocumentsFromLinks, getWebContent } from '../utils/documents';
-import formatChatHistoryAsString from '../utils/formatHistory';
-import { getModelName } from '../utils/modelUtils';
 import { AgentSearch } from './agentSearch';
+import SpeedSearchAgent from './speedSearch';
 
 export interface MetaSearchAgentType {
   searchAndAnswer: (
@@ -34,7 +11,7 @@ export interface MetaSearchAgentType {
     history: BaseMessage[],
     llm: BaseChatModel,
     embeddings: Embeddings,
-    optimizationMode: 'speed' | 'balanced' | 'agent',
+    optimizationMode: 'speed' | 'agent',
     fileIds: string[],
     systemInstructions: string,
     signal: AbortSignal,
@@ -54,623 +31,13 @@ interface Config {
   additionalSearchCriteria?: string;
 }
 
-type BasicChainInput = {
-  chat_history: BaseMessage[];
-  query: string;
-};
-
 class MetaSearchAgent implements MetaSearchAgentType {
   private config: Config;
-  private strParser = new StringOutputParser();
-  private searchQuery?: string;
-  private searxngUrl?: string;
 
   constructor(config: Config) {
     this.config = config;
   }
 
-  /**
-   * Emit a progress event with the given percentage and message
-   */
-  private emitProgress(
-    emitter: eventEmitter,
-    percentage: number,
-    message: string,
-    subMessage?: string,
-  ) {
-    const progressData: any = {
-      message,
-      current: percentage,
-      total: 100,
-    };
-
-    // Add subMessage if provided
-    if (subMessage) {
-      progressData.subMessage = subMessage;
-    }
-
-    emitter.emit(
-      'progress',
-      JSON.stringify({
-        type: 'progress',
-        data: progressData,
-      }),
-    );
-  }
-
-  private async createSearchRetrieverChain(
-    llm: BaseChatModel,
-    systemInstructions: string,
-    emitter: eventEmitter,
-    signal: AbortSignal,
-  ) {
-    // TODO: Don't we want to set this back to default once search is done?
-    (llm as unknown as ChatOpenAI).temperature = 0;
-
-    this.emitProgress(emitter, 10, `Building search query`);
-
-    return RunnableSequence.from([
-      PromptTemplate.fromTemplate(this.config.queryGeneratorPrompt),
-      llm,
-      this.strParser,
-      RunnableLambda.from(async (input: string) => {
-        try {
-          //console.log(`LLM response for initial web search:"${input}"`);
-          const linksOutputParser = new LineListOutputParser({
-            key: 'links',
-          });
-
-          const questionOutputParser = new LineOutputParser({
-            key: 'answer',
-          });
-
-          const links = await linksOutputParser.parse(input);
-          let question = await questionOutputParser.parse(input);
-
-          //console.log('question', question);
-
-          if (question === 'not_needed') {
-            return { query: '', docs: [] };
-          }
-
-          if (links.length > 0) {
-            if (question.length === 0) {
-              question = 'summarize';
-            }
-
-            let docs: Document[] = [];
-
-            const linkDocs = await getDocumentsFromLinks({ links });
-
-            const docGroups: Document[] = [];
-
-            linkDocs.map((doc) => {
-              const URLDocExists = docGroups.find(
-                (d) =>
-                  d.metadata.url === doc.metadata.url &&
-                  d.metadata.totalDocs < 10,
-              );
-
-              if (!URLDocExists) {
-                docGroups.push({
-                  ...doc,
-                  metadata: {
-                    ...doc.metadata,
-                    totalDocs: 1,
-                  },
-                });
-              }
-
-              const docIndex = docGroups.findIndex(
-                (d) =>
-                  d.metadata.url === doc.metadata.url &&
-                  d.metadata.totalDocs < 10,
-              );
-
-              if (docIndex !== -1) {
-                docGroups[docIndex].pageContent =
-                  docGroups[docIndex].pageContent + `\n\n` + doc.pageContent;
-                docGroups[docIndex].metadata.totalDocs += 1;
-              }
-            });
-
-            this.emitProgress(emitter, 20, `Summarizing content`);
-
-            await Promise.all(
-              docGroups.map(async (doc) => {
-                const systemPrompt = systemInstructions
-                  ? `${systemInstructions}\n\n`
-                  : '';
-
-                const res = await llm.invoke(
-                  `${systemPrompt}You are a web search summarizer, tasked with summarizing a piece of text retrieved from a web search. Your job is to summarize the 
-            text into a detailed, 2-4 paragraph explanation that captures the main ideas and provides a comprehensive answer to the query.
-            If the query is \"summarize\", you should provide a detailed summary of the text. If the query is a specific question, you should answer it in the summary.
-            
-            - **Journalistic tone**: The summary should sound professional and journalistic, not too casual or vague.
-            - **Thorough and detailed**: Ensure that every key point from the text is captured and that the summary directly answers the query.
-            - **Not too lengthy, but detailed**: The summary should be informative but not excessively long. Focus on providing detailed information in a concise format.
-
-            The text will be shared inside the \`text\` XML tag, and the query inside the \`query\` XML tag.
-
-            <example>
-            1. \`<text>
-            Docker is a set of platform-as-a-service products that use OS-level virtualization to deliver software in packages called containers. 
-            It was first released in 2013 and is developed by Docker, Inc. Docker is designed to make it easier to create, deploy, and run applications 
-            by using containers.
-            </text>
-
-            <query>
-            What is Docker and how does it work?
-            </query>
-
-            Response:
-            Docker is a revolutionary platform-as-a-service product developed by Docker, Inc., that uses container technology to make application 
-            deployment more efficient. It allows developers to package their software with all necessary dependencies, making it easier to run in 
-            any environment. Released in 2013, Docker has transformed the way applications are built, deployed, and managed.
-            \`
-            2. \`<text>
-            The theory of relativity, or simply relativity, encompasses two interrelated theories of Albert Einstein: special relativity and general
-            relativity. However, the word "relativity" is sometimes used in reference to Galilean invariance. The term "theory of relativity" was based
-            on the expression "relative theory" used by Max Planck in 1906. The theory of relativity usually encompasses two interrelated theories by
-            Albert Einstein: special relativity and general relativity. Special relativity applies to all physical phenomena in the absence of gravity.
-            General relativity explains the law of gravitation and its relation to other forces of nature. It applies to the cosmological and astrophysical
-            realm, including astronomy.
-            </text>
-
-            <query>
-            summarize
-            </query>
-
-            Response:
-            The theory of relativity, developed by Albert Einstein, encompasses two main theories: special relativity and general relativity. Special
-            relativity applies to all physical phenomena in the absence of gravity, while general relativity explains the law of gravitation and its
-            relation to other forces of nature. The theory of relativity is based on the concept of "relative theory," as introduced by Max Planck in
-            1906. It is a fundamental theory in physics that has revolutionized our understanding of the universe.
-            \`
-            </example>
-
-            Everything below is the actual data you will be working with. Good luck!
-
-            <query>
-            ${question}
-            </query>
-
-            <text>
-            ${doc.pageContent}
-            </text>
-
-            Make sure to answer the query in the summary.
-          `,
-                  { signal },
-                );
-
-                const document = new Document({
-                  pageContent: res.content as string,
-                  metadata: {
-                    title: doc.metadata.title,
-                    url: doc.metadata.url,
-                  },
-                });
-
-                docs.push(document);
-              }),
-            );
-
-            return { query: question, docs: docs };
-          } else {
-            if (this.config.additionalSearchCriteria) {
-              question = `${question} ${this.config.additionalSearchCriteria}`;
-            }
-            this.emitProgress(
-              emitter,
-              20,
-              `Searching the web`,
-              `Search Query: ${question}`,
-            );
-
-            const searxngResult = await searchSearxng(question, {
-              language: 'en',
-              engines: this.config.activeEngines,
-            });
-
-            // Store the SearXNG URL for later use in emitting to the client
-            this.searxngUrl = searxngResult.searchUrl;
-
-            const documents = searxngResult.results.map(
-              (result) =>
-                new Document({
-                  pageContent:
-                    result.content ||
-                    (this.config.activeEngines.includes('youtube')
-                      ? result.title
-                      : '') /* Todo: Implement transcript grabbing using Youtubei (source: https://www.npmjs.com/package/youtubei) */,
-                  metadata: {
-                    title: result.title,
-                    url: result.url,
-                    ...(result.img_src && { img_src: result.img_src }),
-                  },
-                }),
-            );
-
-            return { query: question, docs: documents, searchQuery: question };
-          }
-        } catch (error) {
-          console.error('Error in search retriever chain:', error);
-          emitter.emit('error', JSON.stringify({ data: error }));
-          throw error;
-        }
-      }),
-    ]);
-  }
-
-  private async createAnsweringChain(
-    llm: BaseChatModel,
-    fileIds: string[],
-    embeddings: Embeddings,
-    optimizationMode: 'speed' | 'balanced' | 'agent',
-    systemInstructions: string,
-    signal: AbortSignal,
-    emitter: eventEmitter,
-    personaInstructions?: string,
-  ) {
-    return RunnableSequence.from([
-      RunnableMap.from({
-        systemInstructions: () => systemInstructions,
-        query: (input: BasicChainInput) => input.query,
-        chat_history: (input: BasicChainInput) => input.chat_history,
-        date: () => formatDateForLLM(),
-        personaInstructions: () => personaInstructions || '',
-        context: RunnableLambda.from(
-          async (
-            input: BasicChainInput,
-            options?: { signal?: AbortSignal },
-          ) => {
-            // Check if the request was aborted
-            if (options?.signal?.aborted || signal?.aborted) {
-              console.log('Request cancelled by user');
-              throw new Error('Request cancelled by user');
-            }
-
-            const processedHistory = formatChatHistoryAsString(
-              input.chat_history,
-            );
-
-            let docs: Document[] | null = null;
-            let query = input.query;
-
-            if (this.config.searchWeb) {
-              const searchRetrieverChain =
-                await this.createSearchRetrieverChain(
-                  llm,
-                  systemInstructions,
-                  emitter,
-                  signal,
-                );
-              var date = formatDateForLLM();
-
-              const searchRetrieverResult = await searchRetrieverChain.invoke(
-                {
-                  chat_history: processedHistory,
-                  query,
-                  date,
-                  systemInstructions,
-                },
-                { signal: options?.signal },
-              );
-
-              query = searchRetrieverResult.query;
-              docs = searchRetrieverResult.docs;
-
-              // Store the search query in the context for emitting to the client
-              if (searchRetrieverResult.searchQuery) {
-                this.searchQuery = searchRetrieverResult.searchQuery;
-              }
-            }
-
-            const sortedDocs = await this.rerankDocs(
-              query,
-              docs ?? [],
-              fileIds,
-              embeddings,
-              optimizationMode,
-              llm,
-              systemInstructions,
-              emitter,
-              signal,
-            );
-
-            if (options?.signal?.aborted || signal?.aborted) {
-              console.log('Request cancelled by user');
-              throw new Error('Request cancelled by user');
-            }
-
-            this.emitProgress(emitter, 100, `Done`);
-            return sortedDocs;
-          },
-        )
-          .withConfig({
-            runName: 'FinalSourceRetriever',
-          })
-          .pipe(this.processDocs),
-      }),
-      ChatPromptTemplate.fromMessages([
-        ['system', this.config.responsePrompt],
-        new MessagesPlaceholder('chat_history'),
-        ['user', '{query}'],
-      ]),
-      llm,
-      this.strParser,
-    ]).withConfig({
-      runName: 'FinalResponseGenerator',
-    });
-  }
-
-  private async rerankDocs(
-    query: string,
-    docs: Document[],
-    fileIds: string[],
-    embeddings: Embeddings,
-    optimizationMode: 'speed' | 'balanced' | 'agent',
-    llm: BaseChatModel,
-    systemInstructions: string,
-    emitter: eventEmitter,
-    signal: AbortSignal,
-  ): Promise<Document[]> {
-    try {
-      if (docs.length === 0 && fileIds.length === 0) {
-        return docs;
-      }
-
-      if (query.toLocaleLowerCase() === 'summarize') {
-        return docs.slice(0, 15);
-      }
-
-      const filesData = fileIds
-        .map((file) => {
-          const filePath = path.join(process.cwd(), 'uploads', file);
-
-          const contentPath = filePath + '-extracted.json';
-          const embeddingsPath = filePath + '-embeddings.json';
-
-          const content = JSON.parse(fs.readFileSync(contentPath, 'utf8'));
-          const embeddings = JSON.parse(
-            fs.readFileSync(embeddingsPath, 'utf8'),
-          );
-
-          const fileSimilaritySearchObject = content.contents.map(
-            (c: string, i: number) => {
-              return {
-                fileName: content.title,
-                content: c,
-                embeddings: embeddings.embeddings[i],
-              };
-            },
-          );
-
-          return fileSimilaritySearchObject;
-        })
-        .flat();
-
-      let docsWithContent = docs.filter(
-        (doc) => doc.pageContent && doc.pageContent.length > 0,
-      );
-
-      const queryEmbedding = await embeddings.embedQuery(query);
-
-      const getRankedDocs = async (
-        queryEmbedding: number[],
-        includeFiles: boolean,
-        includeNonFileDocs: boolean,
-        maxDocs: number,
-      ) => {
-        let docsToRank = includeNonFileDocs ? docsWithContent : [];
-
-        if (includeFiles) {
-          // Add file documents to the ranking
-          const fileDocs = filesData.map((fileData) => {
-            return new Document({
-              pageContent: fileData.content,
-              metadata: {
-                title: fileData.fileName,
-                url: `File`,
-                embeddings: fileData.embeddings,
-              },
-            });
-          });
-          docsToRank.push(...fileDocs);
-        }
-
-        const similarity = await Promise.all(
-          docsToRank.map(async (doc, i) => {
-            const sim = computeSimilarity(
-              queryEmbedding,
-              doc.metadata?.embeddings
-                ? doc.metadata?.embeddings
-                : (await embeddings.embedDocuments([doc.pageContent]))[0],
-            );
-            return {
-              index: i,
-              similarity: sim,
-            };
-          }),
-        );
-
-        let rankedDocs = similarity
-          .filter(
-            (sim) => sim.similarity > (this.config.rerankThreshold ?? 0.3),
-          )
-          .sort((a, b) => b.similarity - a.similarity)
-          .map((sim) => docsToRank[sim.index]);
-
-        rankedDocs =
-          docsToRank.length > 0 ? rankedDocs.slice(0, maxDocs) : rankedDocs;
-        return rankedDocs;
-      };
-      if (optimizationMode === 'speed' || this.config.rerank === false) {
-        this.emitProgress(
-          emitter,
-          50,
-          `Ranking sources`,
-          this.searchQuery ? `Search Query: ${this.searchQuery}` : undefined,
-        );
-        if (filesData.length > 0) {
-          const sortedFiles = await getRankedDocs(
-            queryEmbedding,
-            true,
-            false,
-            8,
-          );
-
-          return [
-            ...sortedFiles,
-            ...docsWithContent.slice(0, 15 - sortedFiles.length),
-          ];
-        } else {
-          return docsWithContent.slice(0, 15);
-        }
-      } else if (optimizationMode === 'balanced') {
-        this.emitProgress(
-          emitter,
-          40,
-          `Ranking sources`,
-          this.searchQuery ? `Search Query: ${this.searchQuery}` : undefined,
-        );
-        // Get the top ranked attached files, if any
-        let sortedDocs = await getRankedDocs(queryEmbedding, true, false, 8);
-
-        sortedDocs = [
-          ...sortedDocs,
-          ...docsWithContent.slice(0, 15 - sortedDocs.length),
-        ];
-
-        this.emitProgress(
-          emitter,
-          60,
-          `Enriching sources`,
-          this.searchQuery ? `Search Query: ${this.searchQuery}` : undefined,
-        );
-        sortedDocs = await Promise.all(
-          sortedDocs.map(async (doc) => {
-            const webContent = await getWebContent(doc.metadata.url);
-            const chunks =
-              webContent?.pageContent
-                .match(/.{1,500}/g)
-                ?.map((chunk) => chunk.trim()) || [];
-            const chunkEmbeddings = await embeddings.embedDocuments(chunks);
-            const similarities = chunkEmbeddings.map((chunkEmbedding) => {
-              return computeSimilarity(queryEmbedding, chunkEmbedding);
-            });
-
-            const topChunks = similarities
-              .map((similarity, index) => ({ similarity, index }))
-              .sort((a, b) => b.similarity - a.similarity)
-              .slice(0, 5)
-              .map((chunk) => chunks[chunk.index]);
-            const excerpt = topChunks.join('\n\n');
-
-            let newDoc = {
-              ...doc,
-              pageContent: excerpt
-                ? `${excerpt}\n\n${doc.pageContent}`
-                : doc.pageContent,
-            };
-            return newDoc;
-          }),
-        );
-
-        return sortedDocs;
-      }
-    } catch (error) {
-      console.error('Error in rerankDocs:', error);
-      emitter.emit('error', JSON.stringify({ data: error }));
-    }
-    return [];
-  }
-
-  private processDocs(docs: Document[]) {
-    const fullDocs = docs
-      .map(
-        (_, index) =>
-          `<${index + 1}>\n
-<title>${docs[index].metadata.title}</title>\n
-${docs[index].metadata?.url.toLowerCase().includes('file') ? '' : '\n<url>' + docs[index].metadata.url + '</url>\n'}
-<content>\n${docs[index].pageContent}\n</content>\n
-</${index + 1}>\n`,
-      )
-      .join('\n');
-    console.log('Processed docs:', fullDocs);
-    return fullDocs;
-  }
-
-  private async handleStream(
-    stream: AsyncGenerator<StreamEvent, any, any>,
-    emitter: eventEmitter,
-    llm: BaseChatModel,
-    signal: AbortSignal,
-  ) {
-    if (signal.aborted) {
-      return;
-    }
-
-    for await (const event of stream) {
-      if (signal.aborted) {
-        return;
-      }
-
-      if (
-        event.event === 'on_chain_end' &&
-        event.name === 'FinalSourceRetriever'
-      ) {
-        const sourcesData = event.data.output;
-        if (this.searchQuery) {
-          emitter.emit(
-            'data',
-            JSON.stringify({
-              type: 'sources',
-              data: sourcesData,
-              searchQuery: this.searchQuery,
-              searchUrl: this.searxngUrl,
-            }),
-          );
-        } else {
-          emitter.emit(
-            'data',
-            JSON.stringify({ type: 'sources', data: sourcesData }),
-          );
-        }
-      }
-      if (
-        event.event === 'on_chain_stream' &&
-        event.name === 'FinalResponseGenerator'
-      ) {
-        emitter.emit(
-          'data',
-          JSON.stringify({ type: 'response', data: event.data.chunk }),
-        );
-      }
-      if (
-        event.event === 'on_chain_end' &&
-        event.name === 'FinalResponseGenerator'
-      ) {
-        const modelName = getModelName(llm);
-
-        // Send model info before ending
-        emitter.emit(
-          'stats',
-          JSON.stringify({
-            type: 'modelStats',
-            data: {
-              modelName,
-            },
-          }),
-        );
-
-        emitter.emit('end');
-      }
-    }
-  }
-
   /**
    * Execute agent workflow asynchronously with proper streaming support
    */
@@ -719,7 +86,7 @@ ${docs[index].metadata?.url.toLowerCase().includes('file') ? '' : '\n<url>' + do
     history: BaseMessage[],
     llm: BaseChatModel,
     embeddings: Embeddings,
-    optimizationMode: 'speed' | 'balanced' | 'agent',
+    optimizationMode: 'speed' | 'agent',
     fileIds: string[],
     systemInstructions: string,
     signal: AbortSignal,
@@ -728,9 +95,22 @@ ${docs[index].metadata?.url.toLowerCase().includes('file') ? '' : '\n<url>' + do
   ) {
     const emitter = new eventEmitter();
 
-    // Branch to agent search if optimization mode is 'agent'
+    // Branch to speed search if optimization mode is 'speed'
-    if (optimizationMode === 'agent') {
+    if (optimizationMode === 'speed') {
-      // Execute agent workflow asynchronously to maintain streaming
+      const speedSearchAgent = new SpeedSearchAgent(this.config);
+      return speedSearchAgent.searchAndAnswer(
+        message,
+        history,
+        llm,
+        embeddings,
+        systemInstructions,
+        signal,
+        personaInstructions,
+        focusMode,
+      );
+    }
+
+    // Execute agent workflow for 'agent' mode
     this.executeAgentWorkflow(
       llm,
       embeddings,
@@ -743,34 +123,6 @@ ${docs[index].metadata?.url.toLowerCase().includes('file') ? '' : '\n<url>' + do
       signal,
       focusMode || 'webSearch',
     );
-      return emitter;
-    }
-
-    // Existing logic for other optimization modes
-    const answeringChain = await this.createAnsweringChain(
-      llm,
-      fileIds,
-      embeddings,
-      optimizationMode,
-      systemInstructions,
-      signal,
-      emitter,
-      personaInstructions,
-    );
-
-    const stream = answeringChain.streamEvents(
-      {
-        chat_history: history,
-        query: message,
-      },
-      {
-        version: 'v1',
-        // Pass the abort signal to the LLM streaming chain
-        signal,
-      },
-    );
-
-    this.handleStream(stream, emitter, llm, signal);
 
     return emitter;
   }

src/lib/search/speedSearch.ts

@@ -0,0 +1,560 @@
+import type { Embeddings } from '@langchain/core/embeddings';
+import type { BaseChatModel } from '@langchain/core/language_models/chat_models';
+import { BaseMessage } from '@langchain/core/messages';
+import { StringOutputParser } from '@langchain/core/output_parsers';
+import {
+  ChatPromptTemplate,
+  MessagesPlaceholder,
+  PromptTemplate,
+} from '@langchain/core/prompts';
+import {
+  RunnableLambda,
+  RunnableMap,
+  RunnableSequence,
+} from '@langchain/core/runnables';
+import { StreamEvent } from '@langchain/core/tracers/log_stream';
+import { ChatOpenAI } from '@langchain/openai';
+import eventEmitter from 'events';
+import { Document } from 'langchain/document';
+import LineOutputParser from '../outputParsers/lineOutputParser';
+import LineListOutputParser from '../outputParsers/listLineOutputParser';
+import { searchSearxng } from '../searxng';
+import { formatDateForLLM } from '../utils';
+import { getDocumentsFromLinks } from '../utils/documents';
+import formatChatHistoryAsString from '../utils/formatHistory';
+import { getModelName } from '../utils/modelUtils';
+
+export interface SpeedSearchAgentType {
+  searchAndAnswer: (
+    message: string,
+    history: BaseMessage[],
+    llm: BaseChatModel,
+    embeddings: Embeddings,
+    systemInstructions: string,
+    signal: AbortSignal,
+    personaInstructions?: string,
+    focusMode?: string,
+  ) => Promise<eventEmitter>;
+}
+
+interface Config {
+  searchWeb: boolean;
+  rerank: boolean;
+  summarizer: boolean;
+  rerankThreshold: number;
+  queryGeneratorPrompt: string;
+  responsePrompt: string;
+  activeEngines: string[];
+  additionalSearchCriteria?: string;
+}
+
+type BasicChainInput = {
+  chat_history: BaseMessage[];
+  query: string;
+};
+
+class SpeedSearchAgent implements SpeedSearchAgentType {
+  private config: Config;
+  private strParser = new StringOutputParser();
+  private searchQuery?: string;
+  private searxngUrl?: string;
+
+  constructor(config: Config) {
+    this.config = config;
+  }
+
+  /**
+   * Emit a progress event with the given percentage and message
+   */
+  private emitProgress(
+    emitter: eventEmitter,
+    percentage: number,
+    message: string,
+    subMessage?: string,
+  ) {
+    const progressData: any = {
+      message,
+      current: percentage,
+      total: 100,
+    };
+
+    // Add subMessage if provided
+    if (subMessage) {
+      progressData.subMessage = subMessage;
+    }
+
+    emitter.emit(
+      'progress',
+      JSON.stringify({
+        type: 'progress',
+        data: progressData,
+      }),
+    );
+  }
+
+  private async createSearchRetrieverChain(
+    llm: BaseChatModel,
+    systemInstructions: string,
+    emitter: eventEmitter,
+    signal: AbortSignal,
+  ) {
+    // TODO: Don't we want to set this back to default once search is done?
+    (llm as unknown as ChatOpenAI).temperature = 0;
+
+    this.emitProgress(emitter, 10, `Building search query`);
+
+    return RunnableSequence.from([
+      PromptTemplate.fromTemplate(this.config.queryGeneratorPrompt),
+      llm,
+      this.strParser,
+      RunnableLambda.from(async (input: string) => {
+        try {
+          //console.log(`LLM response for initial web search:"${input}"`);
+          const linksOutputParser = new LineListOutputParser({
+            key: 'links',
+          });
+
+          const questionOutputParser = new LineOutputParser({
+            key: 'answer',
+          });
+
+          const links = await linksOutputParser.parse(input);
+          let question = await questionOutputParser.parse(input);
+
+          //console.log('question', question);
+
+          if (question === 'not_needed') {
+            return { query: '', docs: [] };
+          }
+
+          if (links.length > 0) {
+            if (question.length === 0) {
+              question = 'summarize';
+            }
+
+            let docs: Document[] = [];
+
+            const linkDocs = await getDocumentsFromLinks({ links });
+
+            const docGroups: Document[] = [];
+
+            linkDocs.map((doc) => {
+              const URLDocExists = docGroups.find(
+                (d) =>
+                  d.metadata.url === doc.metadata.url &&
+                  d.metadata.totalDocs < 10,
+              );
+
+              if (!URLDocExists) {
+                docGroups.push({
+                  ...doc,
+                  metadata: {
+                    ...doc.metadata,
+                    totalDocs: 1,
+                  },
+                });
+              }
+
+              const docIndex = docGroups.findIndex(
+                (d) =>
+                  d.metadata.url === doc.metadata.url &&
+                  d.metadata.totalDocs < 10,
+              );
+
+              if (docIndex !== -1) {
+                docGroups[docIndex].pageContent =
+                  docGroups[docIndex].pageContent + `\n\n` + doc.pageContent;
+                docGroups[docIndex].metadata.totalDocs += 1;
+              }
+            });
+
+            this.emitProgress(emitter, 20, `Summarizing content`);
+
+            await Promise.all(
+              docGroups.map(async (doc) => {
+                const systemPrompt = systemInstructions
+                  ? `${systemInstructions}\n\n`
+                  : '';
+
+                const res = await llm.invoke(
+                  `${systemPrompt}You are a web search summarizer, tasked with summarizing a piece of text retrieved from a web search. Your job is to summarize the 
+            text into a detailed, 2-4 paragraph explanation that captures the main ideas and provides a comprehensive answer to the query.
+            If the query is \"summarize\", you should provide a detailed summary of the text. If the query is a specific question, you should answer it in the summary.
+            
+            - **Journalistic tone**: The summary should sound professional and journalistic, not too casual or vague.
+            - **Thorough and detailed**: Ensure that every key point from the text is captured and that the summary directly answers the query.
+            - **Not too lengthy, but detailed**: The summary should be informative but not excessively long. Focus on providing detailed information in a concise format.
+
+            The text will be shared inside the \`text\` XML tag, and the query inside the \`query\` XML tag.
+
+            <example>
+            1. \`<text>
+            Docker is a set of platform-as-a-service products that use OS-level virtualization to deliver software in packages called containers. 
+            It was first released in 2013 and is developed by Docker, Inc. Docker is designed to make it easier to create, deploy, and run applications 
+            by using containers.
+            </text>
+
+            <query>
+            What is Docker and how does it work?
+            </query>
+
+            Response:
+            Docker is a revolutionary platform-as-a-service product developed by Docker, Inc., that uses container technology to make application 
+            deployment more efficient. It allows developers to package their software with all necessary dependencies, making it easier to run in 
+            any environment. Released in 2013, Docker has transformed the way applications are built, deployed, and managed.
+            \`
+            2. \`<text>
+            The theory of relativity, or simply relativity, encompasses two interrelated theories of Albert Einstein: special relativity and general
+            relativity. However, the word "relativity" is sometimes used in reference to Galilean invariance. The term "theory of relativity" was based
+            on the expression "relative theory" used by Max Planck in 1906. The theory of relativity usually encompasses two interrelated theories by
+            Albert Einstein: special relativity and general relativity. Special relativity applies to all physical phenomena in the absence of gravity.
+            General relativity explains the law of gravitation and its relation to other forces of nature. It applies to the cosmological and astrophysical
+            realm, including astronomy.
+            </text>
+
+            <query>
+            summarize
+            </query>
+
+            Response:
+            The theory of relativity, developed by Albert Einstein, encompasses two main theories: special relativity and general relativity. Special
+            relativity applies to all physical phenomena in the absence of gravity, while general relativity explains the law of gravitation and its
+            relation to other forces of nature. The theory of relativity is based on the concept of "relative theory," as introduced by Max Planck in
+            1906. It is a fundamental theory in physics that has revolutionized our understanding of the universe.
+            \`
+            </example>
+
+            Everything below is the actual data you will be working with. Good luck!
+
+            <query>
+            ${question}
+            </query>
+
+            <text>
+            ${doc.pageContent}
+            </text>
+
+            Make sure to answer the query in the summary.
+          `,
+                  { signal },
+                );
+
+                const document = new Document({
+                  pageContent: res.content as string,
+                  metadata: {
+                    title: doc.metadata.title,
+                    url: doc.metadata.url,
+                  },
+                });
+
+                docs.push(document);
+              }),
+            );
+
+            return { query: question, docs: docs };
+          } else {
+            if (this.config.additionalSearchCriteria) {
+              question = `${question} ${this.config.additionalSearchCriteria}`;
+            }
+            this.emitProgress(
+              emitter,
+              20,
+              `Searching the web`,
+              `Search Query: ${question}`,
+            );
+
+            const searxngResult = await searchSearxng(question, {
+              language: 'en',
+              engines: this.config.activeEngines,
+            });
+
+            // Store the SearXNG URL for later use in emitting to the client
+            this.searxngUrl = searxngResult.searchUrl;
+
+            const documents = searxngResult.results.map(
+              (result) =>
+                new Document({
+                  pageContent:
+                    result.content ||
+                    (this.config.activeEngines.includes('youtube')
+                      ? result.title
+                      : '') /* Todo: Implement transcript grabbing using Youtubei (source: https://www.npmjs.com/package/youtubei) */,
+                  metadata: {
+                    title: result.title,
+                    url: result.url,
+                    ...(result.img_src && { img_src: result.img_src }),
+                  },
+                }),
+            );
+
+            return { query: question, docs: documents, searchQuery: question };
+          }
+        } catch (error) {
+          console.error('Error in search retriever chain:', error);
+          emitter.emit('error', JSON.stringify({ data: error }));
+          throw error;
+        }
+      }),
+    ]);
+  }
+
+  private async createAnsweringChain(
+    llm: BaseChatModel,
+    embeddings: Embeddings,
+    systemInstructions: string,
+    signal: AbortSignal,
+    emitter: eventEmitter,
+    personaInstructions?: string,
+  ) {
+    return RunnableSequence.from([
+      RunnableMap.from({
+        systemInstructions: () => systemInstructions,
+        query: (input: BasicChainInput) => input.query,
+        chat_history: (input: BasicChainInput) => input.chat_history,
+        date: () => formatDateForLLM(),
+        personaInstructions: () => personaInstructions || '',
+        context: RunnableLambda.from(
+          async (
+            input: BasicChainInput,
+            options?: { signal?: AbortSignal },
+          ) => {
+            // Check if the request was aborted
+            if (options?.signal?.aborted || signal?.aborted) {
+              console.log('Request cancelled by user');
+              throw new Error('Request cancelled by user');
+            }
+
+            const processedHistory = formatChatHistoryAsString(
+              input.chat_history,
+            );
+
+            let docs: Document[] | null = null;
+            let query = input.query;
+
+            if (this.config.searchWeb) {
+              const searchRetrieverChain =
+                await this.createSearchRetrieverChain(
+                  llm,
+                  systemInstructions,
+                  emitter,
+                  signal,
+                );
+              var date = formatDateForLLM();
+
+              const searchRetrieverResult = await searchRetrieverChain.invoke(
+                {
+                  chat_history: processedHistory,
+                  query,
+                  date,
+                  systemInstructions,
+                },
+                { signal: options?.signal },
+              );
+
+              query = searchRetrieverResult.query;
+              docs = searchRetrieverResult.docs;
+
+              // Store the search query in the context for emitting to the client
+              if (searchRetrieverResult.searchQuery) {
+                this.searchQuery = searchRetrieverResult.searchQuery;
+              }
+            }
+
+            const sortedDocs = await this.rerankDocsForSpeed(
+              query,
+              docs ?? [],
+              embeddings,
+              emitter,
+              signal,
+            );
+
+            if (options?.signal?.aborted || signal?.aborted) {
+              console.log('Request cancelled by user');
+              throw new Error('Request cancelled by user');
+            }
+
+            this.emitProgress(emitter, 100, `Done`);
+            return sortedDocs;
+          },
+        )
+          .withConfig({
+            runName: 'FinalSourceRetriever',
+          })
+          .pipe(this.processDocs),
+      }),
+      ChatPromptTemplate.fromMessages([
+        ['system', this.config.responsePrompt],
+        new MessagesPlaceholder('chat_history'),
+        ['user', '{query}'],
+      ]),
+      llm,
+      this.strParser,
+    ]).withConfig({
+      runName: 'FinalResponseGenerator',
+    });
+  }
+
+  /**
+   * Speed-optimized document reranking with simplified logic for web results only
+   */
+  private async rerankDocsForSpeed(
+    query: string,
+    docs: Document[],
+    embeddings: Embeddings,
+    emitter: eventEmitter,
+    signal: AbortSignal,
+  ): Promise<Document[]> {
+    try {
+      if (docs.length === 0) {
+        return docs;
+      }
+
+      if (query.toLocaleLowerCase() === 'summarize') {
+        return docs.slice(0, 15);
+      }
+
+      // Filter out documents with no content
+      let docsWithContent = docs.filter(
+        (doc) => doc.pageContent && doc.pageContent.length > 0,
+      );
+
+      // Speed mode logic - simply return first 15 documents with content
+      // No similarity ranking to prioritize speed
+      this.emitProgress(
+        emitter,
+        50,
+        `Ranking sources`,
+        this.searchQuery ? `Search Query: ${this.searchQuery}` : undefined,
+      );
+
+      return docsWithContent.slice(0, 15);
+    } catch (error) {
+      console.error('Error in rerankDocsForSpeed:', error);
+      emitter.emit('error', JSON.stringify({ data: error }));
+    }
+    return [];
+  }
+
+  private processDocs(docs: Document[]) {
+    const fullDocs = docs
+      .map(
+        (_, index) =>
+          `<${index + 1}>\n
+<title>${docs[index].metadata.title}</title>\n
+${docs[index].metadata?.url.toLowerCase().includes('file') ? '' : '\n<url>' + docs[index].metadata.url + '</url>\n'}
+<content>\n${docs[index].pageContent}\n</content>\n
+</${index + 1}>\n`,
+      )
+      .join('\n');
+    console.log('Processed docs:', fullDocs);
+    return fullDocs;
+  }
+
+  private async handleStream(
+    stream: AsyncGenerator<StreamEvent, any, any>,
+    emitter: eventEmitter,
+    llm: BaseChatModel,
+    signal: AbortSignal,
+  ) {
+    if (signal.aborted) {
+      return;
+    }
+
+    for await (const event of stream) {
+      if (signal.aborted) {
+        return;
+      }
+
+      if (
+        event.event === 'on_chain_end' &&
+        event.name === 'FinalSourceRetriever'
+      ) {
+        const sourcesData = event.data.output;
+        if (this.searchQuery) {
+          emitter.emit(
+            'data',
+            JSON.stringify({
+              type: 'sources',
+              data: sourcesData,
+              searchQuery: this.searchQuery,
+              searchUrl: this.searxngUrl,
+            }),
+          );
+        } else {
+          emitter.emit(
+            'data',
+            JSON.stringify({ type: 'sources', data: sourcesData }),
+          );
+        }
+      }
+      if (
+        event.event === 'on_chain_stream' &&
+        event.name === 'FinalResponseGenerator'
+      ) {
+        emitter.emit(
+          'data',
+          JSON.stringify({ type: 'response', data: event.data.chunk }),
+        );
+      }
+      if (
+        event.event === 'on_chain_end' &&
+        event.name === 'FinalResponseGenerator'
+      ) {
+        const modelName = getModelName(llm);
+
+        // Send model info before ending
+        emitter.emit(
+          'stats',
+          JSON.stringify({
+            type: 'modelStats',
+            data: {
+              modelName,
+            },
+          }),
+        );
+
+        emitter.emit('end');
+      }
+    }
+  }
+
+  async searchAndAnswer(
+    message: string,
+    history: BaseMessage[],
+    llm: BaseChatModel,
+    embeddings: Embeddings,
+    systemInstructions: string,
+    signal: AbortSignal,
+    personaInstructions?: string,
+    focusMode?: string,
+  ) {
+    const emitter = new eventEmitter();
+
+    const answeringChain = await this.createAnsweringChain(
+      llm,
+      embeddings,
+      systemInstructions,
+      signal,
+      emitter,
+      personaInstructions,
+    );
+
+    const stream = answeringChain.streamEvents(
+      {
+        chat_history: history,
+        query: message,
+      },
+      {
+        version: 'v1',
+        // Pass the abort signal to the LLM streaming chain
+        signal,
+      },
+    );
+
+    this.handleStream(stream, emitter, llm, signal);
+
+    return emitter;
+  }
+}
+
+export default SpeedSearchAgent;

src/lib/utils/fileProcessing.ts

@@ -17,7 +17,9 @@ export interface FileData {
  * @param fileIds Array of file IDs to process
  * @returns Array of Document objects with content and embeddings
  */
-export async function processFilesToDocuments(fileIds: string[]): Promise<Document[]> {
+export async function processFilesToDocuments(
+  fileIds: string[],
+): Promise<Document[]> {
   if (fileIds.length === 0) {
     return [];
   }

testAttachments/sporting-events.txt

@@ -0,0 +1,3 @@
+Who won the 2025 Super Bowl?
+Who won the 2023 Formula One Driver's Championship?
+Who won the 2022 World Cup?