🔍 feat: Web Search via OpenAI Responses API

api/server/services/Endpoints/agents/agent.js

@@ -85,7 +85,7 @@ const initializeAgent = async ({
   });
 
   const provider = agent.provider;
-  const { tools, toolContextMap } =
+  const { tools: structuredTools, toolContextMap } =
     (await loadTools?.({
       req,
       res,
@@ -140,12 +140,20 @@ const initializeAgent = async ({
     agent.provider = options.provider;
   }
 
+  /** @type {import('@librechat/agents').GenericTool[]} */
+  let tools = options.tools ?? structuredTools;
   if (
     (agent.provider === Providers.GOOGLE || agent.provider === Providers.VERTEXAI) &&
-    options?.tools?.length &&
-    tools?.length
+    options.tools?.length &&
+    structuredTools?.length
   ) {
     throw new Error(`{ "type": "${ErrorTypes.GOOGLE_TOOL_CONFLICT}"}`);
+  } else if (
+    (agent.provider === Providers.OPENAI || agent.provider === Providers.AZURE) &&
+    options.tools?.length &&
+    structuredTools?.length
+  ) {
+    tools = structuredTools.concat(options.tools);
   }
 
   /** @type {import('@librechat/agents').ClientOptions} */
@@ -173,7 +181,7 @@ const initializeAgent = async ({
     attachments,
     resendFiles,
     toolContextMap,
-    tools: options.tools ?? tools,
+    tools,
     maxContextTokens: (agentMaxContextTokens - maxTokens) * 0.9,
   };
 };

client/src/components/SidePanel/Parameters/DynamicSlider.tsx

@@ -197,7 +197,7 @@ function DynamicSlider({
                   defaultTextProps,
                   cn(
                     optionText,
-                    'reset-rc-number-input reset-rc-number-input-text-right h-auto w-12 border-0 py-1 text-xs group-hover/temp:border-gray-200',
+                    'reset-rc-number-input h-auto w-14 border-0 py-1 pl-1 text-center text-xs group-hover/temp:border-gray-200',
                   ),
                 )}
               />

client/src/hooks/Conversations/useSetIndexOptions.ts

@@ -30,6 +30,14 @@ const useSetIndexOptions: TUseSetOptions = (preset = false) => {
       };
     }
 
+    // Auto-enable Responses API when web search is enabled
+    if (param === 'web_search' && newValue === true) {
+      const currentUseResponsesApi = conversation?.useResponsesApi ?? false;
+      if (!currentUseResponsesApi) {
+        update['useResponsesApi'] = true;
+      }
+    }
+
     setConversation(
       (prevState) =>
         tConvoUpdateSchema.parse({

client/src/locales/en/translation.json

@@ -234,6 +234,7 @@
   "com_endpoint_openai_temp": "Higher values = more random, while lower values = more focused and deterministic. We recommend altering this or Top P but not both.",
   "com_endpoint_openai_topp": "An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We recommend altering this or temperature but not both.",
   "com_endpoint_openai_use_responses_api": "Use the Responses API instead of Chat Completions, which includes extended features from OpenAI. Required for o1-pro, o3-pro, and to enable reasoning summaries.",
+  "com_endpoint_openai_use_web_search": "Enable web search functionality using OpenAI's built-in search capabilities. This allows the model to search the web for up-to-date information and provide more accurate, current responses.",
   "com_endpoint_output": "Output",
   "com_endpoint_plug_image_detail": "Image Detail",
   "com_endpoint_plug_resend_files": "Resend Files",

packages/api/src/endpoints/openai/llm.ts

@@ -1,5 +1,6 @@
 import { ProxyAgent } from 'undici';
 import { KnownEndpoints, removeNullishValues } from 'librechat-data-provider';
+import type { BindToolsInput } from '@langchain/core/language_models/chat_models';
 import type { AzureOpenAIInput } from '@langchain/openai';
 import type { OpenAI } from 'openai';
 import type * as t from '~/types';
@@ -176,6 +177,13 @@ export function getOpenAIConfig(
     delete llmConfig.max_tokens;
   }
 
+  const tools: BindToolsInput[] = [];
+
+  if (modelOptions.web_search) {
+    llmConfig.useResponsesApi = true;
+    tools.push({ type: 'web_search_preview' });
+  }
+
   /**
    * Note: OpenAI Web Search models do not support any known parameters besides `max_tokens`
    */
@@ -216,5 +224,6 @@ export function getOpenAIConfig(
   return {
     llmConfig,
     configOptions,
+    tools,
   };
 }

packages/api/src/types/openai.ts

@@ -1,6 +1,7 @@
 import { z } from 'zod';
 import { openAISchema, EModelEndpoint } from 'librechat-data-provider';
 import type { TEndpointOption, TAzureConfig, TEndpoint } from 'librechat-data-provider';
+import type { BindToolsInput } from '@langchain/core/language_models/chat_models';
 import type { OpenAIClientOptions } from '@librechat/agents';
 import type { AzureOptions } from './azure';
 
@@ -33,6 +34,7 @@ export type ClientOptions = OpenAIClientOptions & {
 export interface LLMConfigResult {
   llmConfig: ClientOptions;
   configOptions: OpenAIConfiguration;
+  tools?: BindToolsInput[];
 }
 
 /**

packages/data-provider/src/parameterSettings.ts

@@ -247,6 +247,19 @@ const openAIParams: Record<string, SettingDefinition> = {
     showDefault: false,
     columnSpan: 2,
   },
+  web_search: {
+    key: 'web_search',
+    label: 'com_ui_web_search',
+    labelCode: true,
+    description: 'com_endpoint_openai_use_web_search',
+    descriptionCode: true,
+    type: 'boolean',
+    default: false,
+    component: 'switch',
+    optionType: 'model',
+    showDefault: false,
+    columnSpan: 2,
+  },
   reasoning_summary: {
     key: 'reasoning_summary',
     label: 'com_endpoint_reasoning_summary',
@@ -596,6 +609,7 @@ const openAI: SettingsConfiguration = [
   baseDefinitions.stop,
   librechat.resendFiles,
   baseDefinitions.imageDetail,
+  openAIParams.web_search,
   openAIParams.reasoning_effort,
   openAIParams.useResponsesApi,
   openAIParams.reasoning_summary,
@@ -618,8 +632,9 @@ const openAICol2: SettingsConfiguration = [
   librechat.resendFiles,
   baseDefinitions.imageDetail,
   openAIParams.reasoning_effort,
-  openAIParams.useResponsesApi,
   openAIParams.reasoning_summary,
+  openAIParams.useResponsesApi,
+  openAIParams.web_search,
 ];
 
 const anthropicConfig: SettingsConfiguration = [

packages/data-provider/src/schemas.ts

@@ -634,6 +634,8 @@ export const tConversationSchema = z.object({
   reasoning_summary: eReasoningSummarySchema.optional().nullable(),
   /* OpenAI: use Responses API */
   useResponsesApi: z.boolean().optional(),
+  /* OpenAI: use Responses API with Web Search */
+  web_search: z.boolean().optional(),
   /* Google: use Search Grounding */
   grounding: z.boolean().optional(),
   /* assistant */
@@ -1071,6 +1073,7 @@ export const openAIBaseSchema = tConversationSchema.pick({
   reasoning_effort: true,
   reasoning_summary: true,
   useResponsesApi: true,
+  web_search: true,
 });
 
 export const openAISchema = openAIBaseSchema

packages/data-schemas/src/types/convo.ts

@@ -47,6 +47,7 @@ export interface IConversation extends Document {
   reasoning_effort?: string;
   reasoning_summary?: string;
   useResponsesApi?: boolean;
+  web_search?: boolean;
   grounding?: boolean;
   // Additional fields
   files?: string[];