Merge 378242763b into 8ed0bcf5ca

client/src/locales/en/translation.json

@@ -352,6 +352,9 @@
   "com_endpoint_top_p": "Top P",
   "com_endpoint_use_active_assistant": "Use Active Assistant",
   "com_endpoint_use_responses_api": "Use Responses API",
+  "com_endpoint_stored_prompt_id": "Stored Prompt ID",
+  "com_endpoint_stored_prompt_id_placeholder": "pmpt_...",
+  "com_endpoint_openai_stored_prompt_id": "OpenAI Playground stored prompt ID (pmpt_...). Routes the conversation through a pre-configured prompt that carries its own system instructions and file-search vector store. Automatically enables the Responses API.",
   "com_endpoint_use_search_grounding": "Grounding with Google Search",
   "com_endpoint_verbosity": "Verbosity",
   "com_error_endpoint_models_not_loaded": "Models for {{0}} could not be loaded. Please refresh the page and try again.",

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

@@ -737,6 +737,68 @@ describe('extractDefaultParams', () => {
   });
 });
 
+describe('Stored Playground Prompt (stored_prompt_id)', () => {
+  it('should inject prompt.id into modelKwargs and force useResponsesApi when stored_prompt_id is set', () => {
+    const result = getOpenAILLMConfig({
+      apiKey: 'test-api-key',
+      streaming: true,
+      modelOptions: {
+        model: 'gpt-4o',
+        stored_prompt_id: 'pmpt_abc123',
+      } as Partial<t.OpenAIParameters>,
+    });
+
+    expect(result.llmConfig).toHaveProperty('useResponsesApi', true);
+    expect(result.llmConfig.modelKwargs).toEqual(
+      expect.objectContaining({ prompt: { id: 'pmpt_abc123' } }),
+    );
+  });
+
+  it('should not set modelKwargs.prompt when stored_prompt_id is absent', () => {
+    const result = getOpenAILLMConfig({
+      apiKey: 'test-api-key',
+      streaming: true,
+      modelOptions: {
+        model: 'gpt-4o',
+        useResponsesApi: true,
+      } as Partial<t.OpenAIParameters>,
+    });
+
+    expect(result.llmConfig.modelKwargs?.prompt).toBeUndefined();
+  });
+
+  it('should preserve a manually-set useResponsesApi=true alongside stored_prompt_id', () => {
+    const result = getOpenAILLMConfig({
+      apiKey: 'test-api-key',
+      streaming: true,
+      modelOptions: {
+        model: 'gpt-4o',
+        useResponsesApi: true,
+        stored_prompt_id: 'pmpt_xyz789',
+      } as Partial<t.OpenAIParameters>,
+    });
+
+    expect(result.llmConfig).toHaveProperty('useResponsesApi', true);
+    expect(result.llmConfig.modelKwargs).toEqual(
+      expect.objectContaining({ prompt: { id: 'pmpt_xyz789' } }),
+    );
+  });
+
+  it('should not add stored_prompt_id itself to llmConfig or modelKwargs as a raw field', () => {
+    const result = getOpenAILLMConfig({
+      apiKey: 'test-api-key',
+      streaming: true,
+      modelOptions: {
+        model: 'gpt-4o',
+        stored_prompt_id: 'pmpt_abc123',
+      } as Partial<t.OpenAIParameters>,
+    });
+
+    expect(result.llmConfig).not.toHaveProperty('stored_prompt_id');
+    expect(result.llmConfig.modelKwargs?.stored_prompt_id).toBeUndefined();
+  });
+});
+
 describe('applyDefaultParams', () => {
   it('should apply defaults only when field is undefined', () => {
     const target: Record<string, unknown> = {

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

@@ -53,6 +53,7 @@ export const knownOpenAIParams = new Set([
   'truncation',
   'include',
   'previous_response_id',
+  'stored_prompt_id',
   // LangChain specific
   '__includeRawResponse',
   'maxConcurrency',
@@ -150,6 +151,7 @@ export function getOpenAILLMConfig({
     reasoning_summary,
     verbosity,
     web_search,
+    stored_prompt_id,
     frequency_penalty,
     presence_penalty,
     ...modelOptions
@@ -274,6 +276,20 @@ export function getOpenAILLMConfig({
     tools.push({ type: 'web_search' });
   }
 
+  if (stored_prompt_id) {
+    /**
+     * Forward a stored OpenAI Playground prompt by ID.
+     * The Responses API accepts `prompt: { id, variables? }` which causes OpenAI
+     * to hydrate the stored system instructions, file-search vector store, and
+     * any tool definitions configured in the Playground.
+     * We force `useResponsesApi` here because Chat Completions does not support
+     * this parameter.
+     */
+    llmConfig.useResponsesApi = true;
+    modelKwargs.prompt = { id: stored_prompt_id };
+    hasModelKwargs = true;
+  }
+
   /**
    * Note: OpenAI reasoning models (o1/o3/gpt-5) do not support temperature and other sampling parameters
    * Exception: gpt-5-chat and versioned models like gpt-5.1 DO support these parameters

packages/data-provider/src/parameterSettings.ts

@@ -268,6 +268,20 @@ const openAIParams: Record<string, SettingDefinition> = {
     showDefault: false,
     columnSpan: 2,
   },
+  stored_prompt_id: {
+    key: 'stored_prompt_id',
+    label: 'com_endpoint_stored_prompt_id',
+    labelCode: true,
+    type: 'string',
+    default: '',
+    component: 'input',
+    placeholder: 'com_endpoint_stored_prompt_id_placeholder',
+    placeholderCode: true,
+    description: 'com_endpoint_openai_stored_prompt_id',
+    descriptionCode: true,
+    optionType: 'model',
+    columnSpan: 4,
+  },
   web_search: {
     key: 'web_search',
     label: 'com_ui_web_search',
@@ -765,6 +779,7 @@ const openAI: SettingsConfiguration = [
   openAIParams.web_search,
   openAIParams.reasoning_effort,
   openAIParams.useResponsesApi,
+  openAIParams.stored_prompt_id,
   openAIParams.reasoning_summary,
   openAIParams.verbosity,
   openAIParams.disableStreaming,
@@ -791,6 +806,7 @@ const openAICol2: SettingsConfiguration = [
   openAIParams.reasoning_summary,
   openAIParams.verbosity,
   openAIParams.useResponsesApi,
+  openAIParams.stored_prompt_id,
   openAIParams.web_search,
   openAIParams.disableStreaming,
   librechat.fileTokenLimit,

packages/data-provider/src/schemas.ts

@@ -770,6 +770,14 @@ export const tConversationSchema = z.object({
   verbosity: eVerbositySchema.optional().nullable(),
   /* OpenAI: use Responses API */
   useResponsesApi: z.boolean().optional(),
+  /**
+   * OpenAI Playground stored prompt ID (e.g. "pmpt_...").
+   * When set, LibreChat forwards `prompt: { id }` in the Responses API request,
+   * routing the call through a pre-configured Playground prompt that can carry
+   * its own system instructions, file-search vector store, and tool definitions.
+   * Implicitly enables `useResponsesApi`.
+   */
+  stored_prompt_id: z.string().optional(),
   /* Anthropic: Effort control */
   effort: eAnthropicEffortSchema.optional().nullable(),
   /* OpenAI Responses API / Anthropic API / Google API */
@@ -880,6 +888,8 @@ export const tQueryParamsSchema = tConversationSchema
     verbosity: true,
     /** @endpoints openAI, custom, azureOpenAI */
     useResponsesApi: true,
+    /** @endpoints openAI, custom, azureOpenAI */
+    stored_prompt_id: true,
     /** @endpoints openAI, anthropic, google */
     web_search: true,
     /** @endpoints openAI, custom, azureOpenAI */
@@ -1178,6 +1188,7 @@ export const openAIBaseSchema = tConversationSchema.pick({
   reasoning_summary: true,
   verbosity: true,
   useResponsesApi: true,
+  stored_prompt_id: true,
   web_search: true,
   disableStreaming: true,
   fileTokenLimit: true,

packages/data-schemas/src/schema/defaults.ts

@@ -139,6 +139,10 @@ export const conversationPreset = {
   useResponsesApi: {
     type: Boolean,
   },
+  /** OpenAI Playground stored prompt ID (pmpt_...) */
+  stored_prompt_id: {
+    type: String,
+  },
   /** OpenAI Responses API / Anthropic API / Google API */
   web_search: {
     type: Boolean,

packages/data-schemas/src/schema/preset.ts

@@ -50,6 +50,7 @@ export interface IPreset extends Document {
   reasoning_summary?: string;
   verbosity?: string;
   useResponsesApi?: boolean;
+  stored_prompt_id?: string;
   web_search?: boolean;
   disableStreaming?: boolean;
   fileTokenLimit?: number;

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

@@ -48,6 +48,7 @@ export interface IConversation extends Document {
   reasoning_summary?: string;
   verbosity?: string;
   useResponsesApi?: boolean;
+  stored_prompt_id?: string;
   web_search?: boolean;
   disableStreaming?: boolean;
   fileTokenLimit?: number;