RHIDP-13056: separating notebooks and lightspeed vector_store

workspaces/lightspeed/.changeset/chilled-guests-search.md

@@ -0,0 +1,6 @@
+---
+'@red-hat-developer-hub/backstage-plugin-lightspeed-backend': minor
+---
+
+All lightspeed query is now called with rhdh-docs vector_store. Notebooks app-config only now requires queryDefaults model and provider
+All files uploaded to lightspeed-stack will be converted to .txt

workspaces/lightspeed/app-config.yaml

@@ -21,12 +21,8 @@ lightspeed:
   notebooks:
     enabled: false
     queryDefaults:
-      model: redhataillama-31-8b-instruct
-      provider_id: vllm
-    sessionDefaults:
-      provider_id: notebooks
-      embedding_model: ${LLAMA_STACK_EMBEDDING_MODEL}
-      embedding_dimension: 768
+      model: ${NOTEBOOKS_QUERY_MODEL}
+      provider_id: ${NOTEBOOKS_QUERY_PROVIDER_ID}
 
 backend:
   # Used for enabling authentication, secret is shared by all backend plugins

workspaces/lightspeed/plugins/lightspeed-backend/README.md

@@ -110,15 +110,8 @@ lightspeed:
     # Required: Query defaults for RAG queries
     # Both model and provider_id must be configured together
     queryDefaults:
-      model: llama3.1-8b-instruct # Model to use for answering queries
-      provider_id: ollama # AI provider for the query model
-
-    # Required: Session defaults for creating vector stores
-    # All three fields are required when Notebooks is enabled
-    sessionDefaults:
-      provider_id: notebooks # Vector store provider ID (must match Llama Stack config)
-      embedding_model: sentence-transformers/all-mpnet-base-v2 # Model for generating embeddings
-      embedding_dimension: 768 # Embedding vector dimension (must match model output)
+      model: ${NOTEBOOKS_QUERY_MODEL} # Model to use for answering queries. Must map to a model available through the provider set in $NOTEBOOKS_QUERY_PROVIDER_ID
+      provider_id: ${NOTEBOOKS_QUERY_PROVIDER_ID} # AI provider for the query model. Must map to a provider enabled in your Lightspeed config.yaml
 
     # Optional: Chunking strategy for document processing
     chunkingStrategy:
@@ -143,11 +136,7 @@ lightspeed:
 - **`queryDefaults.model`** _(required)_: The LLM model to use for answering RAG queries. Must be available in the configured provider.
 - **`queryDefaults.provider_id`** _(required)_: The AI provider identifier for the query model (e.g., `ollama`, `vllm`). Both `model` and `provider_id` must be configured together.
 
-**Session Defaults** _(required when enabled)_:
-
-- **`sessionDefaults.provider_id`** _(required)_: Vector store provider identifier. Must match a provider configured in your Llama Stack instance (e.g., `notebooks`, `chromadb`). This determines where document embeddings are stored.
-- **`sessionDefaults.embedding_model`** _(required)_: The embedding model to use for converting documents to vectors (e.g., `sentence-transformers/all-mpnet-base-v2`). Must be available in Llama Stack.
-- **`sessionDefaults.embedding_dimension`** _(required)_: Dimension of the embedding vectors produced by the embedding model. Must match the model's output dimension (commonly `768`, `384`, or `1536`).
+> **Important**: The `model` and `provider_id` values must map to a provider and model that are actually enabled in your Lightspeed config.yaml configuration. If the provider or model is not available in Lightspeed, queries will fail. For example, if `openai` enabled in Lightspeed via ENABLE_OPENAI, then model must be available, e.g (model=gpt-4o-mini).
 
 **Chunking Strategy** _(optional)_:
 

workspaces/lightspeed/plugins/lightspeed-backend/__fixtures__/lcsHandlers.ts

@@ -268,6 +268,20 @@ export const lcsHandlers: HttpHandler[] = [
     });
   }),
 
+  // Vector stores list endpoint - returns mock RHDH product docs vector store
+  http.get(`${LOCAL_LCS_ADDR}/v1/vector-stores`, () => {
+    return HttpResponse.json({
+      data: [
+        {
+          id: 'vs-rhdh-product-docs',
+          name: 'rhdh-product-docs',
+          provider_id: 'notebooks',
+          metadata: {},
+        },
+      ],
+    });
+  }),
+
   // Catch-all handler for unknown paths
   http.all(`${LOCAL_LCS_ADDR}/*`, ({ request }) => {
     console.log(`Caught request to unknown path: ${request.url}`);

workspaces/lightspeed/plugins/lightspeed-backend/app-config.yaml

@@ -13,13 +13,6 @@
 #      provider_id: ollama # AI provider for query model (e.g., ollama, vllm)
 #      model: llama3.1-8b-instruct # Model to use for answering queries
 #
-#    # REQUIRED when enabled: Session defaults for vector stores
-#    # All three fields are required
-#    sessionDefaults:
-#      provider_id: notebooks # Vector store provider ID (must match Llama Stack config)
-#      embedding_model: sentence-transformers/all-mpnet-base-v2 # Embedding model for documents
-#      embedding_dimension: 768 # Vector dimension (must match embedding model output)
-#
 #    # OPTIONAL: Chunking strategy for document processing
 #    chunkingStrategy:
 #      type: auto # 'auto' (default) or 'static'

workspaces/lightspeed/plugins/lightspeed-backend/src/service/constant.ts

@@ -22,6 +22,7 @@ export const DEFAULT_CHUNKING_STRATEGY_TYPE = 'auto'; // auto chunking
 export const DEFAULT_MAX_CHUNK_SIZE_TOKENS = 512; // 512 tokens
 export const DEFAULT_CHUNK_OVERLAP_TOKENS = 50; // 50 tokens
 export const DEFAULT_LLAMA_STACK_PORT = 8321; // Llama Stack port
+export const DEFAULT_LIGHTSPEED_SERVICE_HOST = '0.0.0.0'; // Lightspeed core service host
 export const DEFAULT_LIGHTSPEED_SERVICE_PORT = 8080; // Lightspeed service port
 export const DEFAULT_MAX_FILE_SIZE_MB = 20 * 1024 * 1024; // 20MB
 export const NOTEBOOKS_SYSTEM_PROMPT =
@@ -36,16 +37,16 @@ Constraints:
 Output Format:
 1. Summary: A 1-2 sentence high-level answer.
 2. Detailed Analysis: A structured breakdown using bullet points.
-3. References: A list of sources used.
+3. References: A list of sources used. References should be in the format of [Document Title] in a new line for each reference.
 
 Disclaimer: Your answers **MUST** be grounded in the provided documents. If the answer isn't present, state: "I don't know based on the provided documents."
+Remember, **ALL** references must be from the provided documents and provided documents only.
 Make no mistakes.
 `.trim();
 
 /**
  * HTTP and networking constants
  */
-export const LIGHTSPEED_SERVICE_HOST = '0.0.0.0'; // Lightspeed core service host
 export const URL_FETCH_TIMEOUT_MS = 30000; // 30 second timeout for URL fetching
 export const USER_AGENT = 'RHDH-AI-Notebooks-Bot/1.0'; // User agent for HTTP requests
 export const MAX_URL_CONTENT_SIZE = 10 * 1024 * 1024; // 10MB max for URL fetched content

workspaces/lightspeed/plugins/lightspeed-backend/src/service/notebooks/VectorStoresOperator.ts

@@ -82,16 +82,45 @@ async function handleHttpError(
  *
  * This class provides the same interface as LlamaStackClient but proxies calls through
  * lightspeed-core REST API instead of calling llama stack directly.
+ *
+ * Implemented as a singleton to ensure single instance across the application.
  */
 export class VectorStoresOperator {
+  private static instance: VectorStoresOperator | null = null;
   private baseURL: string;
   private logger: LoggerService;
 
-  constructor(lightspeedCoreUrl: string, logger: LoggerService) {
+  private constructor(lightspeedCoreUrl: string, logger: LoggerService) {
     this.baseURL = lightspeedCoreUrl;
     this.logger = logger;
   }
 
+  /**
+   * Get the singleton instance of VectorStoresOperator
+   * @param lightspeedCoreUrl - Lightspeed core URL (required on first call)
+   * @param logger - Logger service (required on first call)
+   * @returns The singleton instance
+   */
+  static getInstance(
+    lightspeedCoreUrl: string,
+    logger: LoggerService,
+  ): VectorStoresOperator {
+    if (!VectorStoresOperator.instance) {
+      VectorStoresOperator.instance = new VectorStoresOperator(
+        lightspeedCoreUrl,
+        logger,
+      );
+    }
+    return VectorStoresOperator.instance;
+  }
+
+  /**
+   * Reset the singleton instance (primarily for testing)
+   */
+  static resetInstance(): void {
+    VectorStoresOperator.instance = null;
+  }
+
   /**
    * Vector Stores API - mirrors LlamaStackClient.vectorStores structure
    */

workspaces/lightspeed/plugins/lightspeed-backend/src/service/notebooks/documents/documentService.test.ts

@@ -62,9 +62,10 @@ describe('DocumentService', () => {
         },
       },
     });
-    operator = new VectorStoresOperator(LIGHTSPEED_CORE_ADDR, logger);
+    VectorStoresOperator.resetInstance(); // Reset singleton before each test
+    operator = VectorStoresOperator.getInstance(LIGHTSPEED_CORE_ADDR, logger);
     documentService = new DocumentService(operator, logger, config);
-    sessionService = new SessionService(operator, logger, config);
+    sessionService = new SessionService(operator, logger);
 
     // Create a test session for document operations
     const session = await sessionService.createSession(
@@ -84,7 +85,6 @@ describe('DocumentService', () => {
       const fileId = await documentService.uploadFile(
         'Test content',
         'test-file.txt',
-        'txt',
       );
 
       expect(fileId).toBeDefined();
@@ -93,23 +93,13 @@ describe('DocumentService', () => {
 
     it('should handle upload errors', async () => {
       // Mock a failure by passing invalid content
-      await expect(
-        documentService.uploadFile('', '', 'txt'),
-      ).resolves.toBeDefined();
+      await expect(documentService.uploadFile('', '')).resolves.toBeDefined();
     });
 
     it('should use correct MIME type based on file type', async () => {
-      const fileId1 = await documentService.uploadFile(
-        '{}',
-        'test.json',
-        'json',
-      );
-      const fileId2 = await documentService.uploadFile(
-        'text',
-        'test.txt',
-        'txt',
-      );
-      const fileId3 = await documentService.uploadFile('# MD', 'test.md', 'md');
+      const fileId1 = await documentService.uploadFile('{}', 'test.json');
+      const fileId2 = await documentService.uploadFile('text', 'test.txt');
+      const fileId3 = await documentService.uploadFile('# MD', 'test.md');
 
       expect(fileId1).toBeDefined();
       expect(fileId2).toBeDefined();
@@ -119,11 +109,7 @@ describe('DocumentService', () => {
 
   describe('getFileStatus', () => {
     it('should get file status for existing document', async () => {
-      const fileId = await documentService.uploadFile(
-        'Content',
-        'Test Doc',
-        'text',
-      );
+      const fileId = await documentService.uploadFile('Content', 'Test Doc');
       await documentService.upsertDocument(
         sessionId,
         'Test Doc',
@@ -149,7 +135,6 @@ describe('DocumentService', () => {
       const fileId = await documentService.uploadFile(
         'This is test content',
         'Test Document',
-        'text',
       );
 
       const result = await documentService.upsertDocument(
@@ -169,7 +154,6 @@ describe('DocumentService', () => {
       const fileId1 = await documentService.uploadFile(
         'Original content',
         'Original Title',
-        'text',
       );
       await documentService.upsertDocument(
         sessionId,
@@ -181,7 +165,6 @@ describe('DocumentService', () => {
       const fileId2 = await documentService.uploadFile(
         'Updated content',
         'Original Title',
-        'text',
       );
       const result = await documentService.upsertDocument(
         sessionId,
@@ -199,7 +182,6 @@ describe('DocumentService', () => {
       const fileId1 = await documentService.uploadFile(
         'Content',
         'Original Title',
-        'text',
       );
       await documentService.upsertDocument(
         sessionId,
@@ -211,7 +193,6 @@ describe('DocumentService', () => {
       const fileId2 = await documentService.uploadFile(
         'Updated content',
         'New Title',
-        'text',
       );
       const result = await documentService.upsertDocument(
         sessionId,
@@ -230,7 +211,6 @@ describe('DocumentService', () => {
       const fileId1 = await documentService.uploadFile(
         'Content 1',
         'Document 1',
-        'text',
       );
       await documentService.upsertDocument(
         sessionId,
@@ -242,7 +222,6 @@ describe('DocumentService', () => {
       const fileId2 = await documentService.uploadFile(
         'Content 2',
         'Document 2',
-        'text',
       );
       await documentService.upsertDocument(
         sessionId,
@@ -265,23 +244,15 @@ describe('DocumentService', () => {
     });
 
     it('should filter documents by file type', async () => {
-      const fileId1 = await documentService.uploadFile(
-        'Content',
-        'Text Doc',
-        'text',
-      );
+      const fileId1 = await documentService.uploadFile('Content', 'Text Doc');
       await documentService.upsertDocument(
         sessionId,
         'Text Doc',
         'text',
         fileId1,
       );
 
-      const fileId2 = await documentService.uploadFile(
-        'Content',
-        'PDF Doc',
-        'pdf',
-      );
+      const fileId2 = await documentService.uploadFile('Content', 'PDF Doc');
       await documentService.upsertDocument(
         sessionId,
         'PDF Doc',
@@ -299,7 +270,6 @@ describe('DocumentService', () => {
       const fileId = await documentService.uploadFile(
         'Content',
         'Test Document',
-        'text',
       );
       await documentService.upsertDocument(
         sessionId,
@@ -323,7 +293,6 @@ describe('DocumentService', () => {
       const fileId = await documentService.uploadFile(
         'Content',
         'Test Document',
-        'text',
       );
       await documentService.upsertDocument(
         sessionId,

workspaces/lightspeed/plugins/lightspeed-backend/src/service/notebooks/documents/documentService.ts

@@ -22,7 +22,6 @@ import {
   DEFAULT_CHUNK_OVERLAP_TOKENS,
   DEFAULT_CHUNKING_STRATEGY_TYPE,
   DEFAULT_MAX_CHUNK_SIZE_TOKENS,
-  FILE_TYPE_TO_MIME,
 } from '../../constant';
 import { SessionDocument, UpsertResult } from '../types/notebooksTypes';
 import { VectorStoresOperator } from '../VectorStoresOperator';
@@ -98,27 +97,13 @@ export class DocumentService {
    * @returns File ID from the Files API
    * @throws Error if upload fails
    */
-  /**
-   * Upload a file to the Files API
-   * @param content - File content as string
-   * @param title - File title/name
-   * @param fileType - Optional file type for MIME type detection
-   * @returns File ID from the Files API
-   * @throws Error if upload fails
-   */
-  async uploadFile(
-    content: string,
-    title: string,
-    fileType?: string,
-  ): Promise<string> {
+  async uploadFile(content: string, title: string): Promise<string> {
     try {
       // Determine MIME type from file type or default to text/plain
-      const mimeType = fileType
-        ? FILE_TYPE_TO_MIME[fileType] || 'text/plain'
-        : 'text/plain';
-
+      const mimeType = 'text/plain';
+      const txtFilename = `${title.replace(/\.[^.]+$/, '')}.txt`;
       const file = await this.client.files.create({
-        file: await toFile(Buffer.from(content, 'utf-8'), title, {
+        file: await toFile(Buffer.from(content, 'utf-8'), txtFilename, {
           type: mimeType,
         }),
         purpose: 'assistants',

workspaces/lightspeed/plugins/lightspeed-backend/src/service/notebooks/notebooksRouter.test.ts

@@ -26,6 +26,7 @@ import {
   resetMockStorage,
 } from '../../../__fixtures__/lightspeedCoreHandlers';
 import { createNotebooksRouter } from './notebooksRouters';
+import { VectorStoresOperator } from './VectorStoresOperator';
 
 const mockUserId = 'user:default/guest';
 
@@ -53,6 +54,7 @@ describe('Notebooks Router', () => {
 
   beforeEach(async () => {
     resetMockStorage();
+    VectorStoresOperator.resetInstance(); // Reset singleton before each test
     const logger = mockServices.logger.mock();
     const config = mockServices.rootConfig({
       data: {