@retrivora-ai/rag-engine 1.8.0 → 1.8.2

package/dist/server.d.mts

@@ -1,8 +1,7 @@
-import { j as VectorDBConfig, L as LLMConfig, E as EmbeddingConfig, h as RagConfig, I as IngestDocument, C as ChatMessage, d as ChatResponse, l as RetrievalResult, i as UpsertDocument, V as VectorMatch, G as GraphDBConfig, m as GraphNode, n as Edge, o as GraphSearchResult, k as VectorDBProvider, f as LLMProvider, e as EmbeddingProvider, g as RAGConfig, U as UIConfig, c as ChatOptions } from './index-DPsQodME.mjs';
-import { I as ILLMProvider, E as EmbedOptions } from './DocumentChunker-CFEiRopR.mjs';
-export { C as Chunk, a as ChunkOptions, D as DocumentChunker } from './DocumentChunker-CFEiRopR.mjs';
-import { H as HealthCheckResult, I as IProviderValidator, a as IProviderHealthChecker } from './index-Bb2yEopi.mjs';
-export { C as ConfigValidator, V as ValidationError, b as VectorPlugin, c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from './index-Bb2yEopi.mjs';
+import { n as VectorDBConfig, L as LLMConfig, e as EmbeddingConfig, k as RagConfig, I as ILLMProvider, g as IngestDocument, C as ChatMessage, d as ChatResponse, p as RetrievalResult, m as UpsertDocument, V as VectorMatch, G as GraphDBConfig, q as GraphNode, r as Edge, s as GraphSearchResult, o as VectorDBProvider, h as LLMProvider, f as EmbeddingProvider, j as RAGConfig, U as UIConfig, c as ChatOptions, E as EmbedOptions } from './ILLMProvider-BOJFz3Na.mjs';
+export { C as Chunk, a as ChunkOptions, D as DocumentChunker } from './DocumentChunker-Dh9TvmGG.mjs';
+import { H as HealthCheckResult, I as IProviderValidator, a as IProviderHealthChecker } from './index-D3V9Et2M.mjs';
+export { C as ConfigValidator, V as ValidationError, b as VectorPlugin, c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from './index-D3V9Et2M.mjs';
 import 'next/server';
 
 /**
@@ -106,6 +105,7 @@ declare class ProviderHealthCheck {
  * - Error recovery for transient failures
  * - Multi-tenancy support via namespacing
  * - LRU-bounded embedding cache (max 500 entries, prevents memory leaks)
+ * - Full observability tracing (latency, tokens, hallucination scoring)
  */
 declare class Pipeline {
     private config;
@@ -122,54 +122,49 @@ declare class Pipeline {
     private embeddingCache;
     private initialised;
     constructor(config: RagConfig);
+    /**
+     * Expose the underlying LLM provider (set after initialize()).
+     * Used by the stream handler to pass to UITransformer.analyzeAndDecide().
+     */
+    getLLMProvider(): ILLMProvider | undefined;
     initialize(): Promise<void>;
     /**
      * Ingest documents with automatic chunking, embedding, and batch upsert.
-     * Handles retries for transient failures.
      */
     ingest(documents: IngestDocument[], namespace?: string): Promise<Array<{
         docId: string | number;
         chunksIngested: number;
     }>>;
-    /**
-     * Step 1: Chunk the document content.
-     */
+    /** Step 1: Chunk the document content. */
     private prepareChunks;
     /**
      * Step 2: Generate embeddings for chunks with retry logic.
      * Uses batchEmbed when available for efficiency; falls back to sequential embedding.
      */
     private processEmbeddings;
-    /**
-     * Step 3: Upsert chunks to vector database with retry logic.
-     */
+    /** Step 3: Upsert chunks to vector database with retry logic. */
     private processUpserts;
-    /**
-     * Step 4: Optional graph-based entity extraction and ingestion.
-     */
+    /** Step 4: Optional graph-based entity extraction and ingestion. */
     private processGraphIngestion;
     ask(question: string, history?: ChatMessage[], namespace?: string): Promise<ChatResponse>;
     /**
      * High-performance streaming RAG flow.
-     * Yields text chunks first, then the retrieval metadata at the end.
+     * Yields text chunks first, then the retrieval metadata + observability trace at the end.
      */
     askStream(question: string, history?: ChatMessage[], namespace?: string): AsyncIterable<string | ChatResponse>;
     /**
      * Universal retrieval method combining all enabled providers.
      * Uses an LRU-bounded embedding cache to avoid re-embedding the same query.
      */
+    private generateUiTransformation;
     retrieve(query: string, options: {
         namespace?: string;
         topK?: number;
         filter?: Record<string, unknown>;
     }): Promise<RetrievalResult>;
-    /**
-     * Rewrite the user query for better retrieval performance.
-     */
+    /** Rewrite the user query for better retrieval performance. */
     private rewriteQuery;
-    /**
-     * Generate 3-5 short, relevant questions based on the vector database content.
-     */
+    /** Generate 3-5 short, relevant questions based on the vector database content. */
     getSuggestions(query: string, namespace?: string): Promise<string[]>;
 }
 
@@ -659,17 +654,17 @@ declare class MultiTablePostgresProvider extends BaseVectorProvider {
     private readonly connectionString;
     private tables;
     private searchFields;
+    private readonly uploadTable;
     constructor(config: VectorDBConfig);
     initialize(): Promise<void>;
     /**
-     * Upsert is not supported for MultiTablePostgresProvider as it's designed for
-     * searching across pre-existing tables with varying schemas.
+     * Upsert a document by dynamically provisioning a table and columns.
      */
-    upsert(_doc: UpsertDocument, _namespace?: string): Promise<void>;
+    upsert(doc: UpsertDocument, namespace?: string): Promise<void>;
     /**
-     * Batch upsert is not supported for MultiTablePostgresProvider.
+     * Batch upsert documents by dynamically provisioning tables and columns based on CSV headers.
      */
-    batchUpsert(_docs: UpsertDocument[], _namespace?: string): Promise<void>;
+    batchUpsert(docs: UpsertDocument[], namespace?: string): Promise<void>;
     /**
      * Query all configured tables and merge results, sorted by cosine similarity score.
      */
@@ -871,6 +866,31 @@ declare class UniversalVectorProvider extends BaseVectorProvider {
  * LLMFactory — instantiates the correct ILLMProvider based on llmConfig.provider.
  */
 declare class LLMFactory {
+    /**
+     * Register a custom LLM provider factory at runtime.
+     *
+     * Use this to add support for any LLM backend (Azure OpenAI, Cohere, Mistral,
+     * Bedrock, etc.) without modifying this library's source code.
+     *
+     * @example
+     * // In your Next.js app initialization:
+     * import { LLMFactory } from '@retrivora-ai/rag-engine/server';
+     * import { MyCustomProvider } from './providers/MyCustomProvider';
+     *
+     * LLMFactory.register('my-provider', (config) => new MyCustomProvider(config));
+     *
+     * // Then set in your .env.local:
+     * // LLM_PROVIDER=my-provider
+     */
+    static register(name: string, factory: (config: LLMConfig) => ILLMProvider): void;
+    /**
+     * Unregister a previously registered custom provider.
+     */
+    static unregister(name: string): void;
+    /**
+     * List all registered provider names (built-in + custom).
+     */
+    static listProviders(): string[];
     static create(llmConfig: LLMConfig, embeddingConfig?: EmbeddingConfig): ILLMProvider;
     static getValidator(provider: LLMProvider): IProviderValidator | null;
     static getHealthChecker(provider: LLMProvider): IProviderHealthChecker | null;

package/dist/server.d.ts

@@ -1,8 +1,7 @@
-import { j as VectorDBConfig, L as LLMConfig, E as EmbeddingConfig, h as RagConfig, I as IngestDocument, C as ChatMessage, d as ChatResponse, l as RetrievalResult, i as UpsertDocument, V as VectorMatch, G as GraphDBConfig, m as GraphNode, n as Edge, o as GraphSearchResult, k as VectorDBProvider, f as LLMProvider, e as EmbeddingProvider, g as RAGConfig, U as UIConfig, c as ChatOptions } from './index-DPsQodME.js';
-import { I as ILLMProvider, E as EmbedOptions } from './DocumentChunker-C1GEEosY.js';
-export { C as Chunk, a as ChunkOptions, D as DocumentChunker } from './DocumentChunker-C1GEEosY.js';
-import { H as HealthCheckResult, I as IProviderValidator, a as IProviderHealthChecker } from './index-CkbTzj9J.js';
-export { C as ConfigValidator, V as ValidationError, b as VectorPlugin, c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from './index-CkbTzj9J.js';
+import { n as VectorDBConfig, L as LLMConfig, e as EmbeddingConfig, k as RagConfig, I as ILLMProvider, g as IngestDocument, C as ChatMessage, d as ChatResponse, p as RetrievalResult, m as UpsertDocument, V as VectorMatch, G as GraphDBConfig, q as GraphNode, r as Edge, s as GraphSearchResult, o as VectorDBProvider, h as LLMProvider, f as EmbeddingProvider, j as RAGConfig, U as UIConfig, c as ChatOptions, E as EmbedOptions } from './ILLMProvider-BOJFz3Na.js';
+export { C as Chunk, a as ChunkOptions, D as DocumentChunker } from './DocumentChunker-Dh9TvmGG.js';
+import { H as HealthCheckResult, I as IProviderValidator, a as IProviderHealthChecker } from './index-BwpcaziY.js';
+export { C as ConfigValidator, V as ValidationError, b as VectorPlugin, c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from './index-BwpcaziY.js';
 import 'next/server';
 
 /**
@@ -106,6 +105,7 @@ declare class ProviderHealthCheck {
  * - Error recovery for transient failures
  * - Multi-tenancy support via namespacing
  * - LRU-bounded embedding cache (max 500 entries, prevents memory leaks)
+ * - Full observability tracing (latency, tokens, hallucination scoring)
  */
 declare class Pipeline {
     private config;
@@ -122,54 +122,49 @@ declare class Pipeline {
     private embeddingCache;
     private initialised;
     constructor(config: RagConfig);
+    /**
+     * Expose the underlying LLM provider (set after initialize()).
+     * Used by the stream handler to pass to UITransformer.analyzeAndDecide().
+     */
+    getLLMProvider(): ILLMProvider | undefined;
     initialize(): Promise<void>;
     /**
      * Ingest documents with automatic chunking, embedding, and batch upsert.
-     * Handles retries for transient failures.
      */
     ingest(documents: IngestDocument[], namespace?: string): Promise<Array<{
         docId: string | number;
         chunksIngested: number;
     }>>;
-    /**
-     * Step 1: Chunk the document content.
-     */
+    /** Step 1: Chunk the document content. */
     private prepareChunks;
     /**
      * Step 2: Generate embeddings for chunks with retry logic.
      * Uses batchEmbed when available for efficiency; falls back to sequential embedding.
      */
     private processEmbeddings;
-    /**
-     * Step 3: Upsert chunks to vector database with retry logic.
-     */
+    /** Step 3: Upsert chunks to vector database with retry logic. */
     private processUpserts;
-    /**
-     * Step 4: Optional graph-based entity extraction and ingestion.
-     */
+    /** Step 4: Optional graph-based entity extraction and ingestion. */
     private processGraphIngestion;
     ask(question: string, history?: ChatMessage[], namespace?: string): Promise<ChatResponse>;
     /**
      * High-performance streaming RAG flow.
-     * Yields text chunks first, then the retrieval metadata at the end.
+     * Yields text chunks first, then the retrieval metadata + observability trace at the end.
      */
     askStream(question: string, history?: ChatMessage[], namespace?: string): AsyncIterable<string | ChatResponse>;
     /**
      * Universal retrieval method combining all enabled providers.
      * Uses an LRU-bounded embedding cache to avoid re-embedding the same query.
      */
+    private generateUiTransformation;
     retrieve(query: string, options: {
         namespace?: string;
         topK?: number;
         filter?: Record<string, unknown>;
     }): Promise<RetrievalResult>;
-    /**
-     * Rewrite the user query for better retrieval performance.
-     */
+    /** Rewrite the user query for better retrieval performance. */
     private rewriteQuery;
-    /**
-     * Generate 3-5 short, relevant questions based on the vector database content.
-     */
+    /** Generate 3-5 short, relevant questions based on the vector database content. */
     getSuggestions(query: string, namespace?: string): Promise<string[]>;
 }
 
@@ -659,17 +654,17 @@ declare class MultiTablePostgresProvider extends BaseVectorProvider {
     private readonly connectionString;
     private tables;
     private searchFields;
+    private readonly uploadTable;
     constructor(config: VectorDBConfig);
     initialize(): Promise<void>;
     /**
-     * Upsert is not supported for MultiTablePostgresProvider as it's designed for
-     * searching across pre-existing tables with varying schemas.
+     * Upsert a document by dynamically provisioning a table and columns.
      */
-    upsert(_doc: UpsertDocument, _namespace?: string): Promise<void>;
+    upsert(doc: UpsertDocument, namespace?: string): Promise<void>;
     /**
-     * Batch upsert is not supported for MultiTablePostgresProvider.
+     * Batch upsert documents by dynamically provisioning tables and columns based on CSV headers.
      */
-    batchUpsert(_docs: UpsertDocument[], _namespace?: string): Promise<void>;
+    batchUpsert(docs: UpsertDocument[], namespace?: string): Promise<void>;
     /**
      * Query all configured tables and merge results, sorted by cosine similarity score.
      */
@@ -871,6 +866,31 @@ declare class UniversalVectorProvider extends BaseVectorProvider {
  * LLMFactory — instantiates the correct ILLMProvider based on llmConfig.provider.
  */
 declare class LLMFactory {
+    /**
+     * Register a custom LLM provider factory at runtime.
+     *
+     * Use this to add support for any LLM backend (Azure OpenAI, Cohere, Mistral,
+     * Bedrock, etc.) without modifying this library's source code.
+     *
+     * @example
+     * // In your Next.js app initialization:
+     * import { LLMFactory } from '@retrivora-ai/rag-engine/server';
+     * import { MyCustomProvider } from './providers/MyCustomProvider';
+     *
+     * LLMFactory.register('my-provider', (config) => new MyCustomProvider(config));
+     *
+     * // Then set in your .env.local:
+     * // LLM_PROVIDER=my-provider
+     */
+    static register(name: string, factory: (config: LLMConfig) => ILLMProvider): void;
+    /**
+     * Unregister a previously registered custom provider.
+     */
+    static unregister(name: string): void;
+    /**
+     * List all registered provider names (built-in + custom).
+     */
+    static listProviders(): string[];
     static create(llmConfig: LLMConfig, embeddingConfig?: EmbeddingConfig): ILLMProvider;
     static getValidator(provider: LLMProvider): IProviderValidator | null;
     static getHealthChecker(provider: LLMProvider): IProviderHealthChecker | null;