@retrivora-ai/rag-engine 0.1.4 → 0.1.6

package/dist/DocumentChunker-BEyzadsv.d.mts

@@ -0,0 +1,93 @@
+/**
+ * Generic LLM Provider interface.
+ * Covers both chat completion and embedding generation so a single
+ * provider can handle both responsibilities when appropriate.
+ */
+interface ChatMessage {
+    role: 'user' | 'assistant' | 'system';
+    content: string;
+}
+interface ChatOptions {
+    /** Override max tokens for this specific call */
+    maxTokens?: number;
+    /** Override temperature for this specific call */
+    temperature?: number;
+    /** Stop sequences */
+    stop?: string[];
+}
+interface EmbedOptions {
+    /** Override model for this specific embed call */
+    model?: string;
+}
+interface ILLMProvider {
+    /**
+     * Send a chat completion request.
+     * @param messages – the full conversation history
+     * @param context  – retrieved RAG context to inject
+     * @param options  – optional per-call overrides
+     * @returns        – the assistant's reply text
+     */
+    chat(messages: ChatMessage[], context: string, options?: ChatOptions): Promise<string>;
+    /**
+     * Generate an embedding vector for the given text.
+     * @param text    – text to embed
+     * @param options – optional overrides
+     * @returns       – float array (the embedding)
+     */
+    embed(text: string, options?: EmbedOptions): Promise<number[]>;
+    /**
+     * Generate embedding vectors for multiple texts in a single batch.
+     * @param texts   – array of texts to embed
+     * @param options – optional overrides
+     * @returns       – array of float arrays
+     */
+    batchEmbed(texts: string[], options?: EmbedOptions): Promise<number[][]>;
+    /**
+     * Check if the provider endpoint is reachable.
+     */
+    ping(): Promise<boolean>;
+}
+
+/**
+ * DocumentChunker — splits long text into overlapping chunks
+ * suitable for vector embedding and retrieval.
+ *
+ * Strategy: sentence-aware sliding window
+ *   1. Split on sentence boundaries
+ *   2. Accumulate sentences into chunks up to `chunkSize` characters
+ *   3. Carry over `chunkOverlap` characters from the previous chunk
+ */
+interface Chunk {
+    id: string;
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface ChunkOptions {
+    /** Target chunk size in characters (default 1000) */
+    chunkSize?: number;
+    /** Overlap between consecutive chunks in characters (default 200) */
+    chunkOverlap?: number;
+    /** Source document identifier used as ID prefix */
+    docId?: string | number;
+    /** Extra metadata to attach to every chunk */
+    metadata?: Record<string, unknown>;
+}
+declare class DocumentChunker {
+    private readonly chunkSize;
+    private readonly chunkOverlap;
+    constructor(chunkSize?: number, chunkOverlap?: number);
+    /**
+     * Split a single text string into overlapping chunks.
+     */
+    chunk(text: string, options?: ChunkOptions): Chunk[];
+    /**
+     * Chunk multiple documents at once.
+     */
+    chunkMany(documents: Array<{
+        content: string;
+        docId?: string | number;
+        metadata?: Record<string, unknown>;
+    }>): Chunk[];
+}
+
+export { type ChatMessage as C, DocumentChunker as D, type EmbedOptions as E, type ILLMProvider as I, type ChatOptions as a, type Chunk as b, type ChunkOptions as c };

package/dist/DocumentChunker-BEyzadsv.d.ts

@@ -0,0 +1,93 @@
+/**
+ * Generic LLM Provider interface.
+ * Covers both chat completion and embedding generation so a single
+ * provider can handle both responsibilities when appropriate.
+ */
+interface ChatMessage {
+    role: 'user' | 'assistant' | 'system';
+    content: string;
+}
+interface ChatOptions {
+    /** Override max tokens for this specific call */
+    maxTokens?: number;
+    /** Override temperature for this specific call */
+    temperature?: number;
+    /** Stop sequences */
+    stop?: string[];
+}
+interface EmbedOptions {
+    /** Override model for this specific embed call */
+    model?: string;
+}
+interface ILLMProvider {
+    /**
+     * Send a chat completion request.
+     * @param messages – the full conversation history
+     * @param context  – retrieved RAG context to inject
+     * @param options  – optional per-call overrides
+     * @returns        – the assistant's reply text
+     */
+    chat(messages: ChatMessage[], context: string, options?: ChatOptions): Promise<string>;
+    /**
+     * Generate an embedding vector for the given text.
+     * @param text    – text to embed
+     * @param options – optional overrides
+     * @returns       – float array (the embedding)
+     */
+    embed(text: string, options?: EmbedOptions): Promise<number[]>;
+    /**
+     * Generate embedding vectors for multiple texts in a single batch.
+     * @param texts   – array of texts to embed
+     * @param options – optional overrides
+     * @returns       – array of float arrays
+     */
+    batchEmbed(texts: string[], options?: EmbedOptions): Promise<number[][]>;
+    /**
+     * Check if the provider endpoint is reachable.
+     */
+    ping(): Promise<boolean>;
+}
+
+/**
+ * DocumentChunker — splits long text into overlapping chunks
+ * suitable for vector embedding and retrieval.
+ *
+ * Strategy: sentence-aware sliding window
+ *   1. Split on sentence boundaries
+ *   2. Accumulate sentences into chunks up to `chunkSize` characters
+ *   3. Carry over `chunkOverlap` characters from the previous chunk
+ */
+interface Chunk {
+    id: string;
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface ChunkOptions {
+    /** Target chunk size in characters (default 1000) */
+    chunkSize?: number;
+    /** Overlap between consecutive chunks in characters (default 200) */
+    chunkOverlap?: number;
+    /** Source document identifier used as ID prefix */
+    docId?: string | number;
+    /** Extra metadata to attach to every chunk */
+    metadata?: Record<string, unknown>;
+}
+declare class DocumentChunker {
+    private readonly chunkSize;
+    private readonly chunkOverlap;
+    constructor(chunkSize?: number, chunkOverlap?: number);
+    /**
+     * Split a single text string into overlapping chunks.
+     */
+    chunk(text: string, options?: ChunkOptions): Chunk[];
+    /**
+     * Chunk multiple documents at once.
+     */
+    chunkMany(documents: Array<{
+        content: string;
+        docId?: string | number;
+        metadata?: Record<string, unknown>;
+    }>): Chunk[];
+}
+
+export { type ChatMessage as C, DocumentChunker as D, type EmbedOptions as E, type ILLMProvider as I, type ChatOptions as a, type Chunk as b, type ChunkOptions as c };

package/dist/{LLMFactory-XC55FTD2.mjs → LLMFactory-JFOY2V4X.mjs}

@@ -1,6 +1,7 @@
 import {
   LLMFactory
-} from "./chunk-GD3QJFNN.mjs";
+} from "./chunk-JI6VD5TJ.mjs";
+import "./chunk-UKDXCXW7.mjs";
 import "./chunk-I4E63NIC.mjs";
 export {
   LLMFactory

package/dist/{PineconeProvider-NJ675H7U.mjs → PineconeProvider-ZRAFNFEC.mjs}

@@ -1,6 +1,6 @@
 import {
   PineconeProvider
-} from "./chunk-AALIF3AL.mjs";
+} from "./chunk-ZM6TYIDH.mjs";
 import "./chunk-VOIWNO5O.mjs";
 import "./chunk-I4E63NIC.mjs";
 export {

package/dist/{PostgreSQLProvider-ISNMD3BE.mjs → PostgreSQLProvider-ZNXA67IM.mjs}

@@ -1,6 +1,6 @@
 import {
   PostgreSQLProvider
-} from "./chunk-S5DRHETN.mjs";
+} from "./chunk-IUTAZ7QR.mjs";
 import "./chunk-VOIWNO5O.mjs";
 import "./chunk-I4E63NIC.mjs";
 export {

package/dist/{QdrantProvider-LJWOIGES.mjs → QdrantProvider-NYGHAA7C.mjs}

@@ -1,6 +1,6 @@
 import {
   QdrantProvider
-} from "./chunk-6FODXNUF.mjs";
+} from "./chunk-5K23H7JL.mjs";
 import "./chunk-VOIWNO5O.mjs";
 import "./chunk-I4E63NIC.mjs";
 export {

package/dist/{Pipeline-Bo6CUCox.d.mts → RagConfig-hBGXJmSx.d.mts}

@@ -1,3 +1,25 @@
+interface VectorMatch {
+    id: string | number;
+    score: number;
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface UpsertDocument {
+    id: string | number;
+    vector: number[];
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface IngestDocument {
+    docId: string | number;
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface ChatResponse {
+    reply: string;
+    sources: VectorMatch[];
+}
+
 /**
  * Master configuration interface for Retrivora AI.
  * Each consuming project provides one RagConfig object to drive
@@ -120,96 +142,4 @@ interface RagConfig {
     rag?: RAGConfig;
 }
 
-/**
- * Generic LLM Provider interface.
- * Covers both chat completion and embedding generation so a single
- * provider can handle both responsibilities when appropriate.
- */
-interface ChatMessage {
-    role: 'user' | 'assistant' | 'system';
-    content: string;
-}
-interface ChatOptions {
-    /** Override max tokens for this specific call */
-    maxTokens?: number;
-    /** Override temperature for this specific call */
-    temperature?: number;
-    /** Stop sequences */
-    stop?: string[];
-}
-interface EmbedOptions {
-    /** Override model for this specific embed call */
-    model?: string;
-}
-interface ILLMProvider {
-    /**
-     * Send a chat completion request.
-     * @param messages – the full conversation history
-     * @param context  – retrieved RAG context to inject
-     * @param options  – optional per-call overrides
-     * @returns        – the assistant's reply text
-     */
-    chat(messages: ChatMessage[], context: string, options?: ChatOptions): Promise<string>;
-    /**
-     * Generate an embedding vector for the given text.
-     * @param text    – text to embed
-     * @param options – optional overrides
-     * @returns       – float array (the embedding)
-     */
-    embed(text: string, options?: EmbedOptions): Promise<number[]>;
-    /**
-     * Generate embedding vectors for multiple texts in a single batch.
-     * @param texts   – array of texts to embed
-     * @param options – optional overrides
-     * @returns       – array of float arrays
-     */
-    batchEmbed(texts: string[], options?: EmbedOptions): Promise<number[][]>;
-    /**
-     * Check if the provider endpoint is reachable.
-     */
-    ping(): Promise<boolean>;
-}
-
-interface VectorMatch {
-    id: string;
-    score: number;
-    content: string;
-    metadata?: Record<string, unknown>;
-}
-interface UpsertDocument {
-    id: string;
-    vector: number[];
-    content: string;
-    metadata?: Record<string, unknown>;
-}
-
-interface IngestDocument {
-    docId: string;
-    content: string;
-    metadata?: Record<string, unknown>;
-}
-interface ChatResponse {
-    reply: string;
-    sources: VectorMatch[];
-}
-/**
- * Pipeline — orchestrates the RAG flow: Embed → Search → Augment → Generate.
- */
-declare class Pipeline {
-    private vectorDB;
-    private llmProvider;
-    private embeddingProvider;
-    private chunker;
-    private config;
-    private initialised;
-    constructor(config: RagConfig);
-    initialize(): Promise<void>;
-    ingest(documents: IngestDocument[], namespace?: string): Promise<Array<{
-        docId: string;
-        chunksIngested: number;
-    }>>;
-    ask(question: string, history?: ChatMessage[], namespace?: string): Promise<ChatResponse>;
-    health(): Promise<Record<string, boolean>>;
-}
-
-export { type ChatMessage as C, type EmbedOptions as E, type ILLMProvider as I, type LLMConfig as L, Pipeline as P, type RAGConfig as R, type UIConfig as U, type VectorMatch as V, type ChatOptions as a, type ChatResponse as b, type EmbeddingConfig as c, type IngestDocument as d, type RagConfig as e, type UpsertDocument as f, type VectorDBConfig as g };
+export type { ChatResponse as C, EmbeddingConfig as E, IngestDocument as I, LLMConfig as L, RAGConfig as R, UIConfig as U, VectorMatch as V, RagConfig as a, UpsertDocument as b, VectorDBConfig as c };

package/dist/{Pipeline-Bo6CUCox.d.ts → RagConfig-hBGXJmSx.d.ts}

@@ -1,3 +1,25 @@
+interface VectorMatch {
+    id: string | number;
+    score: number;
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface UpsertDocument {
+    id: string | number;
+    vector: number[];
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface IngestDocument {
+    docId: string | number;
+    content: string;
+    metadata?: Record<string, unknown>;
+}
+interface ChatResponse {
+    reply: string;
+    sources: VectorMatch[];
+}
+
 /**
  * Master configuration interface for Retrivora AI.
  * Each consuming project provides one RagConfig object to drive
@@ -120,96 +142,4 @@ interface RagConfig {
     rag?: RAGConfig;
 }
 
-/**
- * Generic LLM Provider interface.
- * Covers both chat completion and embedding generation so a single
- * provider can handle both responsibilities when appropriate.
- */
-interface ChatMessage {
-    role: 'user' | 'assistant' | 'system';
-    content: string;
-}
-interface ChatOptions {
-    /** Override max tokens for this specific call */
-    maxTokens?: number;
-    /** Override temperature for this specific call */
-    temperature?: number;
-    /** Stop sequences */
-    stop?: string[];
-}
-interface EmbedOptions {
-    /** Override model for this specific embed call */
-    model?: string;
-}
-interface ILLMProvider {
-    /**
-     * Send a chat completion request.
-     * @param messages – the full conversation history
-     * @param context  – retrieved RAG context to inject
-     * @param options  – optional per-call overrides
-     * @returns        – the assistant's reply text
-     */
-    chat(messages: ChatMessage[], context: string, options?: ChatOptions): Promise<string>;
-    /**
-     * Generate an embedding vector for the given text.
-     * @param text    – text to embed
-     * @param options – optional overrides
-     * @returns       – float array (the embedding)
-     */
-    embed(text: string, options?: EmbedOptions): Promise<number[]>;
-    /**
-     * Generate embedding vectors for multiple texts in a single batch.
-     * @param texts   – array of texts to embed
-     * @param options – optional overrides
-     * @returns       – array of float arrays
-     */
-    batchEmbed(texts: string[], options?: EmbedOptions): Promise<number[][]>;
-    /**
-     * Check if the provider endpoint is reachable.
-     */
-    ping(): Promise<boolean>;
-}
-
-interface VectorMatch {
-    id: string;
-    score: number;
-    content: string;
-    metadata?: Record<string, unknown>;
-}
-interface UpsertDocument {
-    id: string;
-    vector: number[];
-    content: string;
-    metadata?: Record<string, unknown>;
-}
-
-interface IngestDocument {
-    docId: string;
-    content: string;
-    metadata?: Record<string, unknown>;
-}
-interface ChatResponse {
-    reply: string;
-    sources: VectorMatch[];
-}
-/**
- * Pipeline — orchestrates the RAG flow: Embed → Search → Augment → Generate.
- */
-declare class Pipeline {
-    private vectorDB;
-    private llmProvider;
-    private embeddingProvider;
-    private chunker;
-    private config;
-    private initialised;
-    constructor(config: RagConfig);
-    initialize(): Promise<void>;
-    ingest(documents: IngestDocument[], namespace?: string): Promise<Array<{
-        docId: string;
-        chunksIngested: number;
-    }>>;
-    ask(question: string, history?: ChatMessage[], namespace?: string): Promise<ChatResponse>;
-    health(): Promise<Record<string, boolean>>;
-}
-
-export { type ChatMessage as C, type EmbedOptions as E, type ILLMProvider as I, type LLMConfig as L, Pipeline as P, type RAGConfig as R, type UIConfig as U, type VectorMatch as V, type ChatOptions as a, type ChatResponse as b, type EmbeddingConfig as c, type IngestDocument as d, type RagConfig as e, type UpsertDocument as f, type VectorDBConfig as g };
+export type { ChatResponse as C, EmbeddingConfig as E, IngestDocument as I, LLMConfig as L, RAGConfig as R, UIConfig as U, VectorMatch as V, RagConfig as a, UpsertDocument as b, VectorDBConfig as c };

package/dist/{chunk-6FODXNUF.mjs → chunk-5K23H7JL.mjs}

@@ -22,6 +22,33 @@ var QdrantProvider = class extends BaseVectorProvider {
   }
   async initialize() {
     await this.ping();
+    await this.ensureIndex();
+  }
+  /**
+   * Ensures that the 'namespace' field has a keyword index for efficient filtering.
+   * Qdrant requires this for search filters to work in many configurations.
+   */
+  async ensureIndex() {
+    var _a, _b;
+    try {
+      await this.http.put(`/collections/${this.indexName}/index`, {
+        field_name: "namespace",
+        field_schema: "keyword"
+      });
+      console.log(`[QdrantProvider] \u2705 Ensured keyword index for "namespace" on collection "${this.indexName}"`);
+    } catch (err) {
+      let status;
+      let data;
+      if (axios.isAxiosError(err)) {
+        status = (_a = err.response) == null ? void 0 : _a.status;
+        data = (_b = err.response) == null ? void 0 : _b.data;
+      }
+      if (status === 409) {
+        return;
+      }
+      const errorMessage = err instanceof Error ? err.message : String(err);
+      console.warn(`[QdrantProvider] \u26A0\uFE0F Could not ensure namespace index (Status: ${status}):`, JSON.stringify(data || errorMessage));
+    }
   }
   async upsert(doc, namespace) {
     await this.batchUpsert([doc], namespace);
@@ -66,7 +93,6 @@ var QdrantProvider = class extends BaseVectorProvider {
       points: [id]
     });
   }
-  // eslint-disable-next-line @typescript-eslint/no-unused-vars
   async deleteNamespace(_namespace) {
     await this.http.post(`/collections/${this.indexName}/points/delete`, {
       filter: {