@retrivora-ai/rag-engine 0.4.3 → 0.4.5

package/dist/{DocumentChunker-BICIjSuG.d.mts → DocumentChunker-3yElxTO3.d.mts}

@@ -73,15 +73,22 @@ interface ChunkOptions {
     docId?: string | number;
     /** Extra metadata to attach to every chunk */
     metadata?: Record<string, unknown>;
+    /** Characters used to split text, in order of priority */
+    separators?: string[];
 }
 declare class DocumentChunker {
     private readonly chunkSize;
     private readonly chunkOverlap;
-    constructor(chunkSize?: number, chunkOverlap?: number);
+    private readonly separators;
+    constructor(chunkSize?: number, chunkOverlap?: number, separators?: string[]);
     /**
-     * Split a single text string into overlapping chunks.
+     * Split a single text string into overlapping chunks using a recursive strategy.
      */
     chunk(text: string, options?: ChunkOptions): Chunk[];
+    /**
+     * Recursively split text based on separators.
+     */
+    private recursiveSplit;
     /**
      * Chunk multiple documents at once.
      */

package/dist/{DocumentChunker-BICIjSuG.d.ts → DocumentChunker-3yElxTO3.d.ts}

@@ -73,15 +73,22 @@ interface ChunkOptions {
     docId?: string | number;
     /** Extra metadata to attach to every chunk */
     metadata?: Record<string, unknown>;
+    /** Characters used to split text, in order of priority */
+    separators?: string[];
 }
 declare class DocumentChunker {
     private readonly chunkSize;
     private readonly chunkOverlap;
-    constructor(chunkSize?: number, chunkOverlap?: number);
+    private readonly separators;
+    constructor(chunkSize?: number, chunkOverlap?: number, separators?: string[]);
     /**
-     * Split a single text string into overlapping chunks.
+     * Split a single text string into overlapping chunks using a recursive strategy.
      */
     chunk(text: string, options?: ChunkOptions): Chunk[];
+    /**
+     * Recursively split text based on separators.
+     */
+    private recursiveSplit;
     /**
      * Chunk multiple documents at once.
      */

package/dist/{MongoDBProvider-ZKW34AEL.mjs → MongoDBProvider-RE3Q5S5B.mjs}

@@ -1,6 +1,6 @@
 import {
   MongoDBProvider
-} from "./chunk-IWHCAQEA.mjs";
+} from "./chunk-PQKTC73Y.mjs";
 import "./chunk-IMP6FUCY.mjs";
 import "./chunk-FWCSY2DS.mjs";
 export {

package/dist/{RagConfig-CVt24lbC.d.ts → RagConfig-BgRDL9Vy.d.mts}

@@ -18,6 +18,22 @@ interface IngestDocument {
 interface ChatResponse {
     reply: string;
     sources: VectorMatch[];
+    graphData?: GraphSearchResult;
+}
+interface GraphNode {
+    id: string;
+    label: string;
+    properties?: Record<string, unknown>;
+}
+interface Edge {
+    source: string;
+    target: string;
+    type: string;
+    properties?: Record<string, unknown>;
+}
+interface GraphSearchResult {
+    nodes: GraphNode[];
+    edges: Edge[];
 }
 
 /**
@@ -29,6 +45,7 @@ interface ChatResponse {
 declare const VECTOR_DB_PROVIDERS: readonly ["pinecone", "pgvector", "postgresql", "mongodb", "milvus", "qdrant", "chromadb", "redis", "weaviate", "rest", "universal_rest", "custom"];
 declare const LLM_PROVIDERS: readonly ["openai", "anthropic", "ollama", "gemini", "rest", "universal_rest", "custom"];
 declare const EMBEDDING_PROVIDERS: readonly ["openai", "ollama", "gemini", "rest", "universal_rest", "custom"];
+declare const GRAPH_DB_PROVIDERS: readonly ["neo4j", "memgraph", "simple", "custom"];
 
 /**
  * Master configuration interface for Retrivora AI.
@@ -62,6 +79,13 @@ interface VectorDBConfig {
      */
     options: Record<string, unknown>;
 }
+type GraphDBProvider = typeof GRAPH_DB_PROVIDERS[number];
+interface GraphDBConfig {
+    /** Which graph database to use */
+    provider: GraphDBProvider;
+    /** Provider-specific options (URI, credentials, etc.) */
+    options: Record<string, unknown>;
+}
 type LLMProvider = typeof LLM_PROVIDERS[number];
 interface LLMConfig {
     /** Which LLM provider to use */
@@ -132,6 +156,8 @@ interface UIConfig {
     visualStyle?: 'glass' | 'solid';
     /** Border radius: 'none', 'sm', 'md', 'lg', 'xl', 'full' */
     borderRadius?: 'none' | 'sm' | 'md' | 'lg' | 'xl' | 'full';
+    /** Whether to allow file uploads directly from the UI */
+    allowUpload?: boolean;
 }
 interface RAGConfig {
     /** Number of top-K chunks retrieved per query */
@@ -142,6 +168,14 @@ interface RAGConfig {
     chunkSize?: number;
     /** Overlap between adjacent chunks in tokens */
     chunkOverlap?: number;
+    /** Characters used to split text, in order of priority */
+    separators?: string[];
+    /** Whether to use query transformation (e.g. HyDE) */
+    useQueryTransformation?: boolean;
+    /** Whether to use graph-based retrieval */
+    useGraphRetrieval?: boolean;
+    /** Whether to perform reranking on retrieved results */
+    useReranking?: boolean;
 }
 interface RagConfig {
     /**
@@ -159,6 +193,8 @@ interface RagConfig {
     ui?: UIConfig;
     /** Optional RAG pipeline tuning knobs */
     rag?: RAGConfig;
+    /** Optional Graph database configuration */
+    graphDb?: GraphDBConfig;
 }
 
-export type { ChatResponse as C, EmbeddingConfig as E, IngestDocument as I, LLMConfig as L, RAGConfig as R, UIConfig as U, VectorMatch as V, EmbeddingProvider as a, LLMProvider as b, RagConfig as c, UpsertDocument as d, VectorDBConfig as e, VectorDBProvider as f };
+export type { ChatResponse as C, EmbeddingConfig as E, GraphDBConfig as G, IngestDocument as I, LLMConfig as L, RAGConfig as R, UIConfig as U, VectorMatch as V, EmbeddingProvider as a, LLMProvider as b, RagConfig as c, UpsertDocument as d, VectorDBConfig as e, VectorDBProvider as f, GraphNode as g, Edge as h, GraphSearchResult as i };

package/dist/{RagConfig-CVt24lbC.d.mts → RagConfig-BgRDL9Vy.d.ts}

@@ -18,6 +18,22 @@ interface IngestDocument {
 interface ChatResponse {
     reply: string;
     sources: VectorMatch[];
+    graphData?: GraphSearchResult;
+}
+interface GraphNode {
+    id: string;
+    label: string;
+    properties?: Record<string, unknown>;
+}
+interface Edge {
+    source: string;
+    target: string;
+    type: string;
+    properties?: Record<string, unknown>;
+}
+interface GraphSearchResult {
+    nodes: GraphNode[];
+    edges: Edge[];
 }
 
 /**
@@ -29,6 +45,7 @@ interface ChatResponse {
 declare const VECTOR_DB_PROVIDERS: readonly ["pinecone", "pgvector", "postgresql", "mongodb", "milvus", "qdrant", "chromadb", "redis", "weaviate", "rest", "universal_rest", "custom"];
 declare const LLM_PROVIDERS: readonly ["openai", "anthropic", "ollama", "gemini", "rest", "universal_rest", "custom"];
 declare const EMBEDDING_PROVIDERS: readonly ["openai", "ollama", "gemini", "rest", "universal_rest", "custom"];
+declare const GRAPH_DB_PROVIDERS: readonly ["neo4j", "memgraph", "simple", "custom"];
 
 /**
  * Master configuration interface for Retrivora AI.
@@ -62,6 +79,13 @@ interface VectorDBConfig {
      */
     options: Record<string, unknown>;
 }
+type GraphDBProvider = typeof GRAPH_DB_PROVIDERS[number];
+interface GraphDBConfig {
+    /** Which graph database to use */
+    provider: GraphDBProvider;
+    /** Provider-specific options (URI, credentials, etc.) */
+    options: Record<string, unknown>;
+}
 type LLMProvider = typeof LLM_PROVIDERS[number];
 interface LLMConfig {
     /** Which LLM provider to use */
@@ -132,6 +156,8 @@ interface UIConfig {
     visualStyle?: 'glass' | 'solid';
     /** Border radius: 'none', 'sm', 'md', 'lg', 'xl', 'full' */
     borderRadius?: 'none' | 'sm' | 'md' | 'lg' | 'xl' | 'full';
+    /** Whether to allow file uploads directly from the UI */
+    allowUpload?: boolean;
 }
 interface RAGConfig {
     /** Number of top-K chunks retrieved per query */
@@ -142,6 +168,14 @@ interface RAGConfig {
     chunkSize?: number;
     /** Overlap between adjacent chunks in tokens */
     chunkOverlap?: number;
+    /** Characters used to split text, in order of priority */
+    separators?: string[];
+    /** Whether to use query transformation (e.g. HyDE) */
+    useQueryTransformation?: boolean;
+    /** Whether to use graph-based retrieval */
+    useGraphRetrieval?: boolean;
+    /** Whether to perform reranking on retrieved results */
+    useReranking?: boolean;
 }
 interface RagConfig {
     /**
@@ -159,6 +193,8 @@ interface RagConfig {
     ui?: UIConfig;
     /** Optional RAG pipeline tuning knobs */
     rag?: RAGConfig;
+    /** Optional Graph database configuration */
+    graphDb?: GraphDBConfig;
 }
 
-export type { ChatResponse as C, EmbeddingConfig as E, IngestDocument as I, LLMConfig as L, RAGConfig as R, UIConfig as U, VectorMatch as V, EmbeddingProvider as a, LLMProvider as b, RagConfig as c, UpsertDocument as d, VectorDBConfig as e, VectorDBProvider as f };
+export type { ChatResponse as C, EmbeddingConfig as E, GraphDBConfig as G, IngestDocument as I, LLMConfig as L, RAGConfig as R, UIConfig as U, VectorMatch as V, EmbeddingProvider as a, LLMProvider as b, RagConfig as c, UpsertDocument as d, VectorDBConfig as e, VectorDBProvider as f, GraphNode as g, Edge as h, GraphSearchResult as i };

package/dist/SimpleGraphProvider-M6T7SE7D.mjs

@@ -0,0 +1,62 @@
+import "./chunk-FWCSY2DS.mjs";
+
+// src/providers/graphdb/BaseGraphProvider.ts
+var BaseGraphProvider = class {
+  constructor(config) {
+    this.config = config;
+  }
+};
+
+// src/providers/graphdb/SimpleGraphProvider.ts
+var SimpleGraphProvider = class extends BaseGraphProvider {
+  constructor() {
+    super(...arguments);
+    this.nodes = /* @__PURE__ */ new Map();
+    this.edges = [];
+  }
+  async initialize() {
+    console.log("[SimpleGraphProvider] Initialised in-memory graph store.");
+  }
+  async addNodes(nodes) {
+    for (const node of nodes) {
+      this.nodes.set(node.id, node);
+    }
+  }
+  async addEdges(edges) {
+    this.edges.push(...edges);
+  }
+  async query(queryText, limit = 5) {
+    const q = queryText.toLowerCase();
+    const matchedNodes = Array.from(this.nodes.values()).filter(
+      (node) => node.id.toLowerCase().includes(q) || node.label.toLowerCase().includes(q) || JSON.stringify(node.properties).toLowerCase().includes(q)
+    ).slice(0, limit);
+    const matchedNodeIds = new Set(matchedNodes.map((n) => n.id));
+    const matchedEdges = this.edges.filter(
+      (edge) => matchedNodeIds.has(edge.source) || matchedNodeIds.has(edge.target)
+    );
+    for (const edge of matchedEdges) {
+      if (!matchedNodeIds.has(edge.source)) {
+        const source = this.nodes.get(edge.source);
+        if (source) matchedNodes.push(source);
+      }
+      if (!matchedNodeIds.has(edge.target)) {
+        const target = this.nodes.get(edge.target);
+        if (target) matchedNodes.push(target);
+      }
+    }
+    return {
+      nodes: Array.from(new Set(matchedNodes)),
+      edges: matchedEdges
+    };
+  }
+  async ping() {
+    return true;
+  }
+  async disconnect() {
+    this.nodes.clear();
+    this.edges = [];
+  }
+};
+export {
+  SimpleGraphProvider
+};

package/dist/{chunk-IWHCAQEA.mjs → chunk-PQKTC73Y.mjs}

@@ -77,7 +77,7 @@ var MongoDBProvider = class extends BaseVectorProvider {
     return results.map((res) => ({
       id: res._id,
       content: res[this.contentKey],
-      metadata: res[this.metadataKey],
+      metadata: res[this.metadataKey] || {},
       score: res.score
     }));
   }

package/dist/{chunk-UUZ3F4WK.mjs → chunk-PRC5CZIZ.mjs}

@@ -77,7 +77,7 @@ function readEnum(env, name, fallback, allowed) {
   throw new Error(`[getRagConfig] ${name} must be one of: ${allowed.join(", ")}`);
 }
 function getRagConfig(env = process.env) {
-  var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A, _B, _C, _D, _E, _F, _G, _H, _I, _J, _K, _L, _M, _N, _O;
+  var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A, _B, _C, _D, _E, _F, _G, _H, _I, _J, _K, _L, _M, _N, _O, _P, _Q;
   const projectId = (_b = (_a = readString(env, "RAG_PROJECT_ID")) != null ? _a : readString(env, "NEXT_PUBLIC_PROJECT_ID")) != null ? _b : "__default__";
   const vectorProvider = readEnum(env, "VECTOR_DB_PROVIDER", "pinecone", VECTOR_DB_PROVIDERS);
   const llmProvider = readEnum(env, "LLM_PROVIDER", "openai", LLM_PROVIDERS);
@@ -174,7 +174,8 @@ function getRagConfig(env = process.env) {
       showSources: ((_I = (_H = readString(env, "NEXT_PUBLIC_SHOW_SOURCES")) != null ? _H : readString(env, "UI_SHOW_SOURCES")) != null ? _I : "true") !== "false",
       welcomeMessage: (_K = (_J = readString(env, "NEXT_PUBLIC_WELCOME_MESSAGE")) != null ? _J : readString(env, "UI_WELCOME_MESSAGE")) != null ? _K : "Hello! I'm your AI assistant. Ask me anything about your documents.",
       visualStyle: (_M = (_L = readString(env, "NEXT_PUBLIC_UI_VISUAL_STYLE")) != null ? _L : readString(env, "UI_VISUAL_STYLE")) != null ? _M : "glass",
-      borderRadius: (_O = (_N = readString(env, "NEXT_PUBLIC_UI_BORDER_RADIUS")) != null ? _N : readString(env, "UI_BORDER_RADIUS")) != null ? _O : "xl"
+      borderRadius: (_O = (_N = readString(env, "NEXT_PUBLIC_UI_BORDER_RADIUS")) != null ? _N : readString(env, "UI_BORDER_RADIUS")) != null ? _O : "xl",
+      allowUpload: ((_Q = (_P = readString(env, "NEXT_PUBLIC_ALLOW_UPLOAD")) != null ? _P : readString(env, "UI_ALLOW_UPLOAD")) != null ? _Q : "false") === "true"
     },
     rag: {
       topK: readNumber(env, "RAG_TOP_K", 5),
@@ -748,52 +749,84 @@ var ConfigValidator = class {
 
 // src/rag/DocumentChunker.ts
 var DocumentChunker = class {
-  constructor(chunkSize = 1e3, chunkOverlap = 200) {
+  constructor(chunkSize = 1e3, chunkOverlap = 200, separators = ["\n\n", "\n", " ", ""]) {
     this.chunkSize = chunkSize;
     this.chunkOverlap = chunkOverlap;
+    this.separators = separators;
   }
   /**
-   * Split a single text string into overlapping chunks.
+   * Split a single text string into overlapping chunks using a recursive strategy.
    */
   chunk(text, options = {}) {
     const {
       chunkSize = this.chunkSize,
       chunkOverlap = this.chunkOverlap,
       docId = `doc_${Date.now()}`,
-      metadata = {}
+      metadata = {},
+      separators = this.separators
     } = options;
-    const cleaned = text.replace(/\r\n/g, "\n").trim();
-    if (!cleaned) return [];
-    const sentences = cleaned.split(new RegExp("(?<=[.!?])\\s+|\\n{2,}")).map((s) => s.trim()).filter(Boolean);
-    const chunks = [];
-    let current = "";
+    const finalChunks = [];
+    const splits = this.recursiveSplit(text, separators, chunkSize);
+    let currentChunk = [];
+    let currentLength = 0;
     let chunkIndex = 0;
-    for (const sentence of sentences) {
-      if ((current + " " + sentence).trim().length <= chunkSize) {
-        current = current ? `${current} ${sentence}` : sentence;
-      } else {
-        if (current) {
-          chunks.push({
-            id: `${docId}_chunk_${chunkIndex++}`,
-            content: current.trim(),
-            metadata: __spreadProps(__spreadValues({}, metadata), { docId, chunkIndex: chunkIndex - 1 })
-          });
-        }
-        if (chunkOverlap > 0 && current.length > chunkOverlap) {
-          current = current.slice(-chunkOverlap) + " " + sentence;
-        } else {
-          current = sentence;
+    for (const split of splits) {
+      if (currentLength + split.length > chunkSize && currentChunk.length > 0) {
+        finalChunks.push({
+          id: `${docId}_chunk_${chunkIndex++}`,
+          content: currentChunk.join("").trim(),
+          metadata: __spreadProps(__spreadValues({}, metadata), { docId, chunkIndex: chunkIndex - 1 })
+        });
+        const overlapItems = [];
+        let overlapLen = 0;
+        for (let i = currentChunk.length - 1; i >= 0; i--) {
+          if (overlapLen + currentChunk[i].length <= chunkOverlap) {
+            overlapItems.unshift(currentChunk[i]);
+            overlapLen += currentChunk[i].length;
+          } else {
+            break;
+          }
         }
+        currentChunk = overlapItems;
+        currentLength = overlapLen;
       }
+      currentChunk.push(split);
+      currentLength += split.length;
     }
-    if (current.trim()) {
-      chunks.push({
+    if (currentChunk.length > 0) {
+      finalChunks.push({
         id: `${docId}_chunk_${chunkIndex}`,
-        content: current.trim(),
+        content: currentChunk.join("").trim(),
         metadata: __spreadProps(__spreadValues({}, metadata), { docId, chunkIndex })
       });
     }
-    return chunks;
+    return finalChunks;
+  }
+  /**
+   * Recursively split text based on separators.
+   */
+  recursiveSplit(text, separators, chunkSize) {
+    const finalSplits = [];
+    let separator = separators[separators.length - 1];
+    let nextSeparators = [];
+    for (let i = 0; i < separators.length; i++) {
+      if (text.includes(separators[i])) {
+        separator = separators[i];
+        nextSeparators = separators.slice(i + 1);
+        break;
+      }
+    }
+    const parts = text.split(separator);
+    for (const part of parts) {
+      if (part.length <= chunkSize) {
+        finalSplits.push(part + separator);
+      } else if (nextSeparators.length > 0) {
+        finalSplits.push(...this.recursiveSplit(part, nextSeparators, chunkSize));
+      } else {
+        finalSplits.push(part);
+      }
+    }
+    return finalSplits;
   }
   /**
    * Chunk multiple documents at once.
@@ -805,6 +838,39 @@ var DocumentChunker = class {
   }
 };
 
+// src/rag/EntityExtractor.ts
+var EntityExtractor = class {
+  constructor(llm) {
+    this.llm = llm;
+  }
+  /**
+   * Extract nodes and edges from a text chunk.
+   */
+  async extract(text) {
+    const prompt = `
+Extract entities and relationships from the following text.
+Format the output as JSON with two keys: "nodes" (array of {id, label, properties}) and "edges" (array of {source, target, type, properties}).
+Use the same ID for the same entity.
+
+Text:
+"${text}"
+
+Output JSON:
+`;
+    const response = await this.llm.chat([
+      { role: "system", content: "You are an expert at knowledge graph extraction. Respond only with valid JSON." },
+      { role: "user", content: prompt }
+    ], "");
+    try {
+      const cleanJson = response.replace(/```json\n?|\n?```/g, "").trim();
+      return JSON.parse(cleanJson);
+    } catch (e) {
+      console.warn("[EntityExtractor] Failed to parse LLM response as JSON:", response);
+      return { nodes: [], edges: [] };
+    }
+  }
+};
+
 // src/llm/providers/OpenAIProvider.ts
 import OpenAI from "openai";
 var OpenAIProvider = class {
@@ -1348,16 +1414,15 @@ var LLMFactory = class _LLMFactory {
 
 // src/core/ProviderRegistry.ts
 var ProviderRegistry = class {
-  /**
-   * Register a custom vector provider class by name.
-   * The name must match the provider value used in VectorDBConfig.provider.
-   *
-   * @example
-   * ProviderRegistry.registerVectorProvider('my-db', MyCustomProvider);
-   */
   static registerVectorProvider(name, providerClass) {
     this.vectorProviders[name] = providerClass;
   }
+  /**
+   * Register a custom graph provider class by name.
+   */
+  static registerGraphProvider(name, providerClass) {
+    this.graphProviders[name] = providerClass;
+  }
   /**
    * Creates a vector database provider based on the configuration.
    * Built-in providers are dynamically imported to avoid bundling all SDKs.
@@ -1378,7 +1443,7 @@ var ProviderRegistry = class {
         return new PostgreSQLProvider(config);
       }
       case "mongodb": {
-        const { MongoDBProvider } = await import("./MongoDBProvider-ZKW34AEL.mjs");
+        const { MongoDBProvider } = await import("./MongoDBProvider-RE3Q5S5B.mjs");
         return new MongoDBProvider(config);
       }
       case "milvus": {
@@ -1412,6 +1477,28 @@ var ProviderRegistry = class {
         );
     }
   }
+  /**
+   * Creates a graph database provider based on the configuration.
+   */
+  static async createGraphProvider(config) {
+    const { provider } = config;
+    if (this.graphProviders[provider]) {
+      return new this.graphProviders[provider](config);
+    }
+    switch (provider) {
+      case "neo4j": {
+        throw new Error("[ProviderRegistry] Neo4j provider not implemented yet.");
+      }
+      case "simple": {
+        const { SimpleGraphProvider } = await import("./SimpleGraphProvider-M6T7SE7D.mjs");
+        return new SimpleGraphProvider(config);
+      }
+      default:
+        throw new Error(
+          `[ProviderRegistry] Unsupported graph provider: "${provider}". Built-in providers: simple. For custom providers, call ProviderRegistry.registerGraphProvider("${provider}", YourClass).`
+        );
+    }
+  }
   /**
    * Creates an LLM provider based on the configuration.
    */
@@ -1420,6 +1507,7 @@ var ProviderRegistry = class {
   }
 };
 ProviderRegistry.vectorProviders = {};
+ProviderRegistry.graphProviders = {};
 
 // src/core/BatchProcessor.ts
 function isTransientError(error) {
@@ -1851,6 +1939,11 @@ var Pipeline = class {
     );
     this.llmProvider = llmProvider;
     this.embeddingProvider = embeddingProvider;
+    if (this.config.graphDb) {
+      this.graphDB = await ProviderRegistry.createGraphProvider(this.config.graphDb);
+      await this.graphDB.initialize();
+      this.entityExtractor = new EntityExtractor(this.llmProvider);
+    }
     await this.vectorDB.initialize();
     this.initialised = true;
   }
@@ -1904,6 +1997,30 @@ var Pipeline = class {
           docId: doc.docId,
           chunksIngested: upsertResult.totalProcessed
         });
+        if (this.graphDB && this.entityExtractor) {
+          console.log(`[Pipeline] Extracting entities for doc ${doc.docId} (${chunks.length} chunks)...`);
+          const extractionOptions = {
+            batchSize: 2,
+            // Low concurrency for LLM extraction
+            maxRetries: 1,
+            initialDelayMs: 500
+          };
+          await BatchProcessor.processBatch(
+            chunks,
+            async (batch) => {
+              for (const chunk of batch) {
+                try {
+                  const { nodes, edges } = await this.entityExtractor.extract(chunk.content);
+                  if (nodes.length > 0) await this.graphDB.addNodes(nodes);
+                  if (edges.length > 0) await this.graphDB.addEdges(edges);
+                } catch (err) {
+                  console.warn(`[Pipeline] Entity extraction failed for chunk:`, err);
+                }
+              }
+            },
+            extractionOptions
+          );
+        }
       } catch (error) {
         console.error(`[Pipeline] Failed to ingest document ${doc.docId}:`, error);
         results.push({ docId: doc.docId, chunksIngested: 0 });
@@ -1912,27 +2029,65 @@ var Pipeline = class {
     return results;
   }
   async ask(question, history = [], namespace) {
-    var _a, _b, _c, _d;
+    var _a, _b, _c, _d, _e, _f;
     await this.initialize();
     const ns = namespace != null ? namespace : this.config.projectId;
     const topK = (_b = (_a = this.config.rag) == null ? void 0 : _a.topK) != null ? _b : 5;
     const scoreThreshold = (_d = (_c = this.config.rag) == null ? void 0 : _c.scoreThreshold) != null ? _d : 0;
     try {
-      const queryVector = await this.embeddingProvider.embed(question, { taskType: "query" });
+      let searchQuery = question;
+      if ((_e = this.config.rag) == null ? void 0 : _e.useQueryTransformation) {
+        searchQuery = await this.rewriteQuery(question, history);
+      }
+      const queryVector = await this.embeddingProvider.embed(searchQuery, { taskType: "query" });
       const fieldHints = extractQueryFieldHints(question);
       const filter = buildQueryFilter(question, fieldHints);
       filter.__entityHints = fieldHints;
       const rawMatches = await this.vectorDB.query(queryVector, topK, ns, filter);
       const sources = rawMatches.filter((m) => m.score >= scoreThreshold);
-      const context = sources.length ? sources.map((m, i) => `[Source ${i + 1}]
+      let graphData;
+      if (this.graphDB && ((_f = this.config.rag) == null ? void 0 : _f.useGraphRetrieval)) {
+        graphData = await this.graphDB.query(searchQuery);
+      }
+      let context = sources.length ? sources.map((m, i) => `[Source ${i + 1}]
 ${m.content}`).join("\n\n---\n\n") : "No relevant context found.";
+      if (graphData && graphData.nodes.length > 0) {
+        const graphContext = graphData.nodes.map(
+          (n) => `Entity: ${n.label} (${n.id})${n.properties ? " - " + JSON.stringify(n.properties) : ""}`
+        ).join("\n");
+        context = `GRAPH KNOWLEDGE:
+${graphContext}
+
+VECTOR CONTEXT:
+${context}`;
+      }
       const messages = [...history, { role: "user", content: question }];
       const reply = await this.llmProvider.chat(messages, context);
-      return { reply, sources };
+      return { reply, sources, graphData };
     } catch (error) {
       throw new Error(`[Pipeline] Chat failed: ${error instanceof Error ? error.message : String(error)}`);
     }
   }
+  /**
+   * Rewrite the user query for better retrieval performance.
+   */
+  async rewriteQuery(question, history) {
+    const prompt = `
+Given the following conversation history and a new question, rewrite the question to be a better search query for a vector database. 
+Focus on extracting the core intent and entities. Do not answer the question, just rewrite it.
+
+History:
+${history.map((m) => `${m.role}: ${m.content}`).join("\n")}
+
+New Question: ${question}
+
+Optimized Search Query:`;
+    const rewrite = await this.llmProvider.chat([
+      { role: "system", content: "You are an assistant that optimizes search queries for RAG systems." },
+      { role: "user", content: prompt }
+    ], "");
+    return rewrite.trim() || question;
+  }
 };
 
 // src/core/ProviderHealthCheck.ts
@@ -2460,7 +2615,7 @@ var DocumentParser = class {
     }
     if (extension === "pdf" || mimeType === "application/pdf") {
       try {
-        const pdf = await import("pdf-parse/lib/pdf-parse.js");
+        const pdf = await import("pdf-parse");
         const buffer = Buffer.isBuffer(file) ? file : Buffer.from(await file.arrayBuffer());
         const data = await pdf.default(buffer);
         return data.text;

package/dist/handlers/index.d.mts

@@ -1,3 +1,3 @@
-import '../RagConfig-CVt24lbC.mjs';
-export { c as createChatHandler, a as createHealthHandler, b as createIngestHandler, d as createUploadHandler } from '../index-OI2--lvT.mjs';
+import '../RagConfig-BgRDL9Vy.mjs';
+export { c as createChatHandler, a as createHealthHandler, b as createIngestHandler, d as createUploadHandler } from '../index-7qeLTPBL.mjs';
 import 'next/server';

package/dist/handlers/index.d.ts

@@ -1,3 +1,3 @@
-import '../RagConfig-CVt24lbC.js';
-export { c as createChatHandler, a as createHealthHandler, b as createIngestHandler, d as createUploadHandler } from '../index-D1hoNXMT.js';
+import '../RagConfig-BgRDL9Vy.js';
+export { c as createChatHandler, a as createHealthHandler, b as createIngestHandler, d as createUploadHandler } from '../index-DowY4_K0.js';
 import 'next/server';