@retrivora-ai/rag-engine 1.8.1 → 1.8.3

package/dist/{ILLMProvider-BfRgI1Xh.d.mts → ILLMProvider-BOJFz3Na.d.mts}

@@ -7,6 +7,8 @@ interface RagMessage extends ChatMessage {
     id: string;
     sources?: VectorMatch[];
     uiTransformation?: unknown;
+    /** Full observability trace emitted by the backend. Only present on assistant messages. */
+    trace?: ObservabilityTrace;
     createdAt: string;
 }
 interface ChatOptions {
@@ -233,6 +235,48 @@ interface RagConfig {
     graphDb?: GraphDBConfig;
 }
 
+/** Per-stage latency breakdown for a single RAG request. */
+interface LatencyBreakdown {
+    embedMs: number;
+    retrieveMs: number;
+    rerankMs?: number;
+    generateMs: number;
+    totalMs: number;
+}
+/** Token usage reported by the LLM (or estimated when the provider doesn't expose it). */
+interface TokenUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+    /** Rough cost in USD — estimated from token counts and model pricing. */
+    estimatedCostUsd?: number;
+}
+/** A single retrieved chunk annotated with its retrieval context. */
+interface RetrievedChunk {
+    id: string | number;
+    score: number;
+    content: string;
+    metadata: Record<string, unknown>;
+    namespace: string;
+}
+/**
+ * Full observability trace for one RAG request.
+ * Emitted by the backend as an SSE frame and stored on each RagMessage.
+ */
+interface ObservabilityTrace {
+    requestId: string;
+    query: string;
+    rewrittenQuery?: string;
+    systemPrompt: string;
+    userPrompt: string;
+    chunks: RetrievedChunk[];
+    latency: LatencyBreakdown;
+    tokens?: TokenUsage;
+    /** 0 = fully grounded, 1 = likely hallucinated. */
+    hallucinationScore?: number;
+    hallucinationReason?: string;
+    timestamp: string;
+}
 interface VectorMatch {
     id: string | number;
     score: number;
@@ -255,6 +299,8 @@ interface ChatResponse {
     sources: VectorMatch[];
     graphData?: GraphSearchResult;
     ui_transformation?: unknown;
+    /** Observability trace — populated by the instrumented Pipeline. */
+    trace?: ObservabilityTrace;
 }
 interface GraphNode {
     id: string;
@@ -331,4 +377,4 @@ interface ILLMProvider {
     ping(): Promise<boolean>;
 }
 
-export type { ChatMessage as C, EmbedOptions as E, GraphDBConfig as G, ILLMProvider as I, LLMConfig as L, Product as P, RagMessage as R, UIConfig as U, VectorMatch as V, UseRagChatOptions as a, UseRagChatReturn as b, ChatOptions as c, ChatResponse as d, EmbeddingConfig as e, EmbeddingProvider as f, IngestDocument as g, LLMProvider as h, RAGConfig as i, RagConfig as j, UpsertDocument as k, VectorDBConfig as l, VectorDBProvider as m, RetrievalResult as n, GraphNode as o, Edge as p, GraphSearchResult as q };
+export type { ChatMessage as C, EmbedOptions as E, GraphDBConfig as G, ILLMProvider as I, LLMConfig as L, ObservabilityTrace as O, Product as P, RagMessage as R, TokenUsage as T, UIConfig as U, VectorMatch as V, UseRagChatOptions as a, UseRagChatReturn as b, ChatOptions as c, ChatResponse as d, EmbeddingConfig as e, EmbeddingProvider as f, IngestDocument as g, LLMProvider as h, LatencyBreakdown as i, RAGConfig as j, RagConfig as k, RetrievedChunk as l, UpsertDocument as m, VectorDBConfig as n, VectorDBProvider as o, RetrievalResult as p, GraphNode as q, Edge as r, GraphSearchResult as s };

package/dist/{ILLMProvider-BfRgI1Xh.d.ts → ILLMProvider-BOJFz3Na.d.ts}

@@ -7,6 +7,8 @@ interface RagMessage extends ChatMessage {
     id: string;
     sources?: VectorMatch[];
     uiTransformation?: unknown;
+    /** Full observability trace emitted by the backend. Only present on assistant messages. */
+    trace?: ObservabilityTrace;
     createdAt: string;
 }
 interface ChatOptions {
@@ -233,6 +235,48 @@ interface RagConfig {
     graphDb?: GraphDBConfig;
 }
 
+/** Per-stage latency breakdown for a single RAG request. */
+interface LatencyBreakdown {
+    embedMs: number;
+    retrieveMs: number;
+    rerankMs?: number;
+    generateMs: number;
+    totalMs: number;
+}
+/** Token usage reported by the LLM (or estimated when the provider doesn't expose it). */
+interface TokenUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+    /** Rough cost in USD — estimated from token counts and model pricing. */
+    estimatedCostUsd?: number;
+}
+/** A single retrieved chunk annotated with its retrieval context. */
+interface RetrievedChunk {
+    id: string | number;
+    score: number;
+    content: string;
+    metadata: Record<string, unknown>;
+    namespace: string;
+}
+/**
+ * Full observability trace for one RAG request.
+ * Emitted by the backend as an SSE frame and stored on each RagMessage.
+ */
+interface ObservabilityTrace {
+    requestId: string;
+    query: string;
+    rewrittenQuery?: string;
+    systemPrompt: string;
+    userPrompt: string;
+    chunks: RetrievedChunk[];
+    latency: LatencyBreakdown;
+    tokens?: TokenUsage;
+    /** 0 = fully grounded, 1 = likely hallucinated. */
+    hallucinationScore?: number;
+    hallucinationReason?: string;
+    timestamp: string;
+}
 interface VectorMatch {
     id: string | number;
     score: number;
@@ -255,6 +299,8 @@ interface ChatResponse {
     sources: VectorMatch[];
     graphData?: GraphSearchResult;
     ui_transformation?: unknown;
+    /** Observability trace — populated by the instrumented Pipeline. */
+    trace?: ObservabilityTrace;
 }
 interface GraphNode {
     id: string;
@@ -331,4 +377,4 @@ interface ILLMProvider {
     ping(): Promise<boolean>;
 }
 
-export type { ChatMessage as C, EmbedOptions as E, GraphDBConfig as G, ILLMProvider as I, LLMConfig as L, Product as P, RagMessage as R, UIConfig as U, VectorMatch as V, UseRagChatOptions as a, UseRagChatReturn as b, ChatOptions as c, ChatResponse as d, EmbeddingConfig as e, EmbeddingProvider as f, IngestDocument as g, LLMProvider as h, RAGConfig as i, RagConfig as j, UpsertDocument as k, VectorDBConfig as l, VectorDBProvider as m, RetrievalResult as n, GraphNode as o, Edge as p, GraphSearchResult as q };
+export type { ChatMessage as C, EmbedOptions as E, GraphDBConfig as G, ILLMProvider as I, LLMConfig as L, ObservabilityTrace as O, Product as P, RagMessage as R, TokenUsage as T, UIConfig as U, VectorMatch as V, UseRagChatOptions as a, UseRagChatReturn as b, ChatOptions as c, ChatResponse as d, EmbeddingConfig as e, EmbeddingProvider as f, IngestDocument as g, LLMProvider as h, LatencyBreakdown as i, RAGConfig as j, RagConfig as k, RetrievedChunk as l, UpsertDocument as m, VectorDBConfig as n, VectorDBProvider as o, RetrievalResult as p, GraphNode as q, Edge as r, GraphSearchResult as s };

package/dist/handlers/index.d.mts

@@ -1,3 +1,3 @@
-import '../ILLMProvider-BfRgI1Xh.mjs';
-export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, k as createSuggestionsHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame, l as sseUIFrame } from '../index-BV0z5mb6.mjs';
+import '../ILLMProvider-BOJFz3Na.mjs';
+export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, k as createSuggestionsHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, l as sseObservabilityFrame, j as sseTextFrame, m as sseUIFrame } from '../index-D3V9Et2M.mjs';
 import 'next/server';

package/dist/handlers/index.d.ts

@@ -1,3 +1,3 @@
-import '../ILLMProvider-BfRgI1Xh.js';
-export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, k as createSuggestionsHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame, l as sseUIFrame } from '../index-1Z4GuYBi.js';
+import '../ILLMProvider-BOJFz3Na.js';
+export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, k as createSuggestionsHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, l as sseObservabilityFrame, j as sseTextFrame, m as sseUIFrame } from '../index-BwpcaziY.js';
 import 'next/server';