@steno-ai/engine 0.1.17 → 0.1.19

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@steno-ai/engine",
-  "version": "0.1.17",
+  "version": "0.1.19",
   "description": "Core memory engine — extraction, retrieval, knowledge graph",
   "license": "MIT",
   "repository": {
@@ -22,6 +22,13 @@
     "src",
     "README.md"
   ],
+  "scripts": {
+    "test": "vitest run",
+    "test:watch": "vitest watch",
+    "typecheck": "tsc --noEmit",
+    "build": "tsc",
+    "clean": "rm -rf dist"
+  },
   "dependencies": {
     "bcryptjs": "^3.0.3",
     "compromise": "^14.15.0",
@@ -31,12 +38,5 @@
     "@types/bcryptjs": "^3.0.0",
     "typescript": "^5.7",
     "vitest": "^3"
-  },
-  "scripts": {
-    "test": "vitest run",
-    "test:watch": "vitest watch",
-    "typecheck": "tsc --noEmit",
-    "build": "tsc",
-    "clean": "rm -rf dist"
   }
-}
+}

package/src/extraction/structured-cross-linker.ts

@@ -105,10 +105,13 @@ export async function findPendingCrossLinks(
   tenantId: string,
   scope: string,
   scopeId: string,
-  options?: { minSimilarity?: number; maxCandidates?: number },
+  options?: { minSimilarity?: number; maxCandidates?: number; maxFacts?: number; maxDurationMs?: number },
 ): Promise<PendingCrossLink[]> {
   const minSim = options?.minSimilarity ?? 0.6;
-  const maxCandidates = options?.maxCandidates ?? 50;
+  const maxCandidates = options?.maxCandidates ?? 20;
+  const maxFacts = options?.maxFacts ?? 10;
+  const maxDurationMs = options?.maxDurationMs ?? 20000;
+  const startTime = Date.now();
 
   // Find structured facts by searching for the "structured" tag content
   // We use keyword search since there's no listFacts method
@@ -117,7 +120,7 @@ export async function findPendingCrossLinks(
     tenantId,
     scope,
     scopeId,
-    limit: 100,
+    limit: maxFacts,
   });
 
   if (recentStructuredFacts.length === 0) return [];
@@ -126,6 +129,12 @@ export async function findPendingCrossLinks(
 
   // For each structured fact, find semantically similar facts from different source types
   for (const match of recentStructuredFacts) {
+    // Time guard — stop before we exceed the budget
+    if (Date.now() - startTime > maxDurationMs) {
+      console.error(`[steno-structured-xlink] Time budget exhausted after ${candidates.length} candidates`);
+      break;
+    }
+
     const fact = match.fact;
     if (!fact.tags?.includes('structured')) continue;
 
@@ -146,10 +155,8 @@ export async function findPendingCrossLinks(
       if (match.fact.id === fact.id) continue;
       if (match.fact.sourceType === fact.sourceType) continue;
 
-      // Skip if already linked (check if edge exists between their entities)
-      // This is a lightweight check — the full edge check happens in processPendingCrossLinks
       candidates.push({
-        entityId: '', // filled by caller
+        entityId: '',
         entityName: '',
         factId: fact.id,
         factContent: fact.content,
@@ -183,8 +190,14 @@ export async function processPendingCrossLinks(
   tenantId: string,
   scope: string,
   scopeId: string,
+  options?: { maxDurationMs?: number; maxFacts?: number },
 ): Promise<{ processed: number; edgesCreated: number }> {
-  const candidates = await findPendingCrossLinks(storage, embedding, tenantId, scope, scopeId);
+  const maxDurationMs = options?.maxDurationMs ?? 20000;
+  const candidates = await findPendingCrossLinks(storage, embedding, tenantId, scope, scopeId, {
+    maxFacts: options?.maxFacts ?? 10,
+    maxDurationMs,
+    maxCandidates: 10,
+  });
 
   if (candidates.length === 0) return { processed: 0, edgesCreated: 0 };
 

package/src/extraction/structured-extractor.ts

@@ -94,6 +94,8 @@ function formatTime(iso: string): string {
 // ---------------------------------------------------------------------------
 
 export function extractStructuredEvent(data: StructuredEvent): ExtractionResult {
+  if (!data?.title) return { facts: [], entities: [], edges: [], tier: 'heuristic', confidence: 1.0, tokensInput: 0, tokensOutput: 0, model: null };
+
   const entities: ExtractedEntity[] = [];
   const edges: ExtractedEdge[] = [];
 
@@ -202,6 +204,8 @@ export function extractStructuredEvent(data: StructuredEvent): ExtractionResult
 }
 
 export function extractStructuredTask(data: StructuredTask): ExtractionResult {
+  if (!data?.title) return { facts: [], entities: [], edges: [], tier: 'heuristic', confidence: 1.0, tokensInput: 0, tokensOutput: 0, model: null };
+
   const entities: ExtractedEntity[] = [];
   const edges: ExtractedEdge[] = [];
 
@@ -269,6 +273,8 @@ export function extractStructuredTask(data: StructuredTask): ExtractionResult {
 }
 
 export function extractStructuredEmail(data: StructuredEmail): ExtractionResult {
+  if (!data?.subject && !data?.from) return { facts: [], entities: [], edges: [], tier: 'heuristic', confidence: 1.0, tokensInput: 0, tokensOutput: 0, model: null };
+
   const entities: ExtractedEntity[] = [];
   const edges: ExtractedEdge[] = [];
 
@@ -340,6 +346,8 @@ export function extractStructuredEmail(data: StructuredEmail): ExtractionResult
 }
 
 export function extractStructuredVault(data: StructuredVault): ExtractionResult {
+  if (!data?.title) return { facts: [], entities: [], edges: [], tier: 'heuristic', confidence: 1.0, tokensInput: 0, tokensOutput: 0, model: null };
+
   const entities: ExtractedEntity[] = [];
   const edges: ExtractedEdge[] = [];