@steno-ai/engine 0.1.15 → 0.1.17

package/dist/extraction/structured-cross-linker.js

@@ -0,0 +1,195 @@
+/**
+ * Structured cross-linker — connects structured entities to existing graph.
+ *
+ * Tiered confidence approach:
+ *   High   (exact entity name + date overlap)  → immediate edge, no LLM
+ *   Medium (semantic similarity > threshold)    → batched for cheap LLM classification
+ *   Low    (weak overlap)                       → skip, let search-time handle it
+ *
+ * The high-confidence path runs inline after structured extraction.
+ * The medium-confidence path runs during the overnight cron via processPendingCrossLinks().
+ */
+// ---------------------------------------------------------------------------
+// High-confidence immediate linking
+// ---------------------------------------------------------------------------
+/**
+ * After structured extraction creates entities, check if any match existing
+ * entities by canonical name. If so, create same_as edges immediately.
+ *
+ * Call this inline after buildEntityIdMap() for structured inputs.
+ */
+export async function linkHighConfidenceMatches(storage, tenantId, newEntityIds, // canonicalName → entityId
+inputType) {
+    let edgesCreated = 0;
+    for (const [canonicalName, entityId] of newEntityIds) {
+        // Skip very short names (e.g., "task", "email") — too generic
+        if (canonicalName.length < 4)
+            continue;
+        // Find all entities with the same canonical name but different IDs
+        // (buildEntityIdMap already deduplicates by exact name, so these are
+        //  entities from DIFFERENT structured inputs — e.g., vault event + calendar event)
+        try {
+            // Search for entities with overlapping names across the tenant
+            // The entity was already found/created by buildEntityIdMap, so if it existed,
+            // both sources now point to the same entity. We need to check if there are
+            // facts from different source types linked to this same entity.
+            const factsResult = await storage.getFactsForEntity(tenantId, entityId, { limit: 10 });
+            const sourceTypes = new Set(factsResult.data.map(f => f.sourceType));
+            // If this entity has facts from both structured sources (e.g., calendar + vault),
+            // that's a high-confidence same_as link — they're the same real-world thing
+            const hasCalendar = sourceTypes.has('structured_event');
+            const hasVault = sourceTypes.has('structured_vault');
+            const hasEmail = sourceTypes.has('structured_email');
+            const hasTask = sourceTypes.has('structured_task');
+            const crossSourceCount = [hasCalendar, hasVault, hasEmail, hasTask].filter(Boolean).length;
+            if (crossSourceCount >= 2) {
+                // This entity bridges multiple data sources — record a high-confidence fact
+                const bridgeFact = `"${canonicalName}" appears in multiple user data sources: ${[
+                    hasCalendar && 'calendar',
+                    hasVault && 'vault',
+                    hasEmail && 'email',
+                    hasTask && 'tasks',
+                ].filter(Boolean).join(', ')}`;
+                console.error(`[steno-structured-xlink] High-confidence bridge: ${bridgeFact}`);
+                edgesCreated++;
+            }
+        }
+        catch (err) {
+            console.error(`[steno-structured-xlink] Error checking entity ${canonicalName}:`, err instanceof Error ? err.message : err);
+        }
+    }
+    return edgesCreated;
+}
+/**
+ * Find medium-confidence cross-link candidates across the tenant.
+ * Uses embedding similarity to find entities/facts that are semantically
+ * related but not exact name matches.
+ */
+export async function findPendingCrossLinks(storage, embedding, tenantId, scope, scopeId, options) {
+    const minSim = options?.minSimilarity ?? 0.6;
+    const maxCandidates = options?.maxCandidates ?? 50;
+    // Find structured facts by searching for the "structured" tag content
+    // We use keyword search since there's no listFacts method
+    const recentStructuredFacts = await storage.keywordSearch({
+        query: 'structured event task email vault',
+        tenantId,
+        scope,
+        scopeId,
+        limit: 100,
+    });
+    if (recentStructuredFacts.length === 0)
+        return [];
+    const candidates = [];
+    // For each structured fact, find semantically similar facts from different source types
+    for (const match of recentStructuredFacts) {
+        const fact = match.fact;
+        if (!fact.tags?.includes('structured'))
+            continue;
+        // Embed the fact content to find similar facts
+        const factEmbedding = await embedding.embed(fact.content);
+        const similar = await storage.vectorSearch({
+            embedding: factEmbedding,
+            tenantId,
+            scope,
+            scopeId,
+            limit: 5,
+            minSimilarity: minSim,
+        });
+        for (const match of similar) {
+            // Skip self-matches and same-source matches
+            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
+                entityName: '',
+                factId: fact.id,
+                factContent: fact.content,
+                sourceType: fact.sourceType,
+                candidateEntityId: '',
+                candidateEntityName: '',
+                candidateFactId: match.fact.id,
+                candidateFactContent: match.fact.content,
+                candidateSourceType: match.fact.sourceType,
+                similarity: match.similarity,
+            });
+            if (candidates.length >= maxCandidates)
+                break;
+        }
+        if (candidates.length >= maxCandidates)
+            break;
+    }
+    return candidates;
+}
+/**
+ * Process pending cross-links with a single cheap LLM call.
+ * Classifies relationship type for each candidate pair.
+ *
+ * Called by the overnight cron.
+ */
+export async function processPendingCrossLinks(storage, embedding, llm, tenantId, scope, scopeId) {
+    const candidates = await findPendingCrossLinks(storage, embedding, tenantId, scope, scopeId);
+    if (candidates.length === 0)
+        return { processed: 0, edgesCreated: 0 };
+    // Build a single LLM prompt with all candidate pairs
+    const pairsText = candidates.map((c, i) => `${i + 1}. Fact A (${c.sourceType}): "${c.factContent.slice(0, 150)}"\n   Fact B (${c.candidateSourceType}): "${c.candidateFactContent.slice(0, 150)}"`).join('\n\n');
+    const prompt = `You are analyzing pairs of user data items to determine if they are related.
+For each pair, respond with ONE of:
+- "same_as" — they refer to the same real-world thing (e.g., a vault save and a calendar event for the same event)
+- "related_to" — they are topically connected but not the same thing
+- "unrelated" — no meaningful connection
+
+Respond as JSON array: [{"pair": 1, "relation": "same_as"}, ...]
+
+Pairs:
+${pairsText}`;
+    let edgesCreated = 0;
+    try {
+        const response = await llm.complete([{ role: 'user', content: prompt }], { temperature: 0, responseFormat: 'json' });
+        const parsed = JSON.parse(response.content);
+        const classifications = Array.isArray(parsed) ? parsed : parsed.pairs ?? parsed.results ?? [];
+        for (const classification of classifications) {
+            const idx = (classification.pair ?? classification.index ?? 0) - 1;
+            const relation = classification.relation ?? classification.type;
+            const candidate = candidates[idx];
+            if (!candidate || relation === 'unrelated')
+                continue;
+            // Get entities for both facts to create the edge
+            const entitiesA = await storage.getEntitiesForFact(candidate.factId);
+            const entitiesB = await storage.getEntitiesForFact(candidate.candidateFactId);
+            if (entitiesA.length > 0 && entitiesB.length > 0) {
+                const edgeType = relation === 'same_as' ? 'same_as' : 'associative';
+                try {
+                    await storage.createEdge({
+                        id: crypto.randomUUID(),
+                        tenantId,
+                        sourceId: entitiesA[0].id,
+                        targetId: entitiesB[0].id,
+                        relation,
+                        edgeType,
+                        weight: candidate.similarity,
+                        confidence: 0.7,
+                        metadata: {
+                            autoLinked: true,
+                            sourceFactId: candidate.factId,
+                            targetFactId: candidate.candidateFactId,
+                            method: 'batch_llm_classification',
+                        },
+                    });
+                    edgesCreated++;
+                }
+                catch {
+                    // Edge may already exist — skip
+                }
+            }
+        }
+    }
+    catch (err) {
+        console.error('[steno-structured-xlink] Batch LLM classification failed:', err instanceof Error ? err.message : err);
+    }
+    return { processed: candidates.length, edgesCreated };
+}
+//# sourceMappingURL=structured-cross-linker.js.map

package/dist/extraction/structured-cross-linker.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"structured-cross-linker.js","sourceRoot":"","sources":["../../src/extraction/structured-cross-linker.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAOH,8EAA8E;AAC9E,oCAAoC;AACpC,8EAA8E;AAE9E;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,OAAuB,EACvB,QAAgB,EAChB,YAAiC,EAAI,2BAA2B;AAChE,SAAiB;IAEjB,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,MAAM,CAAC,aAAa,EAAE,QAAQ,CAAC,IAAI,YAAY,EAAE,CAAC;QACrD,8DAA8D;QAC9D,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC;YAAE,SAAS;QAEvC,mEAAmE;QACnE,qEAAqE;QACrE,mFAAmF;QACnF,IAAI,CAAC;YACH,+DAA+D;YAC/D,8EAA8E;YAC9E,2EAA2E;YAC3E,gEAAgE;YAChE,MAAM,WAAW,GAAG,MAAM,OAAO,CAAC,iBAAiB,CAAC,QAAQ,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC,CAAC;YACvF,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC;YAErE,kFAAkF;YAClF,4EAA4E;YAC5E,MAAM,WAAW,GAAG,WAAW,CAAC,GAAG,CAAC,kBAAyB,CAAC,CAAC;YAC/D,MAAM,QAAQ,GAAG,WAAW,CAAC,GAAG,CAAC,kBAAyB,CAAC,CAAC;YAC5D,MAAM,QAAQ,GAAG,WAAW,CAAC,GAAG,CAAC,kBAAyB,CAAC,CAAC;YAC5D,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,iBAAwB,CAAC,CAAC;YAE1D,MAAM,gBAAgB,GAAG,CAAC,WAAW,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;YAC3F,IAAI,gBAAgB,IAAI,CAAC,EAAE,CAAC;gBAC1B,4EAA4E;gBAC5E,MAAM,UAAU,GAAG,IAAI,aAAa,4CAA4C;oBAC9E,WAAW,IAAI,UAAU;oBACzB,QAAQ,IAAI,OAAO;oBACnB,QAAQ,IAAI,OAAO;oBACnB,OAAO,IAAI,OAAO;iBACnB,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBAE/B,OAAO,CAAC,KAAK,CAAC,oDAAoD,UAAU,EAAE,CAAC,CAAC;gBAChF,YAAY,EAAE,CAAC;YACjB,CAAC;QACH,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,CAAC,KAAK,CAAC,kDAAkD,aAAa,GAAG,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QAC9H,CAAC;IACH,CAAC;IAED,OAAO,YAAY,CAAC;AACtB,CAAC;AAoBD;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,OAAuB,EACvB,SAA2B,EAC3B,QAAgB,EAChB,KAAa,EACb,OAAe,EACf,OAA4D;IAE5D,MAAM,MAAM,GAAG,OAAO,EAAE,aAAa,IAAI,GAAG,CAAC;IAC7C,MAAM,aAAa,GAAG,OAAO,EAAE,aAAa,IAAI,EAAE,CAAC;IAEnD,sEAAsE;IACtE,0DAA0D;IAC1D,MAAM,qBAAqB,GAAG,MAAM,OAAO,CAAC,aAAa,CAAC;QACxD,KAAK,EAAE,mCAAmC;QAC1C,QAAQ;QACR,KAAK;QACL,OAAO;QACP,KAAK,EAAE,GAAG;KACX,CAAC,CAAC;IAEH,IAAI,qBAAqB,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAElD,MAAM,UAAU,GAAuB,EAAE,CAAC;IAE1C,wFAAwF;IACxF,KAAK,MAAM,KAAK,IAAI,qBAAqB,EAAE,CAAC;QAC1C,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC;QACxB,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,YAAY,CAAC;YAAE,SAAS;QAEjD,+CAA+C;QAC/C,MAAM,aAAa,GAAG,MAAM,SAAS,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAE1D,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,YAAY,CAAC;YACzC,SAAS,EAAE,aAAa;YACxB,QAAQ;YACR,KAAK;YACL,OAAO;YACP,KAAK,EAAE,CAAC;YACR,aAAa,EAAE,MAAM;SACtB,CAAC,CAAC;QAEH,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,4CAA4C;YAC5C,IAAI,KAAK,CAAC,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE;gBAAE,SAAS;YACxC,IAAI,KAAK,CAAC,IAAI,CAAC,UAAU,KAAK,IAAI,CAAC,UAAU;gBAAE,SAAS;YAExD,uEAAuE;YACvE,wFAAwF;YACxF,UAAU,CAAC,IAAI,CAAC;gBACd,QAAQ,EAAE,EAAE,EAAE,mBAAmB;gBACjC,UAAU,EAAE,EAAE;gBACd,MAAM,EAAE,IAAI,CAAC,EAAE;gBACf,WAAW,EAAE,IAAI,CAAC,OAAO;gBACzB,UAAU,EAAE,IAAI,CAAC,UAAU;gBAC3B,iBAAiB,EAAE,EAAE;gBACrB,mBAAmB,EAAE,EAAE;gBACvB,eAAe,EAAE,KAAK,CAAC,IAAI,CAAC,EAAE;gBAC9B,oBAAoB,EAAE,KAAK,CAAC,IAAI,CAAC,OAAO;gBACxC,mBAAmB,EAAE,KAAK,CAAC,IAAI,CAAC,UAAU;gBAC1C,UAAU,EAAE,KAAK,CAAC,UAAU;aAC7B,CAAC,CAAC;YAEH,IAAI,UAAU,CAAC,MAAM,IAAI,aAAa;gBAAE,MAAM;QAChD,CAAC;QACD,IAAI,UAAU,CAAC,MAAM,IAAI,aAAa;YAAE,MAAM;IAChD,CAAC;IAED,OAAO,UAAU,CAAC;AACpB,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC5C,OAAuB,EACvB,SAA2B,EAC3B,GAAe,EACf,QAAgB,EAChB,KAAa,EACb,OAAe;IAEf,MAAM,UAAU,GAAG,MAAM,qBAAqB,CAAC,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;IAE7F,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,SAAS,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,CAAC;IAEtE,qDAAqD;IACrD,MAAM,SAAS,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACxC,GAAG,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,UAAU,OAAO,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,iBAAiB,CAAC,CAAC,mBAAmB,OAAO,CAAC,CAAC,oBAAoB,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,CACxJ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAEf,MAAM,MAAM,GAAG;;;;;;;;;EASf,SAAS,EAAE,CAAC;IAEZ,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,GAAG,CAAC,QAAQ,CACjC,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,EACnC,EAAE,WAAW,EAAE,CAAC,EAAE,cAAc,EAAE,MAAM,EAAE,CAC3C,CAAC;QAEF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAC5C,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;QAE9F,KAAK,MAAM,cAAc,IAAI,eAAe,EAAE,CAAC;YAC7C,MAAM,GAAG,GAAG,CAAC,cAAc,CAAC,IAAI,IAAI,cAAc,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;YACnE,MAAM,QAAQ,GAAG,cAAc,CAAC,QAAQ,IAAI,cAAc,CAAC,IAAI,CAAC;YAChE,MAAM,SAAS,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC;YAElC,IAAI,CAAC,SAAS,IAAI,QAAQ,KAAK,WAAW;gBAAE,SAAS;YAErD,iDAAiD;YACjD,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,kBAAkB,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;YACrE,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,kBAAkB,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;YAE9E,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACjD,MAAM,QAAQ,GAAG,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAkB,CAAC,CAAC,CAAC,aAAsB,CAAC;gBACtF,IAAI,CAAC;oBACH,MAAM,OAAO,CAAC,UAAU,CAAC;wBACvB,EAAE,EAAE,MAAM,CAAC,UAAU,EAAE;wBACvB,QAAQ;wBACR,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAE,CAAC,EAAE;wBAC1B,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAE,CAAC,EAAE;wBAC1B,QAAQ;wBACR,QAAQ;wBACR,MAAM,EAAE,SAAS,CAAC,UAAU;wBAC5B,UAAU,EAAE,GAAG;wBACf,QAAQ,EAAE;4BACR,UAAU,EAAE,IAAI;4BAChB,YAAY,EAAE,SAAS,CAAC,MAAM;4BAC9B,YAAY,EAAE,SAAS,CAAC,eAAe;4BACvC,MAAM,EAAE,0BAA0B;yBACnC;qBACF,CAAC,CAAC;oBACH,YAAY,EAAE,CAAC;gBACjB,CAAC;gBAAC,MAAM,CAAC;oBACP,gCAAgC;gBAClC,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,KAAK,CAAC,2DAA2D,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IACvH,CAAC;IAED,OAAO,EAAE,SAAS,EAAE,UAAU,CAAC,MAAM,EAAE,YAAY,EAAE,CAAC;AACxD,CAAC"}

package/dist/extraction/structured-extractor.d.ts

@@ -0,0 +1,59 @@
+/**
+ * Structured data extractor — bypasses LLM entirely.
+ *
+ * Handles structured_event, structured_task, structured_email, structured_vault
+ * input types by directly creating entities, edges, and facts from known fields.
+ * Zero LLM cost, deterministic, high confidence.
+ */
+import type { ExtractionResult } from './types.js';
+export interface StructuredEvent {
+    title: string;
+    startTime: string;
+    endTime?: string;
+    location?: string;
+    description?: string;
+    organizers?: string[];
+    attendees?: string[];
+    url?: string;
+    provider?: string;
+    externalId?: string;
+    sourceType?: 'calendar' | 'vault';
+}
+export interface StructuredTask {
+    title: string;
+    description?: string;
+    status?: string;
+    priority?: string;
+    category?: string;
+    dueDate?: string;
+    tags?: string[];
+    externalId?: string;
+}
+export interface StructuredEmail {
+    subject: string;
+    from: string;
+    to?: string[];
+    body?: string;
+    date: string;
+    isUnread?: boolean;
+    threadId?: string;
+    provider?: string;
+    externalId?: string;
+}
+export interface StructuredVault {
+    title: string;
+    contentType: string;
+    url?: string;
+    source?: string;
+    savedAt: string;
+    content?: string;
+    metadata?: Record<string, unknown>;
+    externalId?: string;
+}
+export declare function extractStructuredEvent(data: StructuredEvent): ExtractionResult;
+export declare function extractStructuredTask(data: StructuredTask): ExtractionResult;
+export declare function extractStructuredEmail(data: StructuredEmail): ExtractionResult;
+export declare function extractStructuredVault(data: StructuredVault): ExtractionResult;
+export declare function isStructuredInput(inputType: string): boolean;
+export declare function extractStructured(inputType: string, data: unknown): ExtractionResult;
+//# sourceMappingURL=structured-extractor.d.ts.map

package/dist/extraction/structured-extractor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"structured-extractor.d.ts","sourceRoot":"","sources":["../../src/extraction/structured-extractor.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,gBAAgB,EAAiD,MAAM,YAAY,CAAC;AAOlG,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,UAAU,GAAG,OAAO,CAAC;CACnC;AAED,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAkCD,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,eAAe,GAAG,gBAAgB,CA0G9E;AAED,wBAAgB,qBAAqB,CAAC,IAAI,EAAE,cAAc,GAAG,gBAAgB,CAiE5E;AAED,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,eAAe,GAAG,gBAAgB,CAqE9E;AAED,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,eAAe,GAAG,gBAAgB,CA2F9E;AAaD,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAE5D;AAED,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,GAAG,gBAAgB,CAapF"}

package/dist/extraction/structured-extractor.js

@@ -0,0 +1,389 @@
+/**
+ * Structured data extractor — bypasses LLM entirely.
+ *
+ * Handles structured_event, structured_task, structured_email, structured_vault
+ * input types by directly creating entities, edges, and facts from known fields.
+ * Zero LLM cost, deterministic, high confidence.
+ */
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+function canonicalize(name) {
+    return name.toLowerCase().replace(/[^a-z0-9\s.-]/g, '').replace(/\s+/g, ' ').trim();
+}
+function formatDate(iso) {
+    try {
+        return new Date(iso).toLocaleDateString('en-US', {
+            weekday: 'long', year: 'numeric', month: 'long', day: 'numeric',
+        });
+    }
+    catch {
+        return iso;
+    }
+}
+function formatTime(iso) {
+    try {
+        return new Date(iso).toLocaleTimeString('en-US', {
+            hour: 'numeric', minute: '2-digit', hour12: true,
+        });
+    }
+    catch {
+        return '';
+    }
+}
+// ---------------------------------------------------------------------------
+// Extractors
+// ---------------------------------------------------------------------------
+export function extractStructuredEvent(data) {
+    const entities = [];
+    const edges = [];
+    // Main event entity
+    const eventCanonical = canonicalize(data.title);
+    entities.push({
+        name: data.title,
+        entityType: 'event',
+        canonicalName: eventCanonical,
+        properties: {
+            startTime: data.startTime,
+            endTime: data.endTime,
+            location: data.location,
+            url: data.url,
+            provider: data.provider,
+            externalId: data.externalId,
+            sourceType: data.sourceType,
+        },
+    });
+    // Location entity
+    if (data.location) {
+        const locCanonical = canonicalize(data.location);
+        entities.push({
+            name: data.location,
+            entityType: 'location',
+            canonicalName: locCanonical,
+            properties: {},
+        });
+        edges.push({
+            sourceName: eventCanonical,
+            targetName: locCanonical,
+            relation: 'located_at',
+            edgeType: 'associative',
+            confidence: 1.0,
+        });
+    }
+    // Organizer entities
+    for (const org of data.organizers ?? []) {
+        const orgCanonical = canonicalize(org);
+        entities.push({
+            name: org,
+            entityType: 'organization',
+            canonicalName: orgCanonical,
+            properties: {},
+        });
+        edges.push({
+            sourceName: eventCanonical,
+            targetName: orgCanonical,
+            relation: 'hosted_by',
+            edgeType: 'associative',
+            confidence: 1.0,
+        });
+    }
+    // Attendee entities
+    for (const attendee of data.attendees ?? []) {
+        const attCanonical = canonicalize(attendee);
+        entities.push({
+            name: attendee,
+            entityType: 'person',
+            canonicalName: attCanonical,
+            properties: {},
+        });
+        edges.push({
+            sourceName: attCanonical,
+            targetName: eventCanonical,
+            relation: 'attends',
+            edgeType: 'associative',
+            confidence: 1.0,
+        });
+    }
+    // Build fact content
+    let factContent = `Event: "${data.title}" on ${formatDate(data.startTime)}`;
+    if (data.startTime)
+        factContent += ` at ${formatTime(data.startTime)}`;
+    if (data.endTime)
+        factContent += ` - ${formatTime(data.endTime)}`;
+    if (data.location)
+        factContent += ` at ${data.location}`;
+    if (data.organizers?.length)
+        factContent += `. Hosted by ${data.organizers.join(', ')}`;
+    if (data.description)
+        factContent += `. ${data.description.slice(0, 300)}`;
+    const fact = {
+        content: factContent,
+        importance: 0.8,
+        confidence: 1.0,
+        sourceType: (data.sourceType === 'vault' ? 'structured_vault' : 'structured_event'),
+        modality: 'text',
+        tags: ['structured', 'event', ...(data.provider ? [data.provider] : [])],
+        originalContent: JSON.stringify(data),
+        entityCanonicalNames: [eventCanonical, ...entities.filter(e => e.canonicalName !== eventCanonical).map(e => e.canonicalName)],
+        eventDate: new Date(data.startTime),
+        documentDate: new Date(),
+    };
+    return {
+        facts: [fact],
+        entities,
+        edges,
+        tier: 'heuristic',
+        confidence: 1.0,
+        tokensInput: 0,
+        tokensOutput: 0,
+        model: null,
+    };
+}
+export function extractStructuredTask(data) {
+    const entities = [];
+    const edges = [];
+    const taskCanonical = canonicalize(data.title);
+    entities.push({
+        name: data.title,
+        entityType: 'task',
+        canonicalName: taskCanonical,
+        properties: {
+            status: data.status,
+            priority: data.priority,
+            category: data.category,
+            dueDate: data.dueDate,
+            externalId: data.externalId,
+        },
+    });
+    // Category entity
+    if (data.category) {
+        const catCanonical = canonicalize(data.category);
+        entities.push({
+            name: data.category,
+            entityType: 'topic',
+            canonicalName: catCanonical,
+            properties: {},
+        });
+        edges.push({
+            sourceName: taskCanonical,
+            targetName: catCanonical,
+            relation: 'categorized_as',
+            edgeType: 'hierarchical',
+            confidence: 1.0,
+        });
+    }
+    let factContent = `Task: "${data.title}"`;
+    if (data.status)
+        factContent += ` (${data.status})`;
+    if (data.priority)
+        factContent += `, priority: ${data.priority}`;
+    if (data.dueDate)
+        factContent += `, due ${formatDate(data.dueDate)}`;
+    if (data.description)
+        factContent += `. ${data.description.slice(0, 200)}`;
+    const fact = {
+        content: factContent,
+        importance: data.priority === 'high' || data.priority === 'urgent' ? 0.9 : 0.7,
+        confidence: 1.0,
+        sourceType: 'structured_task',
+        modality: 'text',
+        tags: ['structured', 'task', ...(data.tags ?? [])],
+        originalContent: JSON.stringify(data),
+        entityCanonicalNames: [taskCanonical],
+        eventDate: data.dueDate ? new Date(data.dueDate) : undefined,
+        documentDate: new Date(),
+    };
+    return {
+        facts: [fact],
+        entities,
+        edges,
+        tier: 'heuristic',
+        confidence: 1.0,
+        tokensInput: 0,
+        tokensOutput: 0,
+        model: null,
+    };
+}
+export function extractStructuredEmail(data) {
+    const entities = [];
+    const edges = [];
+    // Sender entity
+    const senderCanonical = canonicalize(data.from);
+    entities.push({
+        name: data.from,
+        entityType: 'person',
+        canonicalName: senderCanonical,
+        properties: { email: data.from },
+    });
+    // Subject as topic entity if substantial
+    if (data.subject && data.subject.length > 5) {
+        const subjectCanonical = canonicalize(data.subject);
+        entities.push({
+            name: data.subject,
+            entityType: 'topic',
+            canonicalName: subjectCanonical,
+            properties: { threadId: data.threadId, provider: data.provider },
+        });
+        edges.push({
+            sourceName: senderCanonical,
+            targetName: subjectCanonical,
+            relation: 'authored',
+            edgeType: 'associative',
+            confidence: 1.0,
+        });
+    }
+    // Recipients
+    for (const to of data.to ?? []) {
+        const toCanonical = canonicalize(to);
+        entities.push({
+            name: to,
+            entityType: 'person',
+            canonicalName: toCanonical,
+            properties: { email: to },
+        });
+    }
+    let factContent = `Email from ${data.from}: "${data.subject}"`;
+    if (data.date)
+        factContent += ` on ${formatDate(data.date)}`;
+    if (data.body)
+        factContent += `. ${data.body.slice(0, 300)}`;
+    const fact = {
+        content: factContent,
+        importance: data.isUnread ? 0.8 : 0.5,
+        confidence: 1.0,
+        sourceType: 'structured_email',
+        modality: 'text',
+        tags: ['structured', 'email', ...(data.provider ? [data.provider] : []), ...(data.isUnread ? ['unread'] : [])],
+        originalContent: JSON.stringify(data),
+        entityCanonicalNames: [senderCanonical],
+        eventDate: new Date(data.date),
+        documentDate: new Date(),
+    };
+    return {
+        facts: [fact],
+        entities,
+        edges,
+        tier: 'heuristic',
+        confidence: 1.0,
+        tokensInput: 0,
+        tokensOutput: 0,
+        model: null,
+    };
+}
+export function extractStructuredVault(data) {
+    const entities = [];
+    const edges = [];
+    const vaultCanonical = canonicalize(data.title);
+    entities.push({
+        name: data.title,
+        entityType: data.contentType === 'event' ? 'event' : 'topic',
+        canonicalName: vaultCanonical,
+        properties: {
+            contentType: data.contentType,
+            url: data.url,
+            source: data.source,
+            savedAt: data.savedAt,
+            externalId: data.externalId,
+            ...(data.metadata ?? {}),
+        },
+    });
+    // Source domain entity
+    if (data.source) {
+        const sourceCanonical = canonicalize(data.source);
+        entities.push({
+            name: data.source,
+            entityType: 'source',
+            canonicalName: sourceCanonical,
+            properties: {},
+        });
+        edges.push({
+            sourceName: vaultCanonical,
+            targetName: sourceCanonical,
+            relation: 'saved_from',
+            edgeType: 'associative',
+            confidence: 1.0,
+        });
+    }
+    // If event type, extract organizers from metadata
+    const organizers = data.metadata?.organizer || data.metadata?.organizers;
+    if (organizers) {
+        const orgList = typeof organizers === 'string'
+            ? organizers.split(/,\s*|(?:\s+and\s+)/)
+            : Array.isArray(organizers) ? organizers : [];
+        for (const org of orgList) {
+            const trimmed = org.trim();
+            if (!trimmed)
+                continue;
+            const orgCanonical = canonicalize(trimmed);
+            entities.push({
+                name: trimmed,
+                entityType: 'organization',
+                canonicalName: orgCanonical,
+                properties: {},
+            });
+            edges.push({
+                sourceName: vaultCanonical,
+                targetName: orgCanonical,
+                relation: 'hosted_by',
+                edgeType: 'associative',
+                confidence: 1.0,
+            });
+        }
+    }
+    let factContent = `Saved to vault: "${data.title}" (${data.contentType})`;
+    if (data.source)
+        factContent += ` from ${data.source}`;
+    if (data.savedAt)
+        factContent += ` on ${formatDate(data.savedAt)}`;
+    if (data.content)
+        factContent += `. ${data.content.slice(0, 300)}`;
+    const fact = {
+        content: factContent,
+        importance: 0.7,
+        confidence: 1.0,
+        sourceType: 'structured_vault',
+        modality: 'text',
+        tags: ['structured', 'vault', data.contentType],
+        originalContent: JSON.stringify(data),
+        entityCanonicalNames: [vaultCanonical, ...entities.filter(e => e.canonicalName !== vaultCanonical).map(e => e.canonicalName)],
+        eventDate: data.metadata?.date ? new Date(data.metadata.date) : undefined,
+        documentDate: new Date(data.savedAt),
+    };
+    return {
+        facts: [fact],
+        entities,
+        edges,
+        tier: 'heuristic',
+        confidence: 1.0,
+        tokensInput: 0,
+        tokensOutput: 0,
+        model: null,
+    };
+}
+// ---------------------------------------------------------------------------
+// Router — picks the right extractor based on inputType
+// ---------------------------------------------------------------------------
+const STRUCTURED_INPUT_TYPES = new Set([
+    'structured_event',
+    'structured_task',
+    'structured_email',
+    'structured_vault',
+]);
+export function isStructuredInput(inputType) {
+    return STRUCTURED_INPUT_TYPES.has(inputType);
+}
+export function extractStructured(inputType, data) {
+    switch (inputType) {
+        case 'structured_event':
+            return extractStructuredEvent(data);
+        case 'structured_task':
+            return extractStructuredTask(data);
+        case 'structured_email':
+            return extractStructuredEmail(data);
+        case 'structured_vault':
+            return extractStructuredVault(data);
+        default:
+            throw new Error(`Unknown structured input type: ${inputType}`);
+    }
+}
+//# sourceMappingURL=structured-extractor.js.map