ThoughtMCP

/** * Embedding Engine * * Core embedding generation engine implementing five-sector HMD system. * Generates specialized embeddings for Episodic, Semantic, Procedural, * Emotional, and Reflective memory sectors with caching and batch support. * * Requirements: 2.1, 2.2, 2.3, 2.4, 2.5 */ import { generateCacheKey } from "./cache"; import type { EmbeddingModel, EmotionState, EmbeddingCache as IEmbeddingCache, MemoryContent, MemorySector, SectorEmbeddings, TemporalContext, } from "./types"; /** * Five-sector embedding generation engine */ export class EmbeddingEngine { private model: EmbeddingModel; private cache: IEmbeddingCache; private inFlightRequests: Map<string, Promise<number[]>>; constructor(model: EmbeddingModel, cache: IEmbeddingCache) { this.model = model; this.cache = cache; this.inFlightRequests = new Map(); } /** * Generate episodic embedding with temporal context markers */ async generateEpisodicEmbedding(content: string, context: TemporalContext): Promise<number[]> { const cacheKey = generateCacheKey("episodic" as MemorySector, content, context); // Check cache first const cached = this.cache.get(cacheKey); if (cached) { return cached; } // Check for in-flight request const inFlight = this.inFlightRequests.get(cacheKey); if (inFlight) { return inFlight; } // Augment content with temporal markers const augmentedContent = this.augmentEpisodicContent(content, context); // Generate embedding with deduplication const promise = this.generateWithDeduplication(cacheKey, augmentedContent); return promise; } /** * Generate semantic embedding (pure factual content) */ async generateSemanticEmbedding(content: string): Promise<number[]> { const cacheKey = generateCacheKey("semantic" as MemorySector, content); // Check cache first const cached = this.cache.get(cacheKey); if (cached) { return cached; } // Check for in-flight request const inFlight = this.inFlightRequests.get(cacheKey); if (inFlight) { return inFlight; } // Generate embedding with deduplication (no augmentation for semantic) const promise = this.generateWithDeduplication(cacheKey, content); return promise; } /** * Generate procedural embedding with process markers */ async generateProceduralEmbedding(content: string): Promise<number[]> { const cacheKey = generateCacheKey("procedural" as MemorySector, content); // Check cache first const cached = this.cache.get(cacheKey); if (cached) { return cached; } // Check for in-flight request const inFlight = this.inFlightRequests.get(cacheKey); if (inFlight) { return inFlight; } // Augment content with process markers const augmentedContent = this.augmentProceduralContent(content); // Generate embedding with deduplication const promise = this.generateWithDeduplication(cacheKey, augmentedContent); return promise; } /** * Generate emotional embedding with emotion state markers */ async generateEmotionalEmbedding(content: string, emotion: EmotionState): Promise<number[]> { const cacheKey = generateCacheKey("emotional" as MemorySector, content, emotion); // Check cache first const cached = this.cache.get(cacheKey); if (cached) { return cached; } // Check for in-flight request const inFlight = this.inFlightRequests.get(cacheKey); if (inFlight) { return inFlight; } // Augment content with emotion markers const augmentedContent = this.augmentEmotionalContent(content, emotion); // Generate embedding with deduplication const promise = this.generateWithDeduplication(cacheKey, augmentedContent); return promise; } /** * Generate reflective embedding with meta-insight markers */ async generateReflectiveEmbedding(content: string, insights: string[]): Promise<number[]> { const cacheKey = generateCacheKey("reflective" as MemorySector, content, insights); // Check cache first const cached = this.cache.get(cacheKey); if (cached) { return cached; } // Check for in-flight request const inFlight = this.inFlightRequests.get(cacheKey); if (inFlight) { return inFlight; } // Augment content with meta-insight markers const augmentedContent = this.augmentReflectiveContent(content, insights); // Generate embedding with deduplication const promise = this.generateWithDeduplication(cacheKey, augmentedContent); return promise; } /** * Generate all five sector embeddings for a memory */ async generateAllSectorEmbeddings(memory: MemoryContent): Promise<SectorEmbeddings> { // Generate all sectors in parallel const [episodic, semantic, procedural, emotional, reflective] = await Promise.all([ this.generateEpisodicEmbedding( memory.text, memory.context ?? { timestamp: new Date(), sessionId: "default", } ), this.generateSemanticEmbedding(memory.text), this.generateProceduralEmbedding(memory.text), this.generateEmotionalEmbedding( memory.text, memory.emotion ?? { valence: 0, arousal: 0.5, dominance: 0, } ), this.generateReflectiveEmbedding(memory.text, memory.insights ?? []), ]); return { episodic, semantic, procedural, emotional, reflective, }; } /** * Generate embeddings for multiple memories in batch */ async batchGenerateEmbeddings(memories: MemoryContent[]): Promise<SectorEmbeddings[]> { // Process all memories in parallel return Promise.all(memories.map((memory) => this.generateAllSectorEmbeddings(memory))); } /** * Get current model dimension */ getModelDimension(): number { return this.model.getDimension(); } /** * Load a different model */ loadModel(model: EmbeddingModel): void { this.model = model; // Clear cache when switching models this.cache.clear(); } /** * Generate embedding with request deduplication * Prevents duplicate in-flight requests for the same content */ private async generateWithDeduplication(cacheKey: string, content: string): Promise<number[]> { // Create promise for this request const promise = (async () => { try { // Generate embedding const embedding = await this.model.generate(content); // Store in cache this.cache.set(cacheKey, embedding); return embedding; } finally { // Remove from in-flight requests when done this.inFlightRequests.delete(cacheKey); } })(); // Store in-flight request this.inFlightRequests.set(cacheKey, promise); return promise; } /** * Augment content with temporal markers for episodic embedding */ private augmentEpisodicContent(content: string, context: TemporalContext): string { const markers: string[] = []; // Add timestamp marker markers.push(`[TIME: ${context.timestamp.toISOString()}]`); // Add session marker markers.push(`[SESSION: ${context.sessionId}]`); // Add sequence marker if available if (context.sequenceNumber !== undefined) { markers.push(`[SEQUENCE: ${context.sequenceNumber}]`); } // Add duration marker if available if (context.duration !== undefined) { markers.push(`[DURATION: ${context.duration}s]`); } // Add location marker if available if (context.location) { markers.push(`[LOCATION: ${context.location}]`); } // Add participants marker if available if (context.participants && context.participants.length > 0) { markers.push(`[PARTICIPANTS: ${context.participants.join(", ")}]`); } return `${markers.join(" ")} ${content}`; } /** * Augment content with process markers for procedural embedding */ private augmentProceduralContent(content: string): string { // Add procedural markers to emphasize process/steps return `[PROCEDURE] [STEPS] [HOW-TO] ${content}`; } /** * Augment content with emotion markers for emotional embedding */ private augmentEmotionalContent(content: string, emotion: EmotionState): string { const markers: string[] = []; // Add valence marker const valenceLabel = emotion.valence > 0 ? "POSITIVE" : emotion.valence < 0 ? "NEGATIVE" : "NEUTRAL"; markers.push(`[VALENCE: ${valenceLabel}]`); // Add arousal marker const arousalLabel = emotion.arousal > 0.7 ? "HIGH" : emotion.arousal > 0.3 ? "MEDIUM" : "LOW"; markers.push(`[AROUSAL: ${arousalLabel}]`); // Add dominance marker const dominanceLabel = emotion.dominance > 0 ? "DOMINANT" : emotion.dominance < 0 ? "SUBMISSIVE" : "NEUTRAL"; markers.push(`[DOMINANCE: ${dominanceLabel}]`); // Add primary emotion if available if (emotion.primaryEmotion) { markers.push(`[EMOTION: ${emotion.primaryEmotion.toUpperCase()}]`); } return `${markers.join(" ")} ${content}`; } /** * Augment content with meta-insight markers for reflective embedding */ private augmentReflectiveContent(content: string, insights: string[]): string { const markers: string[] = ["[REFLECTION]", "[META-COGNITION]"]; // Add insights as markers if available if (insights.length > 0) { markers.push(`[INSIGHTS: ${insights.join("; ")}]`); } return `${markers.join(" ")} ${content}`; } }