SDOF Knowledge Base

/** * Embedding Service * * This service is responsible for generating vector embeddings from text content * using OpenAI's embedding API with caching for performance optimization. */ import { OpenAI } from 'openai'; import NodeCache from 'node-cache'; import crypto from 'crypto'; import dotenv from 'dotenv'; dotenv.config(); // Load environment variables that don't change during runtime const OPENAI_API_KEY = process.env.OPENAI_API_KEY; const EMBEDDING_CACHE_TTL = parseInt(process.env.EMBEDDING_CACHE_TTL || '3600'); // Default: 1 hour // Initialize OpenAI client const openai = OPENAI_API_KEY ? new OpenAI({ apiKey: OPENAI_API_KEY }) : null; // Initialize cache for embeddings const embeddingCache = new NodeCache({ stdTTL: EMBEDDING_CACHE_TTL, // Time to live in seconds checkperiod: 120, // Check for expired keys every 2 minutes }); /** * Embedding Service class for generating and managing vector embeddings */ export class EmbeddingService { private static instance: EmbeddingService; private readonly openaiModel = process.env.EMBEDDING_MODEL || 'text-embedding-3-large'; /** * Private constructor for singleton pattern */ private constructor() { const embeddingService = process.env.EMBEDDING_SERVICE || 'openai'; // Validate that only supported services are configured if (embeddingService === 'claude') { throw new Error('Claude embeddings are not supported. Anthropic/Claude does not provide an embeddings API. Please use "openai" as the EMBEDDING_SERVICE.'); } if (embeddingService !== 'openai') { console.warn(`Unsupported embedding service: ${embeddingService}. Falling back to OpenAI.`); } console.log(`Embedding service initialized with: openai`); } /** * Get singleton instance */ public static getInstance(): EmbeddingService { if (!EmbeddingService.instance) { EmbeddingService.instance = new EmbeddingService(); } return EmbeddingService.instance; } /** * Generate a hash key for caching */ private generateCacheKey(text: string): string { return crypto.createHash('md5').update(text).digest('hex'); } /** * Generate embedding vector from text using OpenAI */ private async generateOpenAIEmbedding(text: string): Promise<number[]> { if (!openai) { throw new Error('OpenAI API key not configured'); } const response = await openai.embeddings.create({ model: this.openaiModel, input: text, }); return response.data[0].embedding; } /** * Generate embedding vector from text using OpenAI */ public async generateEmbedding(text: string): Promise<number[]> { try { // Check cache first const cacheKey = this.generateCacheKey(text); const cachedEmbedding = embeddingCache.get<number[]>(cacheKey); if (cachedEmbedding) { console.log('Using cached embedding'); return cachedEmbedding; } // Validate embedding service configuration const embeddingService = process.env.EMBEDDING_SERVICE || 'openai'; if (embeddingService === 'claude') { throw new Error('Claude embeddings are not supported. Anthropic/Claude does not provide an embeddings API. Please use "openai" as the EMBEDDING_SERVICE.'); } console.log(`Generating new embedding using OpenAI`); // Generate embedding using OpenAI (only supported service) const embedding = await this.generateOpenAIEmbedding(text); // Cache the result embeddingCache.set(cacheKey, embedding); return embedding; } catch (error) { console.error('Error generating embedding:', error); throw new Error(`Failed to generate embedding: ${error instanceof Error ? error.message : String(error)}`); } } /** * Generate embedding for a knowledge entry by combining title and content */ public async generateEntryEmbedding(title: string, content: string, tags: string[] = []): Promise<number[]> { // Combine title, content, and tags for a more comprehensive embedding const combinedText = `${title}\n\n${content}\n\n${tags.join(', ')}`; return this.generateEmbedding(combinedText); } /** * Get the dimensions of the OpenAI embedding model */ public getDimensions(): number { // OpenAI text-embedding-3-large has 3072 dimensions // OpenAI text-embedding-3-small has 1536 dimensions const model = process.env.EMBEDDING_MODEL || 'text-embedding-3-large'; return model.includes('small') ? 1536 : 3072; } /** * Clear the embedding cache */ public clearCache(): void { embeddingCache.flushAll(); console.log('Embedding cache cleared'); } } // Export singleton instance export default EmbeddingService.getInstance();