Apple Tools MCP

/** * Real Embedding Cache Tests * * Tests embedding caching with real Xenova/all-MiniLM-L6-v2 model */ import { describe, it, expect, beforeAll } from 'vitest' import { checkDataSources, isProductionIndexReady, embed, embedBatch, getEmbedder } from '../helpers/real-data.js' const sources = checkDataSources() describe('Real Embedding Cache', () => { let embedder beforeAll(async () => { embedder = await getEmbedder() }, 60000) describe('Embedding Generation', () => { it('should generate embeddings using real model', async () => { const text = 'This is a test email about quarterly meeting' const vector = await embed(text) expect(vector).toHaveLength(384) expect(vector[0]).toBeTypeOf('number') }, 30000) it('should generate consistent embeddings for same text', async () => { const text = 'Important project update from the team' const vec1 = await embed(text) const vec2 = await embed(text) // Same text should produce identical vectors expect(vec1.length).toBe(vec2.length) // Check each element is the same (within floating point tolerance) for (let i = 0; i < vec1.length; i++) { expect(Math.abs(vec1[i] - vec2[i])).toBeLessThan(1e-6) } }, 30000) }) describe('Batch Embedding', () => { it('should batch embed multiple texts', async () => { const texts = [ 'Meeting about budget review', 'Dinner plans for Friday', 'Project deadline reminder' ] const vectors = await embedBatch(texts) expect(vectors).toHaveLength(3) for (const vec of vectors) { expect(vec).toHaveLength(384) } }, 30000) it('should produce different vectors for different texts', async () => { const texts = [ 'Weather forecast for tomorrow', 'Quarterly financial report' ] const vectors = await embedBatch(texts) // Calculate cosine similarity let dot = 0, norm1 = 0, norm2 = 0 for (let i = 0; i < 384; i++) { dot += vectors[0][i] * vectors[1][i] norm1 += vectors[0][i] * vectors[0][i] norm2 += vectors[1][i] * vectors[1][i] } const similarity = dot / (Math.sqrt(norm1) * Math.sqrt(norm2)) // Different texts should have lower similarity expect(similarity).toBeLessThan(0.9) }, 30000) }) describe('Cache Performance', () => { it('should be faster on repeated embeddings due to model caching', async () => { const text = 'Cache performance test text for embedding' // First call - model may need warmup const start1 = performance.now() await embed(text) const duration1 = performance.now() - start1 // Second call - should be faster (model cached) const start2 = performance.now() await embed(text) const duration2 = performance.now() - start2 // Second call should be faster or similar (not slower) // Allow some variance due to system load expect(duration2).toBeLessThan(duration1 * 2) console.log(`First embed: ${duration1.toFixed(0)}ms, Second: ${duration2.toFixed(0)}ms`) }, 30000) it('should handle rapid sequential embeddings', async () => { const texts = Array.from({ length: 10 }, (_, i) => `Test text number ${i}`) const start = performance.now() for (const text of texts) { await embed(text) } const duration = performance.now() - start // Should complete in reasonable time (< 30s for 10 texts) expect(duration).toBeLessThan(30000) console.log(`10 sequential embeddings: ${duration.toFixed(0)}ms (${(duration/10).toFixed(0)}ms avg)`) }, 60000) }) describe('Model Pipeline', () => { it('should reuse embedding pipeline', async () => { const embedder1 = await getEmbedder() const embedder2 = await getEmbedder() // Should be same instance (cached) expect(embedder1).toBe(embedder2) }, 30000) it('should support pooling options', async () => { // Direct use of embedder with options const text = 'Test with pooling options' const result = await embedder(text, { pooling: 'mean', normalize: true }) expect(result.data).toHaveLength(384) }, 30000) }) }) describe.skipIf(!sources.mail && !sources.messages && !sources.calendar)( 'Search Result Caching', () => { beforeAll(async () => { // System index must already exist - tests never rebuild it const ready = await isProductionIndexReady() if (!ready) { throw new Error('System index not found. Run "npm run build-index" first.') } }, 180000) it('should benefit from repeated searches', async () => { // This tests that LanceDB/embedding caching works in search const { searchProductionIndex } = await import('../helpers/real-data.js') const query = 'meeting tomorrow' // First search const start1 = performance.now() const results1 = await searchProductionIndex(query, 'emails', 5) const duration1 = performance.now() - start1 // Second search (same query) - should benefit from caching const start2 = performance.now() const results2 = await searchProductionIndex(query, 'emails', 5) const duration2 = performance.now() - start2 // Results should be the same expect(results1.length).toBe(results2.length) console.log(`First search: ${duration1.toFixed(0)}ms, Second: ${duration2.toFixed(0)}ms`) }, 60000) it('should handle different query variations', async () => { const { searchProductionIndex } = await import('../helpers/real-data.js') const queries = [ 'project update', 'schedule meeting', 'budget review', 'team discussion' ] for (const query of queries) { const results = await searchProductionIndex(query, 'emails', 3) expect(Array.isArray(results)).toBe(true) } }, 60000) } )