agentdb-performance-optimization

Optimizes AgentDB performance with quantization, HNSW indexing, caching, and batch operations for faster search and lower memory usage.

• Reduces memory by 4-32x using quantization
• Speeds up search with HNSW indexing (150x faster)
• Caches frequently used patterns for faster retrieval
• Enables batch operations for 500x faster insertions

• Quantization: binary (32x memory reduction), scalar (4x), product (8-16x)
• HNSW indexing: O(log n) search, 150x faster than linear scan
• Caching: LRU-based pattern cache with <1ms hit performance
• Batch operations: 500x faster than individual inserts
• Memory reduction: 4-32x with minimal accuracy loss
• Search speed: <100µs for vector retrieval, <1ms for pattern retrieval

1. 1.Run benchmarks with `npx agentdb@latest benchmark`
2. 2.Configure quantization type (binary, scalar, product, or none) in adapter config
3. 3.Set HNSW parameters (M, efConstruction, efSearch) based on dataset size
4. 4.Enable in-memory cache with `cacheSize` and monitor hit rate >80%
5. 5.Use batch insert for 100+ vectors to achieve 500x faster insertion