AgentDB Performance Optimization

Enable binary quantization in the AgentDB adapter configuration to achieve a 32x memory reduction, turning 3GB for 1M vectors into 96MB. Scalar quantization offers a 4x reduction with higher accuracy (98-99%) if binary's 95-98% accuracy is insufficient.

AgentDB builds HNSW indices automatically, giving O(log n) search complexity. Tune hnswM (8 for small, 16 for medium, 32 for large datasets) and hnswEfSearch (50 for speed, 200 for recall) to balance latency against accuracy.

Binary quantization gives 32x memory reduction with 95-98% accuracy, scalar gives 4x reduction with 98-99% accuracy, and product quantization gives 8-16x reduction with 93-97% accuracy. Choose based on dataset size and accuracy requirements.

Yes, quantization trades accuracy for memory and speed. Binary loses roughly 2-5% accuracy, scalar loses 1-2%, and product loses 3-7%. Use quantizationType 'none' for full float32 precision when accuracy is critical.

Slow searches usually come from missing cache or overly strict search parameters. Increase cacheSize to 2000, reduce k in retrieval calls, and lower hnswEfSearch to 50-100 for faster queries at slightly reduced recall.

Avoid binary quantization when your application requires near-perfect recall, such as compliance or medical retrieval, since it retains only 95-98% accuracy. Use scalar quantization or full precision instead for those workloads.