AgentDB Performance Optimization

Enable binary quantization by setting quantizationType to 'binary' in createAgentDBAdapter, which reduces memory 32x (e.g., 1M vectors from 3GB to 96MB) with roughly 2-5% accuracy loss. Scalar quantization offers a milder 4x reduction with 98-99% accuracy.

AgentDB automatically builds HNSW indices, giving O(log n) search around 100µs versus 15ms for linear scans. You can further tune hnswM, hnswEfSearch, and cacheSize, or enable quantization for faster distance computations.

Binary quantization gives 32x memory reduction at 95-98% accuracy, scalar gives 4x at 98-99% accuracy, and product quantization gives 8-16x at 93-97% accuracy. Choose binary for edge deployment, scalar for balanced production use, and product for high-dimensional embeddings.

Yes, quantization trades accuracy for efficiency: binary loses roughly 2-5%, scalar 1-2%, and product 3-7%. Setting quantizationType to 'none' keeps full float32 precision for maximum accuracy on small datasets.

Slow search usually comes from a small cache, high k values, or missing quantization. Increase cacheSize toward 2000, reduce k in retrieveWithReasoning, lower hnswEfSearch, and monitor the cache hit rate via getStats(), aiming for above 80%.

You need Node.js 18 or later and AgentDB v1.0.7+ accessed through the agentic-flow package, plus an existing AgentDB database or application. You can run npx agentdb@latest benchmark to measure baseline performance before tuning.