EmergentDB

A self-optimizing vector database architecture that uses the MAP-Elites quality-diversity algorithm to evolve and select high-performance HNSW index configurations.

EmergentDB replaces manual index tuning with an evolutionary loop called IndexQD, which utilizes the MAP-Elites algorithm to navigate the trade-offs between recall, latency, and memory. By treating the database structure as a phenotype, the system mutates HNSW hyperparameters—specifically M, ef_construction, and ef_search—to populate a multi-dimensional archive of elite configurations. This approach allows the engine to automatically discover optimal graph structures for specific datasets, achieving up to 100% recall and significantly higher throughput than static implementations like ChromaDB or LanceDB by continuously illuminating the most efficient regions of the index parameter space.

https://emergentdb.com