Vector search systems today face a critical dilemma: choose between the blazing speed of in-memory solutions like HNSW or the cost-effective scalability of disk-based approaches like DiskANN. This trade-off has forced organizations to either bear excessive infrastructure costs or accept compromised performance.
CHANNI bridges this gap with a novel multi-level architecture that combines the best aspects of both approaches. By intelligently managing data between memory and disk through its unique cluster-primary representation and nested navigation structure, CHANNI achieves:
