Cosdata

Vector Search Reimagined

Powered by CHANNI, a ground-breaking multi-level vector search index that combines hierarchical graph navigation with intelligent clustering, Cosdata's vector database delivers unmatched performance by combining blazing-fast in-memory search capabilities with disk-based scalability.

Read The CHANNI Paper

Preface: The Need for CHANNI

The Vector Search Challenge

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.

Current Approaches and Their Limitations

HNSW (In-Memory)

✓Exceptional query performance
✓High recall rates
✓Ideal for real-time applications
✗Requires entire index in RAM
✗Costly at scale

Disk-Based ANN

✓Excellent scalability
✓Cost-effective storage
✓Efficient memory usage
✗Higher query latency
✗Complex maintenance

The CHANNI Solution

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:

→Near-HNSW query performance while keeping most data on disk

→Dramatic reduction in infrastructure costs (up to 90% less RAM required)

→Ability to scale to billions of vectors without performance degradation

→Simplified maintenance through intelligent cluster operations

Read the CHANNI Paper

Clustered Hierarchical Approximate Nested Navigable Index (CHANNI)

Read the CHANNI Paper