← Home

sutraDB: Integrated Vector-Graph

sutraDB synthesizes the HNSW vector index directly with the relationship graph. As the vector traversal lands on each similar node, it immediately follows :hasFather edges — it doesn't wait for all vector results first. Graph hops happen during the vector search, not after it. One graph, interleaved traversal.

Instead of two separate graphs glued by JSON, sutraDB reverses the HNSW graph and weaves relationship edges directly into the vector index. When the vector traversal lands on a node similar to "John," it doesn't just record it and move on — it immediately follows :hasFather edges right then and there, while the vector search continues finding more similar names. The graph hops and vector hops are interleaved, not sequential.
Traditional Hybrid
operations
sutraDB
operations
System Boundaries
context switches
System Boundaries
0
context switches
Click "Run Unified Query" to find great-grandfathers of people named like "John"
By unifying the two graphs, sutraDB can answer queries that neither database can answer alone: "find the great-grandfathers of people whose names are semantically similar to John" — combining fuzzy vector similarity with precise multi-hop graph traversal in a single, interleaved operation. Results stream in as they're found, not all at once at the end.