Vector index

A vector index is a data structure that makes similarity search fast by avoiding a full scan. By default Ahnlich compares your query against every stored vector; a vector index pre-structures the vectors so only a fraction need to be checked.

An index skips most of the store instead of scanning every vector.

Ahnlich offers two approaches, from exact-but-linear to approximate-but-fast:

• Exact search — no index; always exact, but cost grows with the store.
• HNSW — approximate, scales to high-dimensional embeddings.

Choosing an index

Situation	Use
Small store, or you need exact results	Exact scan (no index)
Large store of high-dimensional embeddings	HNSW

An index costs extra memory and build time, and approximate indexes (HNSW) trade a little recall for speed — so add one when search latency actually matters.

Adding a vector index

Create one with Create non-linear index (or seed it at store creation); remove it with Drop non-linear index.

Filtering by metadata instead of vectors? That's a different structure — see predicate indexes.

Related

• Similarity metrics
• Similarity search
• Metadata