Module vector 

Vector store — cosine similarity search over memory embeddings.

Provides true vector similarity search using embeddings stored in the Embeddings galaxy (LMDB). Builds an in-memory index from all stored embeddings for fast cosine similarity lookups.

For small-to-medium datasets (<100K memories), brute-force cosine similarity is fast enough (sub-millisecond for 10K vectors at 384 dims). For larger datasets, LanceDB can be added as an optional backend.

Structs

VectorSearchResult

A vector search result.

VectorStore

In-memory vector index for fast cosine similarity search.

Traits

VectorSearchEngine

Trait for pluggable vector search backends.