Struct VectorStore 

pub struct VectorStore { /* private fields */ }

Brute-force vector store (exact nearest-neighbor search).

Suitable for up to ~100K vectors. For larger datasets, use IVF. The flat layout is GPU-friendly — vectors are stored contiguously for efficient parallel distance computation.

Implementations

impl VectorStore

pub fn new(config: VectorStoreConfig) -> Self

Create a new vector store.

pub fn insert(&mut self, entry: VectorEntry) -> Result<(), VectorStoreError>

Insert a vector.

pub fn search( &self, query: &[f32], k: usize, ) -> Result<Vec<SearchResult>, VectorStoreError>

Search for the top-K nearest neighbors.

pub fn delete(&mut self, id: u64) -> bool

Delete a vector by ID.

pub fn len(&self) -> usize

Number of vectors stored.

pub fn is_empty(&self) -> bool

Check if empty.

pub fn flat_vectors(&self) -> &[f32]

Get the flat vector data (for GPU upload).

Returns a contiguous f32 slice of shape [count × dimensions].

pub fn ids(&self) -> &[u64]

Get vector IDs.

pub fn dimensions(&self) -> u32

Dimensionality.