Struct ExactVectorIndex 

pub struct ExactVectorIndex { /* private fields */ }

The native, in-process VectorIndex: exact (brute-force) cosine ranking over an in-memory map — the true top-k, ties broken by id ascending, so the same corpus and query always answer in the same order. For development, tests, and small single-process corpora: data is not persisted, and a restart starts empty.

Implementations

impl ExactVectorIndex

pub fn new(dimensions: usize) -> Self

An empty index whose vectors are all dimensions wide — the width of whatever Embedder is wired alongside it.

Trait Implementations

impl VectorIndex for ExactVectorIndex

fn upsert<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, namespace: &'life1 VectorNamespace, records: &'life2 [VectorRecord], ) -> Pin<Box<dyn Future<Output = Result<(), VectorIndexError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Writes every record into namespace, replacing any existing record with the same id — values and metadata both, never a merge.

fn query<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, namespace: &'life1 VectorNamespace, vector: &'life2 [f32], k: usize, filter: &'life3 VectorFilter, ) -> Pin<Box<dyn Future<Output = Result<Vec<VectorMatch>, VectorIndexError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait,

The at-most-k nearest neighbours of vector in namespace, cosine similarity descending, every match satisfying filter.

fn delete<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, namespace: &'life1 VectorNamespace, ids: &'life2 [String], ) -> Pin<Box<dyn Future<Output = Result<(), VectorIndexError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Removes the records named by ids from namespace. A missing id is not an error: the delete is idempotent.