Struct InMemoryVectorStore 

pub struct InMemoryVectorStore { /* private fields */ }

In-process VectorStore backed by a per-namespace Vec<Slot>.

Cloning is cheap — internal state lives behind Arc<RwLock<...>> so multiple SemanticMemory instances can share one store.

Implementations

impl InMemoryVectorStore

pub fn new(dimension: usize) -> Self

Build an empty store fixed to dimension. Inserts whose vector.len() differs surface Error::InvalidRequest at add time.

pub fn total_slots(&self) -> usize

Total slot count across every namespace. Useful for tests.

Trait Implementations

impl Clone for InMemoryVectorStore

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for InMemoryVectorStore

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl VectorStore for InMemoryVectorStore

fn dimension(&self) -> usize

Vector dimension this index expects.

fn add<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, document: Document, vector: Vec<f32>, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

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

Add a document with its pre-computed vector to the index. Implementations validate vector.len() == self.dimension().

fn search<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, query_vector: &'life3 [f32], top_k: usize, ) -> Pin<Box<dyn Future<Output = Result<Vec<Document>>> + Send + 'async_trait>>

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

Search for the top top_k nearest documents to query_vector.

fn delete<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, doc_id: &'life3 str, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

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

Delete a document by its backend-assigned id. Default impl returns Error::Config — backends without a stable id space must override or document the lifecycle.

fn update<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, doc_id: &'life3 str, document: Document, vector: Vec<f32>, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

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

Replace an existing document’s vector and metadata. Default impl chains delete + add (non-atomic — concurrent searches observe a gap); backends with atomic-update support must override.

fn search_filtered<'life0, 'life1, 'life2, 'life3, 'life4, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, query_vector: &'life3 [f32], top_k: usize, filter: &'life4 VectorFilter, ) -> Pin<Box<dyn Future<Output = Result<Vec<Document>>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait, 'life4: 'async_trait,

Top-K nearest matches with a metadata filter pushed down to the index. Default impl returns entelix_core::Error::Config — silently dropping the filter would return wrong results, so the trait makes the backend’s lack of filter support explicit. Backends with filter support must override.

fn count<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, filter: Option<&'life3 VectorFilter>, ) -> Pin<Box<dyn Future<Output = Result<usize>> + Send + 'async_trait>>

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

Count documents in the namespace, optionally narrowed by a filter. Used by dashboards reporting per-tenant index sizes and by memory-budget enforcement (skip indexing when the namespace is at its cap).

fn list<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, _ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, filter: Option<&'life3 VectorFilter>, limit: usize, offset: usize, ) -> Pin<Box<dyn Future<Output = Result<Vec<Document>>> + Send + 'async_trait>>

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

Enumerate documents in the namespace, optionally narrowed by a filter. limit caps the page size; offset is the page start (cursor-style pagination semantics depend on the backend). Returned documents may omit their vectors — the method is for inspection / pagination, not for retrieval.

fn add_batch<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, ctx: &'life1 ExecutionContext, ns: &'life2 Namespace, items: Vec<(Document, Vec<f32>)>, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

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

Insert many documents at once. Default impl loops over add, polling ExecutionContext::is_cancelled between iterations so a cancelled caller releases the index lock within one add round-trip instead of completing the full batch. Backends that support a native batch endpoint should override — sequential calls amplify network latency by N.