Struct EmbedVec 

pub struct EmbedVec {
    pub index: Arc<RwLock<HnswIndex>>,
    pub storage: Arc<RwLock<VectorStorage>>,
    pub metadata: Arc<RwLock<Vec<Metadata>>>,
    /* private fields */
}

Main embedded vector database struct

Provides HNSW-based approximate nearest neighbor search with optional E8 or H4 lattice quantization for memory efficiency.

Fields

index: Arc<RwLock<HnswIndex>>

HNSW index

storage: Arc<RwLock<VectorStorage>>

Vector storage (raw, E8-quantized, or H4-quantized)

metadata: Arc<RwLock<Vec<Metadata>>>

Metadata storage

Implementations

impl EmbedVec

pub async fn new( dim: usize, distance: Distance, m: usize, ef_construction: usize, ) -> Result<Self>

Create a new in-memory EmbedVec instance

Arguments

• dim - Vector dimension (e.g., 768 for many LLM embeddings)
• distance - Distance metric (Cosine, Euclidean, DotProduct)
• m - HNSW M parameter (connections per layer, typically 16-64)
• ef_construction - HNSW construction parameter (typically 100-500)

Example

use embedvec::{EmbedVec, Distance};

#[tokio::main]
async fn main() {
    let db = EmbedVec::new(768, Distance::Cosine, 32, 200).await.unwrap();
}

pub async fn with_persistence( path: impl AsRef<Path>, dim: usize, distance: Distance, m: usize, ef_construction: usize, ) -> Result<Self>

Create a new EmbedVec with persistence (uses Sled by default)

pub async fn with_backend( config: BackendConfig, dim: usize, distance: Distance, m: usize, ef_construction: usize, ) -> Result<Self>

Create a new EmbedVec with a specific persistence backend

pub fn new_internal( dim: usize, distance: Distance, m: usize, ef_construction: usize, quantization: Quantization, persistence_config: Option<BackendConfig>, ) -> Result<Self>

Internal constructor (public for Python bindings) — with persistence

pub fn builder() -> EmbedVecBuilder

Get a builder for configuring EmbedVec

pub async fn add( &mut self, vector: &[f32], payload: impl Into<Metadata>, ) -> Result<usize>

Add a single vector with metadata

Arguments

• vector - The embedding vector (must match configured dimension)
• payload - Associated metadata (JSON-compatible)

Returns

The assigned vector ID

pub async fn add_many( &mut self, vectors: &[Vec<f32>], payloads: Vec<impl Into<Metadata>>, ) -> Result<()>

Add multiple vectors with metadata in batch

Arguments

• vectors - Slice of embedding vectors
• payloads - Associated metadata for each vector

pub fn add_internal( &mut self, vector: &[f32], payload: Metadata, ) -> Result<usize>

Internal add implementation (public for Python bindings)

pub async fn search( &self, query: &[f32], k: usize, ef_search: usize, filter: Option<FilterExpr>, ) -> Result<Vec<Hit>>

Search for nearest neighbors

Arguments

• query - Query vector
• k - Number of results to return
• ef_search - Search parameter (higher = better recall, slower)
• filter - Optional metadata filter expression

Returns

Vector of Hit results sorted by similarity

pub fn search_internal( &self, query: &[f32], k: usize, ef_search: usize, filter: Option<FilterExpr>, ) -> Result<Vec<Hit>>

Internal search implementation (public for Python bindings)

pub async fn len(&self) -> usize

Get the number of vectors in the database

pub async fn is_empty(&self) -> bool

Check if the database is empty

pub async fn clear(&mut self) -> Result<()>

Clear all vectors and metadata

pub fn quantization(&self) -> &Quantization

Get current quantization mode

pub async fn set_quantization(&mut self, quant: Quantization) -> Result<()>

Set quantization mode (re-quantizes all existing vectors)

pub fn dimension(&self) -> usize

Get vector dimension

pub fn distance(&self) -> Distance

Get distance metric