Struct EmbeddingService 

pub struct EmbeddingService { /* private fields */ }

Main embedding service for generating text embeddings

Uses spawn_blocking to run fastembed’s synchronous operations without blocking the async runtime.

The model is wrapped in Arc<Mutex<TextEmbedding>> to allow safe reuse across async boundaries without recreating the model on each call.

Implementations

impl EmbeddingService

pub fn new(config: EmbeddingConfig) -> Result<Self>

Create a new embedding service with the given configuration

Uses a per-model lock to prevent race conditions when multiple threads try to download/initialize the same model simultaneously.

pub fn with_default_model() -> Result<Self>

Create a new embedding service with the default model

pub fn with_model(model: EmbeddingModelType) -> Result<Self>

Create a new embedding service with a specific model

pub fn model_type(&self) -> EmbeddingModelType

Get the current model type

pub fn dimensions(&self) -> usize

Get the embedding dimensions

pub fn config(&self) -> &EmbeddingConfig

Get the configuration

pub async fn embed_text(&self, text: &str) -> Result<Vec<f32>>

Embed a single text (async via spawn_blocking)

pub async fn embed_texts<S: AsRef<str> + Send + Sync + 'static>( &self, texts: &[S], ) -> Result<Vec<Vec<f32>>>

Embed multiple texts in batches (async via spawn_blocking)

This is more efficient than calling embed_text multiple times as it batches the texts and processes them together.

The model is reused across calls via Arc<Mutex<TextEmbedding>>.

pub async fn embed_sparse<S: AsRef<str> + Send + Sync + 'static>( &self, texts: &[S], ) -> Result<Vec<SparseEmbedding>>

Generate sparse embeddings for hybrid search

The sparse model is reused across calls via Arc<Mutex<SparseTextEmbedding>>.