Skip to main content

VectorStore

tinyagents::harness::embeddings

Trait VectorStore 

Source 
pub trait VectorStore: Send + Sync {
    // Required methods
    fn add<'life0, 'async_trait>(
        &'life0 self,
        id: String,
        vector: Vec<f32>,
        metadata: Value,
    ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;
    fn query<'life0, 'life1, 'async_trait>(
        &'life0 self,
        vector: &'life1 [f32],
        top_k: usize,
    ) -> Pin<Box<dyn Future<Output = Result<Vec<ScoredDoc>>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait,
             'life1: 'async_trait;
}
Expand description

A store of dense vectors that supports nearest-neighbour search.

Implementations associate an id and arbitrary metadata with each vector, and answer top-k similarity queries. This module’s InMemoryVectorStore ranks results by cosine similarity.

Implementations must be Send + Sync so they can be shared behind an Arc.

Required Methods§

Source

fn add<'life0, 'async_trait>( &'life0 self, id: String, vector: Vec<f32>, metadata: Value, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait,

Adds (or overwrites, when id already exists) a vector with associated metadata.

Implementations should reject vectors whose dimensionality does not match the vectors already stored (and zero-length vectors) with TinyAgentsError::Validation, so a store never mixes incomparable vectors.

Source

fn query<'life0, 'life1, 'async_trait>( &'life0 self, vector: &'life1 [f32], top_k: usize, ) -> Pin<Box<dyn Future<Output = Result<Vec<ScoredDoc>>> + Send + 'async_trait>>
where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Returns up to top_k documents most similar to vector, sorted by descending similarity score.

Returns fewer than top_k documents when the store holds fewer entries, and an empty Vec when top_k is 0 or the store is empty (an empty store has no dimensionality to validate against, so any query vector is answered with no hits). Implementations should reject a query vector whose dimensionality does not match the stored vectors with TinyAgentsError::Validation instead of returning meaningless scores.

Dyn Compatibility§

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors§