Trait EmbeddingStore 

pub trait EmbeddingStore: Send + Sync {
    // Required methods
    fn upsert(&self, chunks: &[EmbeddingChunk]) -> Result<usize>;
    fn insert(&self, chunks: &[EmbeddingChunk]) -> Result<usize>;
    fn search(
        &self,
        query_vec: &[f32],
        top_k: usize,
    ) -> Result<Vec<ScoredChunk>>;
    fn delete_by_file(&self, file_paths: &[&str]) -> Result<usize>;
    fn clear(&self) -> Result<()>;
    fn count(&self) -> Result<usize>;

    // Provided methods
    fn search_dual(
        &self,
        query_vec: &[f32],
        top_k: usize,
        doc_weight: f64,
    ) -> Result<Vec<ScoredChunk>> { ... }
    fn get_embedding(
        &self,
        _file_path: &str,
        _symbol_name: &str,
    ) -> Result<Option<EmbeddingChunk>> { ... }
    fn embeddings_for_scored_chunks(
        &self,
        chunks: &[ScoredChunk],
    ) -> Result<Vec<EmbeddingChunk>> { ... }
    fn all_with_embeddings(&self) -> Result<Vec<EmbeddingChunk>> { ... }
    fn embeddings_for_files(
        &self,
        file_paths: &[&str],
    ) -> Result<Vec<EmbeddingChunk>> { ... }
    fn for_each_embedding_batch(
        &self,
        batch_size: usize,
        visitor: &mut dyn FnMut(Vec<EmbeddingChunk>) -> Result<()>,
    ) -> Result<()> { ... }
    fn for_each_file_embeddings(
        &self,
        visitor: &mut dyn FnMut(String, Vec<EmbeddingChunk>) -> Result<()>,
    ) -> Result<()> { ... }
}

Trait for vector embedding storage backends. Implementations handle persistence, indexing, and similarity search.

Required Methods

fn upsert(&self, chunks: &[EmbeddingChunk]) -> Result<usize>

Insert or update embedding chunks. Replaces ALL existing entries.

fn insert(&self, chunks: &[EmbeddingChunk]) -> Result<usize>

Append embedding chunks without clearing existing data.

fn search(&self, query_vec: &[f32], top_k: usize) -> Result<Vec<ScoredChunk>>

Search for chunks similar to the query embedding vector.

fn delete_by_file(&self, file_paths: &[&str]) -> Result<usize>

Delete all embeddings for files matching the given paths.

fn clear(&self) -> Result<()>

Clear all stored embeddings.

fn count(&self) -> Result<usize>

Number of stored chunks.

Provided Methods

fn search_dual( &self, query_vec: &[f32], top_k: usize, doc_weight: f64, ) -> Result<Vec<ScoredChunk>>

Dual-vector search: blend code embedding score with doc embedding score. doc_weight controls the balance (0.0 = code only, 1.0 = doc only).

fn get_embedding( &self, _file_path: &str, _symbol_name: &str, ) -> Result<Option<EmbeddingChunk>>

Retrieve a single stored chunk and embedding by symbol identity.

fn embeddings_for_scored_chunks( &self, chunks: &[ScoredChunk], ) -> Result<Vec<EmbeddingChunk>>

Retrieve stored chunks matching previously ranked search results so callers can batch exact-vector follow-up work without per-result lookups.

fn all_with_embeddings(&self) -> Result<Vec<EmbeddingChunk>>

Retrieve all stored chunks with their embeddings for batch analysis.

fn embeddings_for_files( &self, file_paths: &[&str], ) -> Result<Vec<EmbeddingChunk>>

Retrieve stored chunks for the given files so incremental indexing can reuse unchanged embeddings without materializing the full index.

fn for_each_embedding_batch( &self, batch_size: usize, visitor: &mut dyn FnMut(Vec<EmbeddingChunk>) -> Result<()>, ) -> Result<()>

Stream stored chunks in bounded batches so callers can avoid loading the entire embedding index into memory.

fn for_each_file_embeddings( &self, visitor: &mut dyn FnMut(String, Vec<EmbeddingChunk>) -> Result<()>, ) -> Result<()>

Stream stored chunks grouped by file path for per-file analysis without requiring callers to materialize the entire index first. Full and incremental reindex reconciliation rely on this grouping.

Implementors