Struct Retriever 

pub struct Retriever { /* private fields */ }

Orchestrates the retrieval modes over a store + embedder (+ optional LLM).

Implementations

impl Retriever

pub fn new( store: Arc<dyn VectorStore>, embedder: Arc<dyn Embedder>, chat: Option<Arc<dyn ChatModel>>, ) -> Self

Build a retriever. Pass chat = None to run without an LLM (vector/bm25/hybrid only).

pub fn with_rrf_k(self, k: f32) -> Self

Override the RRF constant (default 60).

pub fn with_multiquery_n(self, n: usize) -> Self

Override the number of Multi-Query rewrites (default 4).

pub async fn retrieve( &self, mode: RetrievalMode, query: &str, k: usize, ) -> Result<Vec<Scored>>

Retrieve the top k chunks for query using mode.

pub async fn vector(&self, query: &str, k: usize) -> Result<Vec<Scored>>

Dense vector search.

pub async fn bm25(&self, query: &str, k: usize) -> Result<Vec<Scored>>

Sparse BM25 keyword search over the whole chunk corpus.

pub async fn hybrid(&self, query: &str, k: usize) -> Result<Vec<Scored>>

Hybrid RAG: fuse dense + sparse results with RRF. Over-fetches each arm so fusion has depth to work with.

pub async fn multi_query(&self, query: &str, k: usize) -> Result<Vec<Scored>>

Multi-Query (Fusion) RAG: the LLM rewrites the question into several diverse queries; each is retrieved (hybrid) and the results are fused with RRF.

pub async fn hyde(&self, query: &str, k: usize) -> Result<Vec<Scored>>

HyDE: the LLM writes a hypothetical answer; its embedding drives the search.

Trait Implementations

impl Clone for Retriever

fn clone(&self) -> Retriever

Returns a duplicate of the value.

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

Performs copy-assignment from source.