Struct Database 

pub struct Database {
    pub data_path: String,
    pub index_path: String,
    pub bm25_index_path: String,
    pub index: Index,
    pub bm25_index: Bm25Index,
    pub bm25_tokenizer: Tokenizer,
    pub hnsw: HNSW,
    pub graph_path: String,
}

Fields

data_path: String

index_path: String

bm25_index_path: String

index: Index

bm25_index: Bm25Index

bm25_tokenizer: Tokenizer

hnsw: HNSW

graph_path: String

Implementations

impl Database

pub fn new( data_path: &str, index_path: &str, bm25_index_path: &str, hnsw_path: &str, graph_path: &str, ) -> Database

pub fn insert_raw(&mut self, vector: Vec<f32>, text: &str, id: Option<&str>)

Insert a record with a pre-computed embedding vector and text metadata. id pins the record to a specific key; pass None to auto-generate a UUID v7. The text is indexed into BM25 for full-text search; the vector is inserted into the HNSW graph.

pub fn text_search(&self, query: &str, k: usize) -> Vec<Record>

pub fn delete(&mut self, id: &str) -> bool

Delete a record by UUID. Returns false if the ID was not found.

pub fn wipe(&mut self)

Delete all records and wipe every index and data file.

pub fn search_scored( &self, query_vector: &[f32], k: usize, ) -> Vec<(f32, Record)>

Brute-force cosine-similarity search over all stored vectors. Returns up to k records sorted by descending score. Useful for recall evaluation against HNSW results.

pub fn search_hnsw(&self, query_vector: &[f32], ef: usize) -> Vec<Record>

HNSW approximate nearest-neighbour search.

Returns records sorted by descending similarity. ef is the exploration factor — higher values improve recall at the cost of speed.

pub fn search(&self, query_vector: &[f32], k: usize) -> Vec<Record>

Brute-force search returning the top-k records (without scores).