Struct RuVectorStore 

pub struct RuVectorStore { /* private fields */ }

RuVector store — manages multiple per-table VectorDB instances.

Implementations

impl RuVectorStore

pub async fn open(path: &Path, dimensions: usize) -> Result<Self, RuVectorError>

Open (or create) a RuVector store at the given directory.

dimensions must equal the output dimension of the embedding model in use. Passing the wrong dimension will cause Dimension mismatch errors on insert. Use VECTOR_DIM as the default (384) when the embedding provider is not yet known, and prefer probing the actual embedder output at startup.

pub async fn open_with_config( path: &Path, dimensions: usize, hnsw: HnswConfig, ) -> Result<Self, RuVectorError>

Open with explicit HNSW tuning (Issue 37). Threading the knobs through here lets brain serve honour storage.hnsw.* from config instead of locking every install to the hardcoded defaults baked into this crate.

pub async fn ensure_tables(&self) -> Result<(), RuVectorError>

Ensure the standard vector tables exist (idempotent).

Retries up to 5 times with exponential backoff (200ms → 3.2s) to handle transient file-lock contention from other brain processes (e.g. a standalone brain mcp holding the redb lock when the daemon starts).

pub async fn add_vectors( &self, table_name: &str, ids: Vec<String>, _contents: Vec<String>, vectors: Vec<Vec<f32>>, _timestamps: Vec<String>, _source_type: &str, ) -> Result<(), RuVectorError>

Add vectors to a table. ids and vectors must have the same length.

pub async fn search( &self, table_name: &str, query_vector: Vec<f32>, top_k: usize, ) -> Result<Vec<VectorResult>, RuVectorError>

Search for the most similar vectors using cosine distance.

Returns results sorted by distance ascending (closest first).

pub async fn delete( &self, table_name: &str, id: &str, ) -> Result<(), RuVectorError>

Delete a vector by ID from a table.

pub async fn delete_batch( &self, table_name: &str, ids: &[&str], ) -> Result<Vec<(String, RuVectorError)>, RuVectorError>

Delete many vectors by ID in one read-lock cycle. Used by handle_forget so a batch of N matched facts collapses from N delete calls (each grabbing + releasing the tables lock) to one. Per-id failures are collected and returned as a Vec of (id, error) pairs so the caller can decide whether to abort or continue; the SQLite side has its own batch flow that doesn’t short-circuit on a single failure either.

pub async fn table_count( &self, table_name: &str, ) -> Result<usize, RuVectorError>

Get the row count for a table.

pub async fn table_names(&self) -> Result<Vec<String>, RuVectorError>

List all open table names.

Trait Implementations

impl Clone for RuVectorStore

fn clone(&self) -> RuVectorStore

Returns a duplicate of the value.

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

Performs copy-assignment from source.