Struct Collection 

pub struct Collection { /* private fields */ }

A single collection: an owned, in-process handle to one collection’s storage. Insert with &mut, query with &. Data is durable via the WAL on every insert; call Collection::flush (or drop the handle) to also persist the dense/sparse indexes for a fast reopen.

Implementations

impl Collection

pub fn insert( &mut self, id: impl Into<VectorId>, vector: Vector, payload: Value, ) -> Result<UpsertResult>

Insert or update a vector with JSON metadata and no text. Timestamps are set to now.

pub fn insert_text( &mut self, id: impl Into<VectorId>, vector: Vector, payload: Value, text: impl Into<String>, ) -> Result<UpsertResult>

Insert or update a vector with JSON metadata and full-text (enables BM25 sparse and hybrid search for this record).

pub fn upsert_record(&mut self, record: VectorRecord) -> Result<UpsertResult>

Insert or update a fully-specified record.

pub fn insert_batch( &mut self, items: Vec<(VectorId, Vector, Value)>, ) -> Result<usize>

Bulk-insert many (id, vector, payload) triples, building the index once at the end. Far faster than repeated Collection::insert for loading a large collection. Returns the number inserted.

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

Delete a record by id.

pub fn query(&self, vector: &[f32], k: usize) -> Result<Vec<SearchResult>>

Dense k-NN search over the vector index.

pub fn query_filtered( &self, vector: &[f32], k: usize, filter: &Value, ) -> Result<Vec<SearchResult>>

Dense k-NN search restricted to records whose payload matches filter, with predicate pushdown: a selective filter scans only the matching vectors (exact, ~proportional to the matches) rather than searching the whole index and discarding non-matches.

filter is the same JSON predicate shape accepted elsewhere, e.g. json!({ "genre": "sci-fi" }) or json!({ "year": { "$gte": 2000 } }).

pub fn query_text(&self, text: &str, k: usize) -> Result<Vec<SearchResult>>

Full-text BM25 search over the sparse index.

pub fn query_hybrid( &self, vector: Option<&[f32]>, text: Option<&str>, k: usize, alpha: f32, ) -> Result<Vec<SearchResult>>

Two-stage hybrid search. At least one of vector / text required. alpha weights dense vs sparse (alpha * dense + (1-alpha) * sparse).

pub fn sql(&self, sql: &str) -> Result<Vec<SearchResult>>

Run a VECTOR_SIM SQL query against this collection.

pub fn len(&self) -> Result<usize>

Number of active (non-deleted) vectors.

pub fn is_empty(&self) -> Result<bool>

Whether the collection has no active vectors.

pub fn stats(&self) -> Result<IndexStats>

Collection statistics (vector count, on-disk bytes, dimension, …).

pub fn name(&self) -> &str

Collection name.

pub fn flush(&mut self) -> Result<()>

Persist indexes and checkpoint the WAL. Data is already durable via the WAL without this; flushing makes the next reopen fast (no index rebuild).

Trait Implementations

impl Drop for Collection

fn drop(&mut self)

Executes the destructor for this type.

fn pin_drop(self: Pin<&mut Self>)

🔬This is a nightly-only experimental API. (pin_ergonomics)

Execute the destructor for this type, but different to Drop::drop, it requires self to be pinned.