Struct VectorStoreOptions 

pub struct VectorStoreOptions { /* private fields */ }

Configuration options for opening or creating a vector store.

Follows the std::fs::OpenOptions pattern for familiar, ergonomic API.

Examples

use omendb::vector::store::VectorStoreOptions;

// Simple persistent store
let store = VectorStoreOptions::default()
    .dimensions(384)
    .open("./vectors")?;

// With custom HNSW parameters
let store = VectorStoreOptions::default()
    .dimensions(384)
    .m(32)
    .ef_construction(400)
    .ef_search(100)
    .open("./vectors")?;

// In-memory store
let store = VectorStoreOptions::default()
    .dimensions(384)
    .build()?;

Implementations

impl VectorStoreOptions

pub fn new() -> Self

Create new options with defaults.

pub fn dimensions(self, dim: usize) -> Self

Set vector dimensionality.

If not set, dimensions will be inferred from:

1. Existing data when opening a persistent store
2. First inserted vector

pub fn m(self, m: usize) -> Self

Set HNSW M parameter (neighbors per node).

Higher M = better recall, more memory. Range: 4-64, default: 16.

pub fn ef_construction(self, ef: usize) -> Self

Set HNSW ef_construction (build quality).

Higher = better graph quality, slower build. Default: 100.

pub fn ef_search(self, ef: usize) -> Self

Set HNSW ef_search (search quality/speed tradeoff).

Higher = better recall, slower search. Default: 100.

pub fn quantization(self, mode: QuantizationMode) -> Self

Enable quantization for memory-efficient storage.

Modes

• QuantizationMode::SQ8: 4x compression, similar speed, ~99% recall (default)
• QuantizationMode::RaBitQ(params): 8x compression, ~0.5x speed, 93-99% recall

Example

// SQ8 (recommended for most cases)
let store = VectorStoreOptions::default()
    .dimensions(768)
    .quantization(QuantizationMode::sq8())
    .open("./vectors")?;

// RaBitQ for higher compression
let store = VectorStoreOptions::default()
    .dimensions(768)
    .quantization(QuantizationMode::rabitq())
    .open("./vectors")?;

pub fn quantization_sq8(self) -> Self

Enable SQ8 quantization (4x compression, similar speed, ~99% recall)

Convenience method for the most common quantization mode.

pub fn quantization_rabitq(self) -> Self

Enable RaBitQ quantization with default 4-bit parameters (8x compression)

pub fn quantization_rabitq_params(self, params: RaBitQParams) -> Self

Enable RaBitQ quantization with custom parameters

pub fn rescore(self, enable: bool) -> Self

Enable/disable rescoring with original vectors (default: true when quantization enabled).

When rescoring is enabled, search uses quantized vectors for fast candidate selection, then reranks candidates using full-precision vectors for accuracy.

Arguments

• enable - Whether to rescore candidates

pub fn oversample(self, factor: f32) -> Self

Set oversampling factor for rescoring (default: 3.0).

When rescoring, fetches k * oversample candidates during quantized search, then returns top k after reranking with full precision.

Higher values improve recall but increase latency.

Arguments

• factor - Oversampling multiplier (must be >= 1.0)

pub fn metric(self, m: &str) -> Result<Self, String>

Set distance metric for similarity search.

Metrics

• "l2" or "euclidean": Euclidean distance (default)
• "cosine": Cosine distance (1 - cosine similarity)
• "dot" or "ip": Inner product (for MIPS)

Errors

Returns error if metric string is not recognized.

Example

let store = VectorStoreOptions::default()
    .dimensions(768)
    .metric("cosine")?
    .open("./vectors")?;

pub fn metric_fn(self, m: Metric) -> Self

Set distance metric directly (no parsing).

pub fn text_search(self, enabled: bool) -> Self

Enable tantivy-based full-text search with default configuration.

When enabled, you can use set_with_text() to index text alongside vectors, and hybrid_search() to search both with RRF fusion.

Uses 50MB writer buffer by default. For custom memory settings, use text_search_config() instead.

pub fn text_search_config(self, config: TextSearchConfig) -> Self

Enable text search with custom configuration.

Example

// Mobile: lower memory
let store = VectorStoreOptions::default()
    .text_search_config(TextSearchConfig { writer_buffer_mb: 15 })
    .open("./db")?;

// Cloud: higher throughput
let store = VectorStoreOptions::default()
    .text_search_config(TextSearchConfig { writer_buffer_mb: 200 })
    .open("./db")?;

pub fn open(&self, path: impl AsRef<Path>) -> Result<VectorStore>

Open or create a persistent vector store at the given path.

Creates the directory if it doesn’t exist. Loads existing data if the store already exists.

pub fn build(&self) -> Result<VectorStore>

Build an in-memory vector store (no persistence).

Trait Implementations

impl Clone for VectorStoreOptions

fn clone(&self) -> VectorStoreOptions

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for VectorStoreOptions

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for VectorStoreOptions

fn default() -> VectorStoreOptions

Returns the “default value” for a type.