Crate vicinity 

Approximate nearest-neighbor search.

vicinity provides Rust indexes and Python bindings for vector search. The default feature set enables HNSW and SIMD distance kernels. Other indexes are available behind feature flags.

Install

vicinity = { version = "0.10.5", features = ["hnsw"] }

Start With HNSW

HNSW is the default in-memory index for dense vectors. Cosine distance expects unit-norm vectors unless auto_normalize(true) is set.

use vicinity::hnsw::HNSWIndex;

let mut index = HNSWIndex::builder(128)
    .m(16)
    .ef_search(50)
    .auto_normalize(true)
    .build()?;

index.add_slice(0, &[0.1; 128])?;
index.add_slice(1, &[0.2; 128])?;
index.build()?;

let results = index.search(&[0.1; 128], 5, 50)?;

Other Indexes

Workload	Start with	Feature
Dense vectors that fit in memory	hnsw::HNSWIndex	hnsw
Raw vectors dominate RAM	ivf_pq::IVFPQIndex	ivf_pq
Frequent writes/deletes	store::UpdatableIndex or FreshGraph	store, fresh_graph
Metadata filters	HNSW post-filtering, ACORN, Curator, or FilteredGraph	hnsw, curator, filtered_graph
Sparse learned retrieval	SparseMIPS	sparse_mips
File-backed search	DiskANN	diskann

The repository README has current benchmark commands. The full feature flag table is in docs/algorithms.md.

Re-exports

pub use distance::DistanceMetric;

pub use error::Result;

pub use error::RetrieveError;

pub use memory::MemoryReport;

Modules

adsampling

ADSampling: adaptive early termination for graph-based ANN search.

classic

Classic ANN methods implementation.

compression

Compression for vector IDs in ANN indexes.

distance

Distance metrics for dense vectors.

error

Error types for vicinity.

filtering

Filtering and lightweight faceting support for vector search.

hnsw

Hierarchical Navigable Small World (HNSW) approximate nearest neighbor search.

lid

Local Intrinsic Dimensionality (LID) estimation.

memory

Memory reporting types for index introspection.

partitioning

Partitioning/clustering for ANN methods.

persistence

Disk persistence for vicinity indexes.

prt

Probabilistic Routing Test (PRT) for graph-based ANN search.

streaming

Streaming updates for vector indices.