Crate dynvec 

Vector storage + HNSW ANN index engine.

dynvec provides three things:

1. A node-local vector row store with two encodings (per-vector int8 quantisation and IEEE 754 half-precision floats) and three distance metrics (euclidean, cosine, dot product).
2. An HNSW approximate-nearest-neighbour index for k-NN queries.
3. A per-table Engine handle, which is the unit the dynomite::vector registry hands out when serving Redis Stack RediSearch FT.* commands.

Module layout

• encoding – vector compression codecs.
• distance – L2 / cosine / dot product scoring.
• index – the HNSW graph.
• storage – row store + per-table HNSW index, with a pluggable storage::Backend trait so the same surface works against an in-memory backend (default) or a Noxu DB (off-by-default noxu feature).
• engine – per-table handle exposed to embedders.
• [api] – the HTTP API (gated on the http feature; kept as a debug surface only).

The distributed k-NN coordinator that used to live here has moved to dynomite::vector::query_fsm, where it can sit against the cluster machinery directly.

Quick start

use std::collections::HashMap;
use dynvec::distance::Distance;
use dynvec::encoding::Codec;
use dynvec::index::HnswParams;
use dynvec::storage::{TableSchema, VectorStore};

let store = VectorStore::in_memory();
store.create_table(TableSchema {
    name: "demo".to_string(),
    dim: 3,
    codec: Codec::Int8Quantized,
    distance: Distance::Cosine,
    hnsw: HnswParams::default(),
}).unwrap();
store
    .upsert("demo", b"a".to_vec(), &[1.0, 0.0, 0.0], HashMap::new())
    .unwrap();
store
    .upsert("demo", b"b".to_vec(), &[0.0, 1.0, 0.0], HashMap::new())
    .unwrap();
let hits = store.search("demo", &[0.95, 0.05, 0.0], 1, None).unwrap();
assert_eq!(hits[0].0.key, b"a");

Re-exports

pub use crate::distance::Distance;

pub use crate::encoding::decode_turbovec;

pub use crate::encoding::distance_turbovec;

pub use crate::encoding::encode_turbovec;

pub use crate::encoding::Codec;

pub use crate::encoding::EncodedVector;

pub use crate::encoding::Encoder;

pub use crate::encoding::Fp16;

pub use crate::encoding::Int8Quantized;

pub use crate::encoding::Turbovec;

pub use crate::engine::Engine;

pub use crate::index::HnswIndex;

pub use crate::index::HnswParams;

pub use crate::index::NodeId;

pub use crate::index::SearchResult;

pub use crate::storage::Backend;

pub use crate::storage::MemoryBackend;

pub use crate::storage::RowKey;

pub use crate::storage::StoreError;

pub use crate::storage::TableSchema;

pub use crate::storage::TableStats;

pub use crate::storage::VectorRow;

pub use crate::storage::VectorStore;

pub use crate::turbo_index::TurboTable;

Modules

distance

Distance metrics for vector search.

encoding

Vector encodings.

engine

Per-table vector engine handle.

index

Approximate nearest-neighbour index.

storage

Vector row storage.

turbo_index

turbovec-backed approximate nearest-neighbour table.