Struct EmbeddingIndex 

pub struct EmbeddingIndex<T: Clone> { /* private fields */ }

Combined NSW graph index with per-entry metadata storage.

T is any cloneable metadata type — e.g. a String payload, a struct, or a raw identifier.

use oxibonsai_runtime::embedding_index::EmbeddingIndex;

let mut idx: EmbeddingIndex<String> = EmbeddingIndex::new(3);
idx.insert(vec![1.0, 0.0, 0.0], "vec-a".to_string());
idx.insert(vec![0.0, 1.0, 0.0], "vec-b".to_string());

let results = idx.search(&[1.0, 0.0, 0.0], 1);
assert_eq!(results[0].1, &"vec-a".to_string());

Implementations

impl<T: Clone> EmbeddingIndex<T>

pub fn new(dim: usize) -> Self

Create a new index for dim-dimensional vectors with default NSW config.

pub fn new_with_config(dim: usize, config: NswConfig) -> Self

Create a new index with a custom NswConfig.

pub fn insert(&mut self, vector: Vec<f32>, meta: T) -> usize

Insert a vector with associated metadata.

Returns the stable numeric ID assigned to this entry.

pub fn search(&self, query: &[f32], top_k: usize) -> Vec<(NswSearchResult, &T)>

Search for the top-top_k nearest neighbors of query.

Returns a Vec of (NswSearchResult, &T) pairs sorted by descending cosine similarity.

pub fn len(&self) -> usize

Number of entries in the index.

pub fn is_empty(&self) -> bool

Returns true if the index contains no entries.