Struct VectorIndex 

pub struct VectorIndex { /* private fields */ }

Vector index using simple brute-force search

This implementation uses linear scan which is efficient for small collections (< 10,000 vectors). For larger collections, consider HNSW or IVF indexing.

Implementations

impl VectorIndex

pub fn new(dimension: usize) -> Self

Create new index

pub fn add(&mut self, chunk_id: String, embedding: Vec<f32>) -> Result<()>

Add embedding to index.

The embedding is L2-normalized at insert time so that cosine similarity reduces to a simple dot product during search.

pub fn remove(&mut self, chunk_id: &str)

Remove embedding by chunk ID

pub fn search(&self, query: &[f32], k: usize) -> Vec<(String, f32)>

Search for similar embeddings

Uses a min-heap to efficiently track the top-k results without sorting the entire result set. Also applies early termination when all top-k results have similarity > 0.95.

Because all stored embeddings are L2-normalized at insert time, cosine similarity is just the dot product (no per-query sqrt needed).

pub fn cosine_similarity(a: &[f32], b: &[f32]) -> f32

Cosine similarity between two arbitrary vectors.

Normalizes both inputs before computing the dot product. Kept for external callers and tests; the hot search path uses dot_product on pre-normalized vectors instead.

pub fn len(&self) -> usize

Get index size

pub fn is_empty(&self) -> bool

Check if empty

pub fn clear(&mut self)

Clear index

pub fn verify_index_integrity(&self) -> Vec<String>

Verify index integrity, returning a list of issues found.

Checks for:

• Mismatched embedding dimensions
• NaN or Inf values in vectors
• Duplicate chunk IDs
• Empty embedding vectors

pub fn check_health(&self) -> IndexHealth

Check overall health of the index.