k-d-trees

K-d trees are what make sub-100ms vector search possible in a browser extension. This post explains how they work, why they matter for semantic history search, and how TraceMind uses Voy's WASM k-d tree to search tens of thousands of embeddings almost instantly.

Searching thousands of 384-dimensional vectors in milliseconds inside a browser requires smarter data structures than linear scan. Here is how Voy, k-d trees, and approximate nearest-neighbor search make it possible.