A from-scratch HNSW approximate-nearest-neighbor index — the algorithm behind vector databases. Move your cursor over the field to query the index; warren returns its nearest neighbors and we compare them to exact search, live. This runs the real Python library (with NumPy) in your browser via Pyodide.
Gray dots are the indexed vectors. As you move the cursor (the query), blue rings mark the true nearest neighbors from exact brute-force search and filled dots mark what warren's HNSW returns — when they coincide, recall is 100%. "DB scanned" is the fraction of vectors whose distance warren actually computed, the sublinear-search win. Python fetched verbatim from src/warren.