Key points:

- HNSW indexing with O(log n) search - SQ8 quantization (3.6x memory reduction) - 329µs search latency at 100k vectors - 213 KB gzipped bundle - 24x faster than voy (fastest pure-WASM competitor) - Works in browsers, Node.js, and edge devices

New in v0.4.0: - Complete docs: tutorial, tuning guide, troubleshooting - Migration from hnswlib/FAISS/Pinecone - Interactive benchmark dashboard - Chaos testing (15 edge cases) - P99 latency tracking in CI

GitHub: https://github.com/matte1782/edgevec

The use case is offline-first AI applications where you want vector search without network latency or sending embeddings to a cloud service.