iaavas/simile-search

Built on Transformers.js for client-side embeddings, Simile employs HNSW indexing for O(log n) search performance on large datasets and supports quantization (float16/int8) to reduce memory footprint by up to 75%. The engine scores results by blending semantic similarity, fuzzy matching, and keyword matching with configurable weights, while supporting asynchronous indexing and persistent snapshots to avoid recomputation across application restarts.

Available on npm.