I built simile-search after struggling with semantic search for small projects and product catalogs where running a vector DB or calling APIs felt like overkill.

simile-search is an offline-first semantic search library: - No database - No external APIs - Local embeddings + fuzzy matching - Works well with messy human input (names, typos, partial queries)

It’s designed for cases like local apps, small catalogs, demos, and environments with limited infra.

GitHub: https://github.com/iaavas/simile-search npm: https://www.npmjs.com/package/simile-search

I’d really appreciate feedback on the approach, performance tradeoffs, and where this might break down.