Semantic search over your own notes — entirely in your browser. Paste any set of passages, ask questions in plain words, and get results ranked by meaning, not keywords. Embeddings are computed locally; the text never leaves the page.
Model
all-MiniLM-L6-v2 · 384-dim sentence embeddings
detecting device…
~23 MB one-time download, cached by your browser afterwards.
Corpus
load sample corpus
Query
How it works: every passage and the query are mapped to 384-dimensional vectors by a sentence-transformer model running locally (mean-pooled, L2-normalized). Results are ranked by cosine similarity, which captures meaning — "the meeting about hiring" matches "we decided to postpone recruitment" even though they share no keyword. Server-side multimodal embedders exist for production scale; this page proves the whole idea works with zero infrastructure, zero cost, and zero data exposure. Clearing your browser storage removes the cached model; your text is never stored anywhere.