Search

Searching is done by Find and Rank in browse, and by ask from the command line. All three find notes by meaning, not exact wording — they never generate text; they only ever return lines you already wrote, verbatim. As your file grows, ask and Find will return more alternate results, up to 10. Rank lists every candidate.

Each note is embedded (turned into a vector) by a local model. Your question gets embedded the same way, and llote ranks every note by cosine similarity to it — closest meaning wins, not closest spelling.

The default model is BGE Small EN v1.5. Settings can switch you to a quantized (smaller/faster) or larger BGE model; changing models rebuilds the search index. llote setup downloads the current choice ahead of time; otherwise the first ask or browse search does it for you, which needs internet access once.

Embeddings are cached. Each note's embedding is stored in an index file in app data, keyed by a hash of its timestamp and text, which cuts down on ask times. Switching models, or clearing the index from Settings, starts that cache over.