v0.1.20
latestDeep search answers in about a second, and short queries find the right note again.
Desktop
- Smart search now keeps its engine resident between queries: the first deep search of a session takes a couple of seconds and the rest land in about one, instead of reloading 1.2 GB of models every time.
- The engine releases that memory after five idle minutes and shuts down entirely after ten, so nothing stays resident while you are not searching.
- Short queries like a single word no longer come back empty — the closest note is always shown, because a two-character query scores low against a long note even when it is exactly the one you meant.
- Switching the active Vault now rebuilds the smart search index instead of quietly returning nothing.
- Deep search results no longer show internal snippet markers.