v2.20.1
Fixed
- Dashboard took ~20s to load —
/api/memoriesand/api/tagsawaited the embedding-model warmup (a multi-hundred-MB pipeline init) even though listing is a pure database read. Warmup removed from those handlers; the endpoints now respond in milliseconds. Semantic search still warms up on demand.
Full Changelog: v2.20.0...v2.20.1