Patch release with three fixes.
Fixes
zotero_get_search_database_statusno longer reports "0 documents / not initialized" against a populated database — ChromaDB ≥1.x's embedding-function conflict check rejected the status reader's no-op embedding function; it now identifies as"default", which short-circuits the check for any persisted backend (#362, #364).- Semantic search with the reranker enabled no longer times out — the cross-encoder was reloaded from disk on every request (~30s per call); it is now cached process-wide and warmed up in the background at server start, so reranked searches are sub-second after the first load (#283, #365).
- Ollama embeddings use the current
/api/embedendpoint instead of the deprecated/api/embeddingsroute, sending the whole batch in a single request instead of one request per document (#349, #360).
No re-index needed: stored vectors are unaffected by any of these changes.
Full details in CHANGELOG.md.