The default embedding model (all-MiniLM-L6-v2) is English-centric. Cross-lingual search (e.g. querying in Russian while memories are stored in English) works partially but with noticeable quality drop.

Questions:

1. Is there a plan to support paraphrase-multilingual-MiniLM-L12-v2 or another multilingual model?
2. Is this configurable already (e.g. via mempalace.yaml or env var)?
3. Is the quality degradation on cross-lingual queries a known limitation?

This isn't mentioned in the README. Would be helpful to document even if the answer is "not supported yet".