Vector embedding models for on-device RAG, semantic search, and retrieval-augmented generation. Compact enough to run locally on mobile hardware.