Use smaller local models to extract key metadata or summaries from documents to handle tasks without requiring large-scale vector storage.
