Pattern

Semantic response caching

Skip the model call entirely when a new request means the same thing as one already answered, by matching it against stored requests on embedding similarity and returning the stored response.

Classification Reusing stored model responses for new requests whose meaning matches an earlier one.