Generate an embedding for a search query, to be compared with the embeddings of the memories.
Generate an embedding for a search query, to be compared with the embeddings of the memories.
Some embedding models embed a query and a document differently, and a query is only matched correctly if it is embedded as one. A memory store calls this, not embed, for the text it searches with. The default delegates to embed, so a service whose model embeds both alike (and every existing implementation) does not need to override it.