An embedding is a vector of numbers that represents the meaning of a text. Two texts close in meaning have close vectors, which enables semantic search. The query and document vectors must come from the same model.
Example
embedding("refund") ≈ embedding("refund an order")