Embeddings that represent individual tokens (words or subwords) rather than entire documents, allowing fine-grained matching during search.