Data structure that accelerates similarity search, for example HNSW graphs or IVF lists over embedding vectors.