A payload index is a secondary index built over the metadata — the payload — attached to vectors, so that conditions on that metadata can be evaluated quickly during search. Just as a relational database indexes columns to speed up filtered queries, a vector database indexes payload fields to speed up filtered vector search.
Without such an index, applying a metadata condition would mean checking each candidate vector’s payload one by one, which is slow, especially for selective or high-cardinality filters. A payload index lets the database rapidly identify which vectors satisfy a condition — often by maintaining structures like sorted lists or bitmaps per field — so the filter can be combined with similarity search efficiently.
Payload indexes are central to making filtered and hybrid search performant. They underpin techniques like bitmap filtering and enable the database to apply conditions during traversal rather than as a slow afterthought, helping avoid the recall cliff. The fields you choose to index should match the conditions your queries filter on most, since each index adds storage and write cost in exchange for faster filtered reads.