Bitmap filtering is a technique for applying metadata conditions to vector search extremely quickly by representing sets of matching records as compressed bitmaps. A bitmap is a sequence of bits where each position corresponds to a stored vector, set to 1 if that vector matches a condition and 0 otherwise.
The power of this representation is that combining filters becomes a matter of fast bitwise operations. To find vectors matching category = news AND year = 2026, the engine intersects the two bitmaps with a single hardware-accelerated AND operation across many bits at once. This is far faster than checking conditions record by record.
Modern implementations use compressed bitmap formats such as Roaring Bitmaps, which stay compact even for very large or sparse sets. Bitmap filtering is a core reason some vector databases handle complex, high-cardinality metadata filters with little performance penalty, making it central to efficient filtered and hybrid search.