PathSeer: Adaptive Filtered Vector Search

Weaviate 1.40 adds PathSeer, a strategy for vector searches that also apply structured filters. It combines approaches from the existing sweeping and ACORN strategies rather than replacing them.

Weaviate's implementation benchmarks report a strong recall-throughput trade-off across the tested workloads, with build time of 1.10× and memory use of 1.06× compared with sweeping or ACORN. It adds no new user-facing tuning settings.

PathSeer only activates on indexes built with the extra neighbour data it needs.