PathSeer filtered ANN algorithm ships in Weaviate 1.40
Filtered vector search usually forces a trade-off between pre-filtering and post-filtering, and PathSeer removes the need to pick a side or tune per query. Teams running production RAG or hybrid search can enable it on existing HNSW collections with a one-line config change and no reindexing. Weaviate is open source, so the implementation can be compared directly against alternatives like Qdrant's filtered search handling.
Weaviate 1.40 introduces PathSeer, a new filtered approximate nearest neighbour (ANN) algorithm that dynamically chooses a search strategy without query-specific tuning. In the team's benchmarks, PathSeer was the fastest strategy across almost every workload tested, including all six production-like conjunction filters. It works with existing HNSW indexes, so collections can switch via "filterStrategy": "pathseer" without reindexing. A full technical breakdown is published on the Weaviate blog.
It is not often researchers design a new filtered ANN algorithm and also implement it in Weaviate but we are happy when this occurs! PathSeer, now available in Weaviate 1.40, makes that decision dynamically to improve search without query-specific tuning. In our benchmarks, PathSeer was the fastest strategy across almost every workload tested, including all six production-like conjunction filters. It also works with existing HNSW indexes, so you can switch an existing collection without reindexing: "filterStrategy": "pathseer" Read the full breakdown: weaviate.io/blog/pathseer-… 💬 1 🔄 2 ❤️ 6 👀 289 📊 3 ⚡