Firebolt Implements Native HNSW Vector Search
Firebolt describes native vector search indexing using HNSW and the USearch library to enable approximate nearest neighbor (ANN) queries over embeddings. The company says Similarweb reduced query latency by 100x, from tens of seconds to about 300 ms (99th percentile), enabling sub-second semantic search across hundreds of millions of vectors. Firebolt supports in-memory and disk-backed load strategies and exposes a SQL API for vector_search and index creation.
Key Points
- 1Implements HNSW-based ANN indexes using USearch to support large-scale vector searches.
- 2Reduces query latency dramatically—Similarweb saw 100x improvement to ~300 ms (99th percentile).
- 3Enables sub-second semantic search on hundreds of millions of embeddings within analytics clusters.
Scoring Rationale
Official, practical feature with measurable 100x latency gains; limited novelty beyond established ANN/HNSW techniques compared to prior solutions.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
