Hybrid search in Spice.ai merges vector similarity (semantic) and full-text BM25 (keyword) results into one ranked output. Both search types run in parallel, and their ranks are fused using Reciprocal Rank Fusion (RRF) for optimal relevance.

You can query this through the /v1/search API or with SQL functions like vector_search() and text_search(). Because results are treated as tables, you can filter, join, and aggregate just like any SQL dataset.

Traditional vector databases like Pinecone, Qdrant, or Weaviate store data in dedicated clusters that you must provision and maintain. Spice.ai takes a unified approach that combines vector, text, and relational search within a single runtime.

You can join vector results with structured data, apply SQL filters, and run inference on top of them, all without moving data or managing multiple systems. For developers, this means a single query layer that delivers the power of a vector database, the flexibility of SQL, and the speed of an accelerated cache.

Use S3 Vectors when you need to store and query embeddings at large scale. It’s ideal for workloads with millions or billions of vectors that don’t need always-on compute. By offloading storage and similarity search to S3 Vectors, you get the scalability and durability of S3 with sub-second queries via transient compute. Spice manages the entire lifecycle: embedding, indexing, filtering, and query orchestration.