Supabase has introduced Vector Buckets, specialized storage containers designed for vector data, expanding storage options beyond pgvector and S3 Vectors. This new feature combines the durability and cost-efficiency of Amazon S3 with built-in similarity search, allowing developers to store large volumes of vectors (up to tens of millions) while maintaining performance. Hot, latency-sensitive vectors are recommended for pgvector, while warm or cold vectors can be stored in Vector Buckets. Architecturally, Vector Buckets are based on S3-compatible storage and support multiple indexes per bucket. The workflow involves creating buckets and indexes, storing vectors with embeddings and metadata, and querying using similarity search. Ideal use cases include semantic search, recommendation systems, and Retrieval-Augmented Generation (RAG). Vector Buckets are currently free during Public Alpha, with egress costs applying.
مشاهده مطلب اصلی