Vector Buckets

Vector Buckets provide specialized S3-backed storage for vector embeddings with built-in similarity search. Store tens of millions of vectors per index with cosine, Euclidean, or L2 distance metrics.

Key benefits

1. Massive scale: Store tens of millions of vectors per index.
2. Built-in similarity search: Query vectors using cosine, Euclidean, or L2 distance.
3. Metadata filtering: Filter search results by associated metadata.
4. Batch operations: Process up to 500 vectors per request.
5. S3 reliability: Built on S3-compatible storage infrastructure.
6. Complementary to pgvector: Use both for different use cases.

When to use Vector Buckets vs pgvector

Use pgvector for lowest latency, core relational model vectors, transactional guarantees, and small to medium datasets. Use Vector Buckets for millions of vectors, S3-style durability, AI-heavy apps (RAG, semantic search), and separate storage tiers.

Hybrid approach

Keep hot vectors in pgvector, archive large collections in Vector Buckets, and query both from Postgres via Foreign Data Wrapper.

Vector Buckets are valuable for:

• Semantic search at scale
• Recommendation systems
• RAG (Retrieval-Augmented Generation)
• Image and video similarity search
• Large-scale embedding archives

Vector Buckets provide the scalable vector storage required for AI-powered applications.