Skip to content

S3 Vectors Concepts

Amazon S3 Vectors
Amazon S3 Vectors — a purpose-built vector storage capability within S3 that enables similarity search over embeddings without a separate vector database.
is a purpose-built vector storage capability within S3. It enables similarity search over embeddings without managing a separate vector database (no Pinecone, no pgvector, no OpenSearch).

S3 Vectors core concepts — vector buckets, indexes, dimensions, and distance metrics

A Vector Bucket
A specialized S3 bucket type (aws_s3vectors_vector_bucket) that hosts vector indexes for similarity search.
is a specialized S3 bucket type that hosts vector indexes. Think of it as the “database” that contains your “tables” (indexes).

  • Created via aws_s3vectors_vector_bucket Terraform resource
  • Named with a globally unique name (this demo uses a random prefix)
  • Contains one or more vector indexes

A Vector Index
A named index (aws_s3vectors_index) within a vector bucket defining dimension, distance metric, and data type.
is a named index within a vector bucket. It defines how vectors are stored and searched:

PropertyValue in this demoMeaning
Dimension1024Number of floats per vector. Must match embedding model output exactly.
Distance metriccosine

Measures angle between vectors. Best for text embeddings (direction matters more than magnitude).

Data typefloat3232-bit floating point precision for each value.

Each stored vector consists of:

  • Key — A unique identifier (e.g., article-a1b2c3d4-chunk-000)
  • Data — The float array (1024 values for this demo)
  • Metadata — Arbitrary key-value pairs (title, URL, date, feed, etc.)

Metadata is returned with search results, eliminating the need for a separate database lookup.

S3 Vectors compared to traditional vector databases — infrastructure, provisioning, auth, scaling, and pricing