Skip to content
Amazon S3 Vectors RAG

Vector Operations

API operations

Section titled “API operations”

Five API calls power this demo. The Ingest Lambda
AWS Lambda — serverless compute. This project uses two functions: ingest (RSS → embed → store) and query (search → answer). writes and maintains vectors; the Query Lambda searches them. Both share GetVectors and ListIndexes.

S3 Vectors API — which Lambda calls PutVectors, QueryVectors, GetVectors, DeleteVectors, and ListIndexes

PutVectors

Section titled “PutVectors”

Store one or more vectors with metadata:

s3vectors.put_vectors(
    vectorBucketName=config.vector_bucket,
    indexName=config.vector_index,
    vectors=[
        {
            "key": "article-a1b2c3d4-chunk-000",
            "data": {"float32": embedding},  # 1024 floats
            "metadata": {
                "source": "aws-whats-new-a1b2c3d4.txt",
                "title": "Amazon S3 Vectors is now GA",
                "url": "https://aws.amazon.com/...",
                "published": "2025-06-01T12:00:00Z",
                "feed": "aws-whats-new",
                "chunk": "0",
            },
        }
    ],
)

Key points:

  • key must be unique within the index (used for upsert and retrieval)
  • data must match the index dimension exactly (1024 floats)
  • metadata is arbitrary — store whatever you’ll need at search time

QueryVectors

Section titled “QueryVectors”

Find the most similar vectors to a query:

response = s3vectors.query_vectors(
    vectorBucketName=config.vector_bucket,
    indexName=config.vector_index,
    queryVector={"float32": query_embedding},
    topK=15,
    returnMetadata=True,
)


for vector in response.get("vectors", []):
    key = vector["key"]
    similarity = vector.get("similarity", 0.0)
    metadata = vector.get("metadata", {})
    title = metadata.get("title", "")

Key points:

  • queryVector is the embedded question (same model, same dimension)
  • topK controls how many results to return
  • returnMetadata=True includes stored metadata in results
  • Results are sorted by similarity score (highest first)

GetVectors

Section titled “GetVectors”

Retrieve specific vectors by their keys:

response = s3vectors.get_vectors(
    vectorBucketName=config.vector_bucket,
    indexName=config.vector_index,
    keys=["article-a1b2c3d4-chunk-000"],
    returnMetadata=True,
)

Boto3 client

Section titled “Boto3 client”

The S3 Vectors service has its own boto3 client:

import boto3


s3vectors = boto3.client("s3vectors", region_name="ap-southeast-2")