SvectorDB

SvectorDB is a serverless vector database designed to run on AWS, where you pay per request instead of provisioning capacity. It is aimed at developers who need similarity search for AI apps, scaling from a single vector to around a million.

Items hold a key, a value and a vector, and you can query by a vector or by the key of an existing item to find its nearest neighbours. Hybrid search lets you filter results with Lucene or Elasticsearch-style key-value queries, and upserts and deletions show up immediately with no eventual consistency.

Built-in vectorizers handle text and images, or you can bring your own embeddings. Client libraries exist for JavaScript and Python, an OpenAPI specification covers other languages, and CloudFormation support lets you define databases next to the rest of your AWS infrastructure.