Semantic search for Amazon Keyspaces

 Keyspaces table ──change stream──▶ keyvector ──▶ embed ──▶ S3 Vectors index
                                                   │
                         your Sentence-Transformers model (TEI), Bedrock, or any OpenAI-compatible API

 keyvector query "…" ──▶ embed ──▶ nearest rows ──▶ read them back from Keyspaces ──▶ current results

Why teams use it

No migration

Keyspaces stays your source of truth and your application keeps writing to it unchanged. Turn on the table’s change stream and the index follows.

The index stays right

Inserts, updates, deletes and expiries all reach the index. A failed batch is retried, never skipped, and a restart resumes where it stopped. Tested against live AWS.

Your data stays in your account

Runs on your ECS under an IAM role that can read one table and write one index. With a self-hosted model, your text never leaves your VPC.

Pick your model

Any open Sentence-Transformers model on Hugging Face TEI, Amazon Bedrock (Titan, Cohere), or an OpenAI-compatible API.

Pays only for real changes

A row is embedded only when its text changes. Updates to other columns cost nothing, and metadata changes reuse the stored vector.

Open source

Apache-2.0. Read every line before it touches your data, and change it if you need to.

Start here

Why KeyvectorWho it's for, the alternatives, and when it isn't the right fit.

Quick startIndex a table and run your first semantic query.

Deploy to your accountOne CloudFormation stack per table, least-privilege IAM.

Dev notesWhat we learned building it, including what broke against real AWS.