@theoven/vector

Vector search for Oven — an embedded scan, pgvector or Qdrant behind one contract.

Part of Oven — the batteries-included framework for Bun.

Install

bun add @theoven/vector

Usage

import { sqliteVector, vector } from '@theoven/vector'

app.use(vector(sqliteVector({ url: './vectors.db' })))

export default async ({ body, vector }) => {
  const embedding = await embed(body.question)
  return vector.query(embedding, { k: 5, filter: { source: 'handbook' } })
}

Defaults to SQLite, so retrieval works with nothing to provision — the same argument as the SQLite database default.

Stores

| | | | --- | --- | | sqliteVector | in this package; scans in process, no server | | @theoven/vector-pg | pgvector, for apps already on Postgres | | @theoven/vector-qdrant | a dedicated vector database |

score is cosine similarity on every one of them — 1 identical, 0 unrelated — so switching store does not invert your thresholds.

The embedded store scans

sqlite-vec cannot be loaded under Bun: the bundled SQLite is built without dynamic extension support. So this compares in JavaScript. Measured at 1536 dimensions:

| vectors | per query | | --- | --- | | 1,000 | 3.4 ms | | 10,000 | 17 ms | | 50,000 | 88 ms | | 200,000 | 359 ms |

Linear, and fine to roughly 50,000. Past that, change the store.

Documentation

https://theoven.app/docs/bricks/vector/

License

MIT