@theoven/vector-pg

pgvector as a vector store for Oven.

A store for @theoven/vector, so everything on ctx.vector works unchanged — and score stays cosine similarity, where higher is more similar.

bun add @theoven/vector @theoven/vector-pg

import { vector } from '@theoven/vector'
import { pgVector } from '@theoven/vector-pg'

app.use(vector(pgVector({ url: env.string('DATABASE_URL'), dimensions: 1536 })))

dimensions is required because pgvector fixes the vector size when the column is created — unlike the embedded store, which learns it from the first write.

The appeal is one fewer service to operate. The cost is that the index is approximate and has to be created deliberately — see the brick page.

Documentation

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

License

MIT