bare-sqlite-vector

SQLite Vector bindings for Bare. Statically embeds the sqlite-vector extension from https://github.com/sqliteai/sqlite-vector and registers it against an existing bare-sqlite connection from https://github.com/holepunchto/bare-sqlite, adding vector encoding, nearest-neighbour search, and quantization-based ANN search to ordinary SQLite tables.

npm i bare-sqlite-vector

Usage

const { DatabaseSync } = require('bare-sqlite')
const vector = require('bare-sqlite-vector')

const db = new DatabaseSync(':memory:')
vector.register(db)

db.exec(`
  CREATE TABLE points (id INTEGER PRIMARY KEY, v BLOB);
  INSERT INTO points (v) VALUES (vector_as_f32('[1.0, 0.0, 0.0, 0.0]'));
  INSERT INTO points (v) VALUES (vector_as_f32('[0.0, 1.0, 0.0, 0.0]'));
`)

db.prepare("SELECT vector_init('points', 'v', 'type=FLOAT32,dimension=4,distance=L2')").get()

const rows = db
  .prepare(
    "SELECT rowid, distance FROM vector_full_scan('points', 'v', vector_as_f32('[0.9, 0.1, 0.0, 0.0]'), 2)"
  )
  .all()

API

vector.register(db)

Register the vector extension's SQL functions and virtual table modules on the given bare-sqlite connection. db must be an open DatabaseSync instance from bare-sqlite.

Registration is per-connection; each new DatabaseSync that needs vector functions must call register again. The call is idempotent: Invoking it more than once on the same connection is a no-op.

Throws DATABASE_NOT_OPEN if db has been closed.

License

Apache-2.0 and Elastic-2.0