Native Vector Search for SQLite
No extensions needed.
Jökull Sólberg, CTO
TripToJapan
"Forget vector databases. It's just a column type.
It's great being able to index and query Vector distances alongside other data in the same table as opposed to dealing with sync and drift between databases."
Intuitive SQL
For AI applications
- Embeddings are just a column.
- Zero setup, no extensions - it just works.
- When you insert a new row the index is updated automatically. You can query the table and embeddings using SQL.
- You can efficiently do approximate nearest neighbor search with SQL functions.
- You can also do exact neighbor search without needing an index.
import { createClient } from '@libsql/client';
const db = createClient({
url: 'file:local.db',
});
await db.batch([
"CREATE TABLE movies (title, year, emb F32_BLOB(3))",
"CREATE INDEX movies_idx ON movies (
libsql_vector_idx(emb, 'metric=cosine')
)",
"INSERT INTO movies
VALUES ('Napoleon', 2023, vector('[1,2,3]'))"
]);
await db.execute(
"SELECT title, year
FROM vector_top_k('movies_idx', vector('[4,5,6]'), 3)
JOIN movies
ON movies.rowid = id;"
);- It’s just a file, and vectors are now just another column type.
- Works on all platforms: mobile, desktop, web, embedded systems, servers, etc.
- RAG now available using only SQLite.
- Efficient resource usage with low memory footprint.
- No need for separate vector databases.
- Combine relational and vector data for richer more powerful queries.
- It works everywhere, via a local file or over http
- Extremely cost efficient
- Natively serverless
- Built for low latency
- Read from local embedded replicas
- Effortless multi region edge replication
