Get started

pnpm add @anvia/core @anvia/pgvector pg pgvector

import { retrieveDocuments } from '@anvia/core/vector-store';
import { PgVectorClient } from '@anvia/pgvector';
const storeClient = new PgVectorClient({
    client: pool
});
const store = storeClient.vectorStore({
    tableName: 'support_docs',
    dimensions: 1536,
    metric: "cosine"
});
await store.validate();
await store.upsert({
    documents: documents
});
const results = await retrieveDocuments({ store, model: embeddings, query, topK: 5 });

The table and vector extension must exist before validate(). Use ensure() when this process may create missing infrastructure; both methods check the configured dimensions.

Without client, pass connectionString; the adapter otherwise creates a pg pool using standard environment behavior.

Next

• Collections and indexing
• Search and filters
• Production