Get started

pnpm add @anvia/core @anvia/milvus @zilliz/milvus2-sdk-node

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

Documents must already contain dense embeddings. Without a client, the adapter connects to localhost:19530; inject a configured client for any remote deployment.

ensure() creates and loads a missing collection, while validate() requires the collection to exist and its vector field to match the configured dimensions, then loads it. The metric is not compared against the existing collection.

Next

• Collections and indexing
• Search and filters
• Production