Milvus

The Milvus plugin registers retrievers and indexers backed by a Milvus server (v2 REST API) for Retrieval-Augmented Generation.

Installation

<dependency>
    <groupId>com.google.genkit</groupId>
    <artifactId>genkit-plugin-milvus</artifactId>
    <version>1.0.0-SNAPSHOT</version>
</dependency>

Requirements

• A running Milvus server. For local development:

  curl -sfL https://raw.githubusercontent.com/milvus-io/milvus/master/scripts/standalone_embed.sh -o standalone_embed.sh
  bash standalone_embed.sh start

• Java 21+
• An embedder (e.g. from the Google GenAI plugin)

Usage

MilvusPlugin registers a retriever and indexer named milvus/<collectionName> for each configured collection.

import com.google.genkit.plugins.milvus.MilvusCollectionConfig;
import com.google.genkit.plugins.milvus.MilvusPlugin;

Genkit genkit = Genkit.builder()
    .plugin(GoogleGenAIPlugin.create(apiKey))
    .plugin(
        MilvusPlugin.builder()
            .url("http://localhost:19530")   // default; serves both gRPC and REST
            .token(System.getenv("MILVUS_TOKEN")) // optional; required for auth-enabled servers / Zilliz Cloud
            .addCollection(
                MilvusCollectionConfig.builder()
                    .collectionName("films")
                    .embedderName("googleai/gemini-embedding-001")
                    .metric(MilvusCollectionConfig.Metric.COSINE)
                    .createCollectionIfNotExists(true) // default
                    .build())
            .build())
    .build();

// Index and retrieve
genkit.index("milvus/films", documents);
List<Document> results = genkit.retrieve("milvus/films", "a Christopher Nolan sci-fi film");

Configuration

Tune per-collection settings with MilvusCollectionConfig:

• collectionName — the Milvus collection name (required)
• embedderName — the embedder used to vectorize documents and queries (required)
• dimension — the embedding dimension used when creating the collection (default 768). The store also probes the embedder on first use and creates the collection with the model’s actual output dimension, so this is only a fallback.
• metric — COSINE (default), L2, or INNER_PRODUCT
• createCollectionIfNotExists — auto-create the collection (default true)
• addAdditionalMetadata(key, value) — metadata merged into every indexed document

Collections are created in Milvus “quick setup” mode (auto id primary key, a vector field, and dynamic fields). The document text is stored under text and its metadata as a JSON string under metadata.

Sample

See the milvus sample.