Google Gen AI Embeddings (Experimental)

https://github.com/googleapis/java-genai

This integration uses the official Google Gen AI SDK for Java (com.google.genai:google-genai). It is marked Experimental: the API and implementation may change in future releases.

Maven Dependency

xml

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-google-genai</artifactId>
    <version>1.19.0-beta29</version>
</dependency>

API Key

Get an API key for free here: https://ai.google.dev/gemini-api/docs/api-key .

Models available

See the available embedding models, for example:

gemini-embedding-001 — text-only; supports task types and output dimensionality (128–3072).
gemini-embedding-2 — natively multimodal; does not use the task type parameter (see Google AI Gemini Embeddings for how task instructions work with Gemini Embedding 2).

GoogleGenAiEmbeddingModel

Basic Usage

java

EmbeddingModel embeddingModel = GoogleGenAiEmbeddingModel.builder()
    .apiKey(System.getenv("GOOGLE_AI_GEMINI_API_KEY"))
    .modelName("gemini-embedding-001")
    .build();

Response<Embedding> response = embeddingModel.embed("Hello, world!");
Embedding embedding = response.content();

Configuring the Embedding Model

java

EmbeddingModel embeddingModel = GoogleGenAiEmbeddingModel.builder()
    .apiKey(System.getenv("GOOGLE_AI_GEMINI_API_KEY"))
    .modelName("gemini-embedding-001")
    .taskType(GoogleGenAiEmbeddingModel.TaskTypeEnum.RETRIEVAL_DOCUMENT) // default task type
    .outputDimensionality(768)      // reduce the embedding size (for models that support it)
    .titleMetadataKey("title")      // metadata key used as the document title for RETRIEVAL_DOCUMENT
    .maxRetries(3)
    .timeout(Duration.ofSeconds(30))
    .build();

Request/response API and capabilities

Besides the convenience methods and the builder-level taskType(...), GoogleGenAiEmbeddingModel supports the request/response API with per-call parameters:

Input type: EmbeddingInputType.QUERY / DOCUMENT is mapped to the SDK's RETRIEVAL_QUERY / RETRIEVAL_DOCUMENT task type, so you can embed queries and documents differently without configuring two model instances. (This applies to models that support task types, such as gemini-embedding-001.)
Dimensions: a per-call dimensions(...) overrides the builder's outputDimensionality, for models that support reducing the output size.
Multimodal (gemini-embedding-2): natively embeds interleaved text + image into a single embedding. Earlier models (e.g. gemini-embedding-001) are text-only. Images must be provided as base64 (ImageContent).
Listeners: configure via GoogleGenAiEmbeddingModel.builder().listeners(...) to observe requests, responses, and errors.

java

EmbeddingResponse response = embeddingModel.embed(EmbeddingRequest.builder()
    .input("What is the capital of France?")
    .inputType(EmbeddingInputType.QUERY) // embed as a query
    .dimensions(256)                     // reduce output dimensionality
    .build());

List<Embedding> embeddings = response.embeddings();

Multimodal example (Gemini Embedding 2 — text and image fused into one embedding):

java

EmbeddingModel embeddingModel = GoogleGenAiEmbeddingModel.builder()
    .apiKey(System.getenv("GOOGLE_AI_GEMINI_API_KEY"))
    .modelName("gemini-embedding-2")
    .build();

EmbeddingResponse response = embeddingModel.embed(EmbeddingRequest.builder()
    .input(TextContent.from("a photo of a cat"), ImageContent.from(base64Image, "image/png"))
    .build());

Embedding embedding = response.embeddings().get(0);

See Embedding Model for the request/response API, and Observability for listeners.

Learn more

For more details on the Gemini embedding models, see the documentation.