Built-in LangChain4j Providers

Maven Coordinates

No additional dependencies are required beyond the LangChain4j integration core dependencies.

Content Retriever

Provider key: lc4j-content-retriever.

In LangChain4j RAG, ContentRetriever is the component that takes a user query, retrieves relevant content from an underlying data source, and returns ranked content used to augment the prompt.

In Helidon, this provider creates LangChain4j content retrievers from configuration. If type is not set, Helidon uses the default embedding-store-content-retriever (ContentRetrieverType.EMBEDDING_STORE_CONTENT_RETRIEVER) and wires it using the configured embedding model and embedding store.

In a typical RAG setup (see RAG), a named retriever references:

• an EmbeddingModel (embedding-model)
• an EmbeddingStore<TextSegment> (embedding-store)

Each entry under langchain4j.content-retrievers becomes a named singleton declarative service bean in the Helidon service registry. You can attach it to AI services or agents using @Ai.ContentRetriever("name"), or inject it directly by name.

langchain4j:
  content-retrievers:
    foo-bar-content-retriever:
      provider: lc4j-content-retriever1
      type: embedding-store-content-retriever2
      embedding-store: foo-bar-inmemory-embedding-store3
      embedding-model: foo-bar-embedding-model4
      max-results: 10
      min-score: 0.6

@Ai.Service
@Ai.ChatModel("foo-bar-chat-model")
@Ai.ContentRetriever("foo-bar-content-retriever")1
public interface FooBarExpert {
    String askFoo(String foo);
}

@Service.Singleton
public class RetrieverConsumer {
    RetrieverConsumer(@Service.Named("foo-bar-content-retriever") ContentRetriever retriever) {1
    }
}

Configuration properties:

Configuration options

In-Memory Embedding Store

Provider key: lc4j-in-memory.

In LangChain4j in-memory embedding store integration, InMemoryEmbeddingStore is an in-process vector store implementation suitable for local or lightweight use cases.

In Helidon, this provider creates InMemoryEmbeddingStore<TextSegment> instances from langchain4j.embedding-stores.<name> configuration entries.

Each entry becomes a named singleton declarative service bean in the Helidon service registry. That named embedding store can be referenced by configured content retrievers and can also be injected by name into other service beans.

If from-file is configured, Helidon initializes the store by loading previously persisted embeddings and segments using LangChain4j InMemoryEmbeddingStore.fromFile(...). If from-file is not configured, the store starts empty.

langchain4j:
  embedding-stores:
    foo-bar-inmemory-embedding-store:
      provider: lc4j-in-memory1
      # optional: preload persisted store content
      from-file: "target/foo-bar-inmemory-embedding-store.json"2

  content-retrievers:
    foo-bar-content-retriever:
      provider: lc4j-content-retriever
      embedding-model: foo-bar-embedding-model
      embedding-store: foo-bar-inmemory-embedding-store3

@Service.Singleton
public class EmbeddingStoreLifecycle {
    private final InMemoryEmbeddingStore<TextSegment> store;

    EmbeddingStoreLifecycle(@Service.Named("foo-bar-inmemory-embedding-store")
                            InMemoryEmbeddingStore<TextSegment> store) {
        this.store = store;
    }

    @Service.PreDestroy1
    void persistEmbeddingStore() {
        store.serializeToFile(Path.of("target/foo-bar-inmemory-embedding-store.json"));2
    }
}

Configuration properties:

Configuration options

Additional Information

• LangChain4j Integration