Customizable HTTP Client

Some LangChain4j modules (currently OpenAI and Ollama) support customizing the HTTP clients used to call the LLM provider API.

The langchain4j-http-client module implements an HttpClient SPI, which is used by those modules to call the LLM provider's REST API. This means the underlying HTTP client can be customized, and any other HTTP client can be integrated by implementing the HttpClient SPI.

Currently, there are 2 out-of-the-box implementations:

• JdkHttpClient in the langchain4j-http-client-jdk module. It is used by default when a supported module (e.g., langchain4j-open-ai) is used.
• SpringRestClient in the langchain4j-http-client-spring-restclient. It is used by default when a supported module's Spring Boot starter (e.g., langchain4j-open-ai-spring-boot-starter) is used.

Customizing JDK's HttpClient

HttpClient.Builder httpClientBuilder = HttpClient.newBuilder()
        .sslContext(...);

JdkHttpClientBuilder jdkHttpClientBuilder = JdkHttpClient.builder()
        .httpClientBuilder(httpClientBuilder);

OpenAiChatModel model = OpenAiChatModel.builder()
        .httpClientBuilder(jdkHttpClientBuilder)
        .apiKey(System.getenv("OPENAI_API_KEY"))
        .modelName("gpt-4o-mini")
        .build();

Customizing Spring's RestClient

RestClient.Builder restClientBuilder = RestClient.builder()
        .requestFactory(new HttpComponentsClientHttpRequestFactory());

SpringRestClientBuilder springRestClientBuilder = SpringRestClient.builder()
        .restClientBuilder(restClientBuilder)
        .streamingRequestExecutor(new VirtualThreadTaskExecutor());

OpenAiChatModel model = OpenAiChatModel.builder()
        .httpClientBuilder(springRestClientBuilder)
        .apiKey(System.getenv("OPENAI_API_KEY"))
        .modelName("gpt-4o-mini")
        .build();