Package com.sap.ai.sdk.foundationmodels.openai

Class OpenAiClient

java.lang.Object
com.sap.ai.sdk.foundationmodels.openai.OpenAiClient
public final class OpenAiClient extends Object
Client for interacting with OpenAI models.

  • Method Details

    • forModel

      @Nonnull public static OpenAiClient forModel(@Nonnull OpenAiModel foundationModel) throws DeploymentResolutionException
      Create a new OpenAI client for the given foundation model, using the default resource group.
      Parameters:
      foundationModel - the OpenAI model which is deployed.
      Returns:
      a new OpenAI client.
      Throws:
      DeploymentResolutionException - if no deployment for the given model was found in the default resource group.

    • realtimeClient

      @Nonnull public static OpenAiRealtimeClient realtimeClient()
      Creates and configures OpenAI Realtime API client.
      Returns:
      created client.

    • withCustomDestination

      @Nonnull public static OpenAiClient withCustomDestination(@Nonnull com.sap.cloud.sdk.cloudplatform.connectivity.Destination destination)
      Create a new OpenAI client with a custom destination, allowing for a custom resource group or otherwise custom destination. The destination needs to be configured with a URL pointing to an OpenAI model deployment. Typically, such a destination should be obtained using AiCoreService.getInferenceDestination(String).

      Example: 

      
 var destination = new AiCoreService().getInferenceDestination("custom-rg").forModel(GPT_4O);
 OpenAiClient.withCustomDestination(destination);
      Parameters:
      destination - the specific HttpDestination to use.
      Returns:
      a new OpenAI client.
      See Also:

    • withSystemPrompt

      @Nonnull public OpenAiClient withSystemPrompt(@Nonnull String systemPrompt)
      Add a system prompt before user prompts.

      Note: The system prompt is ignored on chat completions invoked with OpenAiChatCompletionPrompt.

      Parameters:
      systemPrompt - the system prompt.
      Returns:
      the client.

    • withHeader

      @Nonnull public OpenAiClient withHeader(@Nonnull String key, @Nonnull String value)
      Create a new OpenAI client with a custom header added to every call made with this client.
      Parameters:
      key - the key of the custom header to add.
      value - the value of the custom header to add.
      Returns:
      a new client.
      Since:
      1.11.0

    • withHeaders

      @Nonnull public OpenAiClient withHeaders(@Nonnull Map<String,String> headers)
      Create a new openAI client with multiple custom headers added to every call made with this client.
      Parameters:
      headers - a map of key value pairs for the custom headers to add.
      Returns:
      a new client.
      Since:
      1.22.0

    • chatCompletion

      @Nonnull public OpenAiChatCompletionResponse chatCompletion(@Nonnull OpenAiChatCompletionRequest request) throws OpenAiClientException
      Generate a completion for the given conversation and request parameters.
      Parameters:
      request - the completion request.
      Returns:
      the completion output.
      Throws:
      OpenAiClientException - if the request fails.
      Since:
      1.4.0

    • chatCompletion

      @Nonnull public CreateChatCompletionResponse chatCompletion(@Nonnull CreateChatCompletionRequest request) throws OpenAiClientException
      Generate a completion for the given low-level request object.
      Parameters:
      request - the completion request.
      Returns:
      the completion output.
      Throws:
      OpenAiClientException - if the request fails.
      Since:
      1.4.0

    • streamChatCompletion

      @Nonnull public Stream<String> streamChatCompletion(@Nonnull String prompt) throws OpenAiClientException
      Stream a completion for the given string prompt as user.

      Returns a lazily populated stream of text chunks. To access more details about the individual chunks, use streamChatCompletionDeltas(OpenAiChatCompletionRequest).

      The stream should be consumed using a try-with-resources block to ensure that the underlying HTTP connection is closed.

      Example: 

      
 try (var stream = client.streamChatCompletion("...")) {
       stream.forEach(System.out::println);
 }

      Please keep in mind that using a terminal stream operation like Stream.forEach(java.util.function.Consumer<? super T>) will block until all chunks are consumed. Also, for obvious reasons, invoking BaseStream.parallel() on this stream is not supported.

      Parameters:
      prompt - a text message.
      Returns:
      a stream of text chunks.
      Throws:
      OpenAiClientException - if the request fails or if the finish reason is content_filter.
      See Also:

    • streamChatCompletionDeltas

      @Nonnull public Stream<OpenAiChatCompletionDelta> streamChatCompletionDeltas(@Nonnull OpenAiChatCompletionRequest request) throws OpenAiClientException
      Stream a completion for the given conversation and request parameters.

      Returns a lazily populated stream of delta objects. To simply stream the text chunks use streamChatCompletion(String)

      The stream should be consumed using a try-with-resources block to ensure that the underlying HTTP connection is closed.

      Example: 

      
 try (var stream = client.streamChatCompletionDeltas(prompt)) {
       stream
           .peek(delta -> System.out.println(delta.getUsage()))
           .map(OpenAiChatCompletionDelta::getDeltaContent)
           .forEach(System.out::println);
 }

      Please keep in mind that using a terminal stream operation like Stream.forEach(java.util.function.Consumer<? super T>) will block until all chunks are consumed. Also, for obvious reasons, invoking BaseStream.parallel() on this stream is not supported.

      Parameters:
      request - the prompt, including a list of messages.
      Returns:
      a stream of message deltas.
      Throws:
      OpenAiClientException - if the request fails or if the finish reason is content_filter.
      Since:
      1.4.0
      See Also:

    • streamChatCompletionDeltas

      @Nonnull public Stream<OpenAiChatCompletionDelta> streamChatCompletionDeltas(@Nonnull CreateChatCompletionRequest request) throws OpenAiClientException
      Stream a completion for the given low-level request object. Returns a lazily populated stream of delta objects.
      Parameters:
      request - the completion request.
      Returns:
      a stream of message deltas.
      Throws:
      OpenAiClientException - if the request fails or if the finish reason is content_filter.
      Since:
      1.4.0
      See Also:

    • embedding

      @Nonnull public OpenAiEmbeddingResponse embedding(@Nonnull OpenAiEmbeddingRequest request) throws OpenAiClientException
      Get a vector representation of a given request that can be easily consumed by machine learning models and algorithms using high-level request object.
      Parameters:
      request - the request with input text.
      Returns:
      the embedding response convenience object.
      Throws:
      OpenAiClientException - if the request fails.
      Since:
      1.4.0
      See Also:

    • embedding

      @Nonnull public EmbeddingsCreate200Response embedding(@Nonnull EmbeddingsCreateRequest request) throws OpenAiClientException
      Get a vector representation of a given inputs using low-level request.
      Parameters:
      request - the request with input text.
      Returns:
      the embedding output.
      Throws:
      OpenAiClientException - if the request fails.
      Since:
      1.4.0
      See Also: