Adding AI services to Semantic Kernel

One of the main features of Semantic Kernel is its ability to add different AI services to the kernel. This allows you to easily swap out different AI services to compare their performance and to leverage the best model for your needs. In this article, we will provide sample code for adding different AI services to the kernel.

If you want to see any of these samples in a complete solution, you can check them out in the public documentation repository.

Language	Link to final solution
C#	Open example in GitHub
Java	Open example in GitHub
Python	Open solution in GitHub

Azure OpenAI

With Azure OpenAI, you can deploy most of the OpenAI models to the cloud and use them in your Semantic Kernel project. Depending on the model that you want to use, however, you will either need to add the model as a text generation service or as a chat completion service.

The following table shows which service you should use for each model.

Model type	Model
Text generation	text-ada-001
Text generation	text-babbage-001
Text generation	text-curie-001
Text generation	text-davinci-001
Text generation	text-davinci-002
Text generation	text-davinci-003
Chat Completion	gpt-3.5-turbo
Chat Completion	gpt-4

Chat completion deployments

C#

To add an Azure OpenAI chat completion service to your Semantic Kernel project, you will need to use the AddAzureChatCompletionService method.

Kernel kernel = Kernel.CreateBuilder()
                      .AddAzureOpenAIChatCompletion(modelId, endpoint, apiKey)
                      .Build();

Java

To add an Azure OpenAI chat completion service to your Semantic Kernel project, you will need to create a ChatCompletionService.

ChatCompletionService chatCompletionService = ChatCompletionService.builder()
    .withModelId(CHAT_MODEL_ID)
    .withOpenAIAsyncClient(client)
    .build();

Then use the withAIService method of the Kernel.Builder to add the service to the kernel.

Kernel kernel = Kernel.builder()
    .withAIService(ChatCompletionService.class, chatCompletionService)
    .withAIService(TextGenerationService.class, textGenerationService)
    .build();

Python

To add an Azure OpenAI chat completion service to your Semantic Kernel project, you will need to use the AzureChatCompletion class.

kernel.add_service(
    AzureChatCompletion(
        service_id=service_id,
        deployment_name=deployment_name,
        endpoint=endpoint,
        api_key=api_key,
    ),
)

Text generation deployments

C#

To add an Azure OpenAI text generation service to your Semantic Kernel project, you will need to use the AddAzureTextCompletionService method.

kernel = Kernel.CreateBuilder()
               .AddAzureOpenAITextGeneration(textModelId, endpoint, apiKey)
               .Build();

Java

To add an Azure OpenAI text generation service to your Semantic Kernel project, you will need to create a TextGenerationService.

TextGenerationService textGenerationService = TextGenerationService.builder()
    .withModelId(TEXT_MODEL_ID)
    .withOpenAIAsyncClient(client)
    .build();

Then use the withAIService method of the Kernel.Builder to add the service to the kernel.

Kernel kernel = Kernel.builder()
    .withAIService(ChatCompletionService.class, chatCompletionService)
    .withAIService(TextGenerationService.class, textGenerationService)
    .build();

Python

To add an Azure OpenAI text generation service to your Semantic Kernel project, you will need to use the AzureTextCompletion class.

kernel.add_service(
    AzureTextCompletion(
        service_id=service_id,
        deployment_name=deployment_name,
        endpoint=endpoint,
        api_key=api_key,
    ),
)

OpenAI

Similar to Azure OpenAI, depending on the model that you want to use from OpenAI, you will either need to add the model as a text generation service or as a chat completion service.

The following table shows which service you should use for each model.

Model type	Model
Text generation	text-ada-001
Text generation	text-babbage-001
Text generation	text-curie-001
Text generation	text-davinci-001
Text generation	text-davinci-002
Text generation	text-davinci-003
Chat Completion	gpt-3.5-turbo
Chat Completion	gpt-4

Chat completion models

C#

To add an OpenAI text generation service to your Semantic Kernel project, you will need to use the AddOpenAIChatCompletionService method.

kernel = Kernel.CreateBuilder()
               .AddOpenAIChatCompletion(openAImodelId, openAIapiKey)
               .Build();

Java

To add an Azure OpenAI chat completion service to your Semantic Kernel project, you will need to create a ChatCompletionService.

ChatCompletionService chatCompletionService = ChatCompletionService.builder()
    .withModelId(CHAT_MODEL_ID)
    .withOpenAIAsyncClient(client)
    .build();

Then use the withAIService method of the Kernel.Builder to add the service to the kernel.

Kernel kernel = Kernel.builder()
    .withAIService(ChatCompletionService.class, chatCompletionService)
    .withAIService(TextGenerationService.class, textGenerationService)
    .build();

Python

To add an OpenAI text generation service to your Semantic Kernel project, you will need to use the OpenAIChatCompletion class.

kernel.add_service(
    OpenAIChatCompletion(
        service_id=service_id,
        ai_model_id=model_id,
        api_key=api_key,
        org_id=org_id,
    ),
)

Text generation models

C#

To add an OpenAI text generation service to your Semantic Kernel project, you will need to use the AddOpenAITextCompletionService method.

kernel = Kernel.CreateBuilder()
               .AddOpenAITextGeneration(openAItextModelId, openAIapiKey)
               .Build();

Java

To add an Azure OpenAI text generation service to your Semantic Kernel project, you will need to create a TextGenerationService.

TextGenerationService textGenerationService = TextGenerationService.builder()
    .withModelId(TEXT_MODEL_ID)
    .withOpenAIAsyncClient(client)
    .build();

Then use the withAIService method of the Kernel.Builder to add the service to the kernel.

Kernel kernel = Kernel.builder()
    .withAIService(ChatCompletionService.class, chatCompletionService)
    .withAIService(TextGenerationService.class, textGenerationService)
    .build();

Python

To add an OpenAI text generation service to your Semantic Kernel project, you will need to use the OpenAITextCompletion class.

kernel.add_service(
    OpenAITextCompletion(
        service_id=service_id,
        ai_model_id=model_id,
        api_key=api_key,
        org_id=org_id,
    ),
)

Next steps

Now that you know how to add different AI services to your Semantic Kernel project, you can learn now to add telemetry and logging to the kernel.

Add telemetry and logging