{"id":4899,"date":"2025-05-28T06:18:24","date_gmt":"2025-05-28T13:18:24","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/semantic-kernel\/?p=4899"},"modified":"2025-05-28T06:18:24","modified_gmt":"2025-05-28T13:18:24","slug":"semantic-kernel-and-microsoft-extensions-ai-better-together-part-2","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/agent-framework\/semantic-kernel-and-microsoft-extensions-ai-better-together-part-2\/","title":{"rendered":"Semantic Kernel and Microsoft.Extensions.AI: Better Together, Part 2"},"content":{"rendered":"

This is Part 2 of our series on integrating Microsoft.Extensions.AI with Semantic Kernel. In\u00a0Part 1<\/a>, we explored the relationship between these technologies and how they complement each other. Now, let’s dive into practical examples showing how to use Microsoft.Extensions.AI abstractions with Semantic Kernel in non-agent scenarios.<\/em><\/p>\n

Getting Started with Microsoft.Extensions.AI and Semantic Kernel<\/h2>\n

Before we dive into examples, let’s understand what we’ll be working with. Microsoft.Extensions.AI provides foundational abstractions like\u00a0IChatClient<\/code>\u00a0and\u00a0IEmbeddingGenerator<string, Embedding<float>><\/code>, while Semantic Kernel builds upon these to provide higher-level functionality like plugins, prompt templates, and workflow automation.<\/p>\n

In this post, we’ll focus on practical scenarios where you can leverage both technologies together for common AI tasks.<\/p>\n

1. Basic Chat Completion with IChatClient<\/h2>\n

Let’s start with the most fundamental scenario: chat completion. Semantic Kernel now natively supports Microsoft.Extensions.AI’s\u00a0IChatClient<\/code>\u00a0interface.<\/p>\n

Using Kernel Builder<\/h3>\n
using<\/span> Microsoft.Extensions.AI;\r\nusing<\/span> Microsoft.SemanticKernel;\r\n\r\n\/\/ Create a kernel with OpenAI chat client<\/span>\r\nvar<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddOpenAIChatClient(\"gpt-4o\"<\/span>, \"your-api-key\"<\/span>)\r\n    .Build();\r\n\r\n\/\/ Simple chat completion<\/span>\r\nvar<\/span> response = await<\/span> kernel.InvokePromptAsync(\"What is the capital of France?\"<\/span>);\r\nConsole.WriteLine(response);\r\n<\/code><\/pre>\n
Using a Chat Client directly with Azure OpenAI<\/h3>\n
var<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddAzureOpenAIChatClient(\r\n        deploymentName: \"gpt-4o\"<\/span>,\r\n        endpoint: \"https:\/\/your-resource.openai.azure.com\/\"<\/span>,\r\n        apiKey: \"your-api-key\"<\/span>)\r\n    .Build();\r\n\r\nvar<\/span> client = kernel.GetRequiredService<IChatClient>();\r\nvar<\/span> response = await<\/span> client.GetResponseAsync([new<\/span>(ChatRole.User, \"Hello, AI!\"<\/span>)]);\r\nConsole.WriteLine(response.Text);\r\n<\/code><\/pre>\n
Using Dependency Injection<\/h3>\n
using<\/span> Microsoft.Extensions.DependencyInjection;\r\nusing<\/span> Microsoft.Extensions.AI;\r\nusing<\/span> Microsoft.SemanticKernel;\r\n\r\nvar<\/span> services = new<\/span> ServiceCollection();\r\n\r\n\/\/ Register the chat client<\/span>\r\nservices.AddOpenAIChatClient(\"gpt-4o\"<\/span>, \"your-api-key\"<\/span>);\r\n\r\n\/\/ Register Semantic Kernel<\/span>\r\nservices.AddKernel();\r\n\r\nvar<\/span> serviceProvider = services.BuildServiceProvider();\r\nvar<\/span> kernel = serviceProvider.GetRequiredService<Kernel>();\r\n\r\nvar<\/span> response = await<\/span> kernel.InvokePromptAsync(\"Tell me about artificial intelligence.\"<\/span>);\r\nConsole.WriteLine(response);\r\n<\/code><\/pre>\n
Converting Between IChatCompletionService and IChatClient<\/h3>\n
Sometimes you might need to work with both interfaces. Semantic Kernel provides seamless conversion:<\/p>\n
using<\/span> Microsoft.Extensions.AI;\r\nusing<\/span> Microsoft.SemanticKernel;\r\nusing<\/span> Microsoft.SemanticKernel.ChatCompletion;\r\n\r\n\/\/ Get the chat completion service<\/span>\r\nvar<\/span> chatService = kernel.GetRequiredService<IChatCompletionService>();\r\n\r\n\/\/ Convert to IChatClient when needed<\/span>\r\nIChatClient chatClient = chatService.AsChatClient();\r\n\r\n\/\/ Or convert back<\/span>\r\nIChatCompletionService backToService = chatClient.AsChatCompletionService();\r\n<\/code><\/pre>\n
2. Embedding Generation with IEmbeddingGenerator<\/h2>\n
Semantic Kernel has moved from its own\u00a0ITextEmbeddingGenerationService<\/code>\u00a0to Microsoft.Extensions.AI’s\u00a0IEmbeddingGenerator<string, Embedding<float>><\/code>. Here’s how to use it:<\/p>\n
Basic Embedding Generation<\/h3>\n
using<\/span> Microsoft.Extensions.AI;\r\nusing<\/span> Microsoft.SemanticKernel;\r\n\r\n#pragma<\/span> warning<\/span> disable SKEXP0010 \/\/ Type is for evaluation<\/span>\r\n\r\nvar<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddOpenAIEmbeddingGenerator(\"text-embedding-ada-002\"<\/span>, \"your-api-key\"<\/span>)\r\n    .Build();\r\n\r\nvar<\/span> embeddingGenerator = kernel.GetRequiredService<IEmbeddingGenerator<string<\/span>, Embedding<float<\/span>>>>();\r\n\r\n\/\/ Generate embeddings<\/span>\r\nvar<\/span> embeddings = await<\/span> embeddingGenerator.GenerateAsync([\r\n    \"Semantic Kernel is a lightweight, open-source development kit.\"<\/span>,\r\n    \"Microsoft.Extensions.AI provides foundational AI abstractions.\"<\/span>\r\n]);\r\n\r\nforeach<\/span> (var<\/span> embedding in<\/span> embeddings)\r\n{\r\n    Console.WriteLine($\"Generated embedding with {embedding.Vector.Length}<\/span> dimensions\"<\/span>);\r\n}\r\n<\/code><\/pre>\n
Working with Azure OpenAI Embeddings<\/h3>\n
using<\/span> Microsoft.Extensions.AI;\r\nusing<\/span> Microsoft.SemanticKernel;\r\n\r\n#pragma<\/span> warning<\/span> disable SKEXP0010<\/span>\r\n\r\nvar<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddAzureOpenAIEmbeddingGenerator(\r\n        deploymentName: \"text-embedding-ada-002\"<\/span>,\r\n        endpoint: \"https:\/\/your-resource.openai.azure.com\/\"<\/span>,\r\n        apiKey: \"your-api-key\"<\/span>)\r\n    .Build();\r\n\r\nvar<\/span> embeddingGenerator = kernel.GetRequiredService<IEmbeddingGenerator<string<\/span>, Embedding<float<\/span>>>>();\r\n\r\n\/\/ Generate embeddings with custom dimensions (if supported by model)<\/span>\r\nvar<\/span> embeddings = await<\/span> embeddingGenerator.GenerateAsync(\r\n    [\"Custom text for embedding\"<\/span>],\r\n    new<\/span> EmbeddingGenerationOptions { Dimensions = 1536<\/span> });\r\n\r\nConsole.WriteLine($\"Generated {embeddings.Count}<\/span> embeddings\"<\/span>);\r\n<\/code><\/pre>\n
3. Function Calling Integration<\/h2>\n
One of the most powerful features is how Semantic Kernel’s function calling integrates with Microsoft.Extensions.AI. Since\u00a0KernelFunction<\/code>\u00a0is now an\u00a0AIFunction<\/code>, the integration is seamless.<\/p>\n
Creating and Using Kernel Functions<\/h3>\n
using<\/span> Microsoft.Extensions.AI;\r\nusing<\/span> Microsoft.SemanticKernel;\r\nusing<\/span> System.ComponentModel;\r\n\r\nvar<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddOpenAIChatClient(\"gpt-4o\"<\/span>, \"your-api-key\"<\/span>)\r\n    .Build();\r\n\r\n\/\/ Import the function as a plugin<\/span>\r\nkernel.ImportPluginFromType<WeatherPlugin>();\r\n\r\n\/\/ Use function calling<\/span>\r\nvar<\/span> settings = new<\/span> PromptExecutionSettings\r\n{\r\n    FunctionChoiceBehavior = FunctionChoiceBehavior.Auto()\r\n};\r\n\r\nvar<\/span> response = await<\/span> kernel.InvokePromptAsync(\r\n    \"What's the weather like in Seattle and what time is it?\"<\/span>,\r\n    new<\/span>(settings));\r\n\r\nConsole.WriteLine(response);\r\n\r\npublic<\/span> class<\/span> WeatherPlugin<\/span>\r\n{\r\n    [KernelFunction, Description(\"Get the current weather for a city\"<\/span>)<\/span>]\r\n    public<\/span> static<\/span> string<\/span> GetWeather<\/span>([Description(\"The city name\"<\/span><\/span>)] string<\/span> city)<\/span>\r\n    {\r\n        return<\/span> $\"The weather in {city}<\/span> is sunny and 72\u00b0F\"<\/span>;\r\n    }\r\n\r\n    [KernelFunction, Description(\"Get the current time\"<\/span>)<\/span>]\r\n    public<\/span> static<\/span> string<\/span> GetCurrentTime<\/span>()<\/span>\r\n    {\r\n        return<\/span> DateTime.Now.ToString(\"yyyy-MM-dd HH:mm:ss\"<\/span>);\r\n    }\r\n}\r\n<\/code><\/pre>\n
Working with KernelFunction Directly<\/h3>\n
var<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddOpenAIChatClient(\"gpt-4o\"<\/span>, \"your-api-key\"<\/span>)\r\n    .Build();\r\n\r\n\/\/ Create a function from a method<\/span>\r\nvar<\/span> weatherFunction = KernelFunctionFactory.CreateFromMethod(\r\n    () => \"Sunny and 75\u00b0F\"<\/span>,\r\n    \"GetWeather\"<\/span>,\r\n    \"Gets the current weather\"<\/span>);\r\n\r\n\/\/ KernelFunction is already an AIFunction, so you can use it directly<\/span>\r\nvar<\/span> chatOptions = new<\/span> ChatOptions\r\n{\r\n    Tools = [weatherFunction], \/\/ KernelFunction works directly as AITool<\/span>\r\n    ToolMode = ChatToolMode.Auto\r\n};\r\n\r\nvar<\/span> chatClient = kernel.GetRequiredService<IChatClient>();\r\nvar<\/span> messages = new<\/span> List<ChatMessage>\r\n{\r\n    new<\/span>(ChatRole.User, \"What's the weather like?\"<\/span>)\r\n};\r\n\r\nvar<\/span> response = await<\/span> chatClient.GetResponseAsync(messages, chatOptions);\r\nConsole.WriteLine(response.Text);\r\n<\/code><\/pre>\n
4. Content Type Conversions<\/h2>\n
Using\u00a0InvokeAsync<T><\/code>\u00a0with Microsoft.Extensions.AI Types<\/h3>\n
var<\/span> kernel = Kernel.CreateBuilder()\r\n    .AddOpenAIChatClient(\"gpt-4o\"<\/span>, \"your-api-key\"<\/span>)\r\n    .Build();\r\n\/\/ Get Microsoft.Extensions.AI ChatResponse directly<\/span>\r\nvar<\/span> chatResponse = await<\/span> kernel.InvokeAsync<ChatResponse>(\r\n    kernel.CreateFunctionFromPrompt(\"Tell me a joke\"<\/span>));\r\n\r\nConsole.WriteLine($\"Model: {chatResponse.ModelId}<\/span>\"<\/span>);\r\nConsole.WriteLine($\"Content: {chatResponse.Text}<\/span>\"<\/span>);\r\n\r\n\/\/ Get List<ChatMessage> for conversation history<\/span>\r\nvar<\/span> message = await<\/span> kernel.InvokeAsync<ChatMessage>(\r\n    kernel.CreateFunctionFromPrompt(\"Start a conversation about AI\"<\/span>));\r\n\r\nConsole.WriteLine($\"Message Role: {message.Role}<\/span>\"<\/span>);\r\nConsole.WriteLine($\"Message Content: {message.Text}<\/span>\"<\/span>);\r\n\r\n\/\/ Get Microsoft.Extensions.AI TextContent directly<\/span>\r\nvar<\/span> textContent = await<\/span> kernel.InvokeAsync<Microsoft.Extensions.AI.TextContent>(\r\n    kernel.CreateFunctionFromPrompt(\"Start a conversation about AI\"<\/span>));\r\n\r\nConsole.WriteLine($\"Text Content: {textContent.Text}<\/span>\"<\/span>);\r\n<\/code><\/pre>\n
5. Service Selection and Dependency Injection<\/h2>\n
Semantic Kernel’s service selection works seamlessly with Microsoft.Extensions.AI services.<\/p>\n
Multiple Chat Providers<\/h3>\n
var<\/span> services = new<\/span> ServiceCollection();\r\n\r\n\/\/ Register multiple chat clients<\/span>\r\nservices.AddOpenAIChatClient(\"gpt-4\"<\/span>, \"openai-key\"<\/span>, serviceId: \"OpenAI\"<\/span>);\r\nservices.AddAzureOpenAIChatClient(\r\n    \"gpt-4\"<\/span>,\r\n    \"https:\/\/your-resource.openai.azure.com\/\"<\/span>,\r\n    \"azure-key\"<\/span>,\r\n    serviceId: \"AzureOpenAI\"<\/span>);\r\n\r\nservices.AddKernel();\r\n\r\nvar<\/span> serviceProvider = services.BuildServiceProvider();\r\nvar<\/span> kernel = serviceProvider.GetRequiredService<Kernel>();\r\n\r\n\/\/ Use specific service<\/span>\r\nvar<\/span> settings = new<\/span> PromptExecutionSettings { ServiceId = \"AzureOpenAI\"<\/span> };\r\nvar<\/span> response = await<\/span> kernel.InvokePromptAsync<ChatResponse>(\r\n    \"Explain machine learning\"<\/span>,\r\n    new<\/span>(settings));\r\n\r\nConsole.WriteLine(\"Model: \"<\/span> + response.ModelId);\r\nConsole.WriteLine(\"Content: \"<\/span> + response.Text);\r\n<\/code><\/pre>\n
Conclusion<\/h2>\n
The integration between Microsoft.Extensions.AI and Semantic Kernel provides a powerful foundation for building AI applications. By leveraging both technologies, you get:<\/p>\n