{"id":1664,"date":"2023-12-05T15:14:33","date_gmt":"2023-12-05T23:14:33","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/semantic-kernel\/?p=1664"},"modified":"2024-01-10T15:12:55","modified_gmt":"2024-01-10T23:12:55","slug":"release-candidate-1-for-the-semantic-kernel-net-sdk-is-now-live","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/agent-framework\/release-candidate-1-for-the-semantic-kernel-net-sdk-is-now-live\/","title":{"rendered":"Release Candidate 1 for the Semantic Kernel .NET SDK is now live."},"content":{"rendered":"

Semantic Kernel v1.0<\/a> has shipped and the contents of this blog entry is now out of date.<\/p><\/blockquote>\n

Since the interface is getting extremely close to its final v1.0.0 structure, we\u2019re excited to release v1.0.0 RC1 of the .NET Semantic Kernel SDK. During the next two weeks we\u2019ll be focused on bug fixes and making minor<\/em> adjustments to finish the shape of the API.<\/p>\n

In this blog we\u2019ll share\u2026<\/p>\n

    \n
  1. Just<\/em> how much easier it is to get started.<\/a><\/li>\n
  2. Improvements to the kernel<\/a><\/li>\n
  3. Making function invocation easier<\/a><\/li>\n
  4. Creating and sharing prompts with YAML.<\/a><\/li>\n<\/ol>\n

    Automated function calling makes getting started easy.<\/h2>\n

    With the latest round of updates, we took great care to make the SDK as simple to use for new and existing users. This included renaming many of our classes and interfaces to better align with the rest of the industry and upgrading custom classes to existing .NET implementations.<\/p>\n

    To highlight just how much easier Semantic Kernel has gotten, I want to share what I\u2019m most proud of: our work simplifying function calling with OpenAI. With function calling, the model can tell the program which function should be called next to satisfy a user\u2019s need, but setting up OpenAI function calling has required multiple steps. You had to\u2026<\/p>\n

      \n
    1. Describe your functions<\/li>\n
    2. Call the model<\/li>\n
    3. Review the results to see if a function call request was being made<\/li>\n
    4. Parse the data necessary to make the call<\/li>\n
    5. Perform the operation<\/li>\n
    6. Add the results back to the chat history<\/li>\n
    7. And then start the operation over again\u2026<\/li>\n<\/ol>\n

      With Semantic Kernel, however, we have all<\/em> the information needed to completely automate this entire process, so we\u2019ve done just that. Take, for example, a simple app that allows a user to turn a light bulb on and off with an AI assistant.<\/strong><\/p>\n

      In V1.0.0 RC1, you\u2019ll start by creating your plugins with the [KernelFunction]<\/code> attribute.<\/p>\n

      public class LightPlugin\r\n{\r\n\u00a0\u00a0\u00a0 public bool IsOn { get; set; }\r\n\r\n \u00a0\u00a0 [KernelFunction, Description(\"Gets the state of the light.\")]\r\n \u00a0\u00a0 public string GetState() => IsOn ? \"on\" : \"off\";\r\n\r\n \u00a0\u00a0 [KernelFunction, Description(\"Changes the state of the light.'\")]\r\n \u00a0\u00a0 public string ChangeState(bool newState)\r\n \u00a0\u00a0 {\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 IsOn = newState;\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 var state = GetState();\r\n\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 \/\/ Print the state to the console\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Console.ForegroundColor = ConsoleColor.DarkBlue;\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Console.WriteLine($\"[Light is now {state}]\");\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Console.ResetColor();\r\n\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 return state;\r\n \u00a0\u00a0 }\r\n}\r\n\r\n<\/pre>\n
      You can then easily add your services and plugins to a single kernel.<\/p>\n
      var builder = Kernel.CreateBuilder();\r\nbuilder.Services.AddAzureOpenAIChatCompletion(\"gpt-35-turbo\", \"gpt-3.5-turbo\", endpoint, apiKey)(kernelSettings)\r\nbuilder.Plugins.AddFromType<LightPlugin>();\r\nKernel kernel = builder.Build()<\/pre>\n
      Finally, you can invoke a prompt that uses the new plugin you just authored. This is where the updates to Semantic Kernel really start to shine! Since all the necessary information is stored in the kernel, you can automatically call the registered functions using the AutoInvokeKernelFunctions<\/code> option.<\/p>\n
      \/\/ Enable auto invocation of kernel functions\r\nOpenAIPromptExecutionSettings settings = new()\r\n{\r\n \u00a0\u00a0 FunctionCallBehavior = FunctionCallBehavior.AutoInvokeKernelFunctions\r\n};\r\n\r\n\/\/ Start a chat session\r\nwhile (true)\r\n{\r\n \u00a0\u00a0 \/\/ Get the user's message\r\n \u00a0\u00a0 Console.Write(\"User > \");\r\n \u00a0\u00a0 var userMessage = Console.ReadLine()!;\r\n\r\n \u00a0\u00a0 \/\/ Invoke the kernel\r\n \u00a0\u00a0 var results = await kernel.InvokePromptAsync(userMessage, new(settings));\r\n\r\n \u00a0\u00a0 \/\/ Print the results\r\n \u00a0\u00a0 Console.WriteLine($\"Assistant > {results}\u201d);\r\n}<\/pre>\n
      When you run the program, you can now ask the agent to turn the lights on and off.<\/p>\n
      \"Image<\/a><\/p>\n
      We could make our agent even more efficient, though. To \u201ctoggle\u201d the lights, the AI must currently make two function calls: 1) get the current state and then 2) change the state. This essentially doubles the number of tokens and time required to fulfill the request.<\/p>\n
      \"Image<\/a><\/p>\n
      With Semantic Kernel templates, you can instead pre-emptively provide the LLM with the current state. Notice how we can tell the AI about the current state with a system message below.<\/p>\n
      \/\/ Invoke the kernel\r\nvar results = await kernel.InvokePromptAsync(@$\"\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 <message role=\"\"system\"\">The current state of the light is \"\"{{{{LightPlugin.GetState}}}}\"\"<\/message>\r\n \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 <message role=\"\"user\"\">{userMessage}.<\/message>\",\r\n \u00a0\u00a0 new(settings)\r\n);<\/pre>\n
      Now, when we ask the AI to toggle the lights, it only needs to make a single function call!<\/p>\n
      \"Image<\/a><\/p>\n
      We believe the updates we\u2019ve made to function calling in Semantic Kernel make it much<\/em> easier than what we had before. We also believe it\u2019s much easier than the other popular open-source SDKs available today.<\/p>\n
      This was just a quick overview of what it will look like to build an AI application with Semantic Kernel. To learn more about the changes we made as a team, please continue reading.<\/p>\n
      \n\n\n\n
       <\/p>\n
      \"Image<\/a>\nSergey Menshykh<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
      The kernel is now at the center of everything<\/em>.<\/h2>\n
      In our original blog post for v1.0.0<\/a>, we shared how we wanted to increase the value of the kernel and<\/em> make it easier to use. We believe we\u2019ve done just that by making it the<\/em> property bag for your entire AI application.<\/p>\n