Assistants for Semantic Kernel

This is assistant support for the Semantic Kernel.

This enables the usage of assistants for the Semantic Kernel without relying on OpenAI Assistant APIs. It runs locally planners and plugins for the assistants.

It provides different scenarios for the usage of assistants such as:

• Assistant with Semantic Kernel plugins
• Multi-Assistant conversation

As the assistants are using the Semantic Kernel, you can use your own model for the assistants and host them locally (see: Bring you own model for more details.).

Usage

1. Create you agent description file in yaml:

   name: Mathematician
   description: A mathematician that resolves given maths problems.
   instructions: |
     You are a mathematician.
     Given a math problem, you must answer it with the best calculation formula.
     No need to show your work, just give the answer to the math problem.
     Use calculation results.
   input_parameter: 
     default_value: ""
     description: |
         The word mathematics problem to solve in 2-3 sentences.
         Make sure to include all the input variables needed along with their values and units otherwise the math function will not be able to solve it.
   execution_settings:
     planner: Handlebars
     prompt_settings: 
       temperature: 0.0
       top_p: 1
       max_tokens: 2000
   

2. Instanciate your assistant in your code:

    string azureOpenAIChatCompletionDeployment = configuration["AzureOpenAIDeploymentName"]!;
    string azureOpenAIEndpoint = configuration["AzureOpenAIEndpoint"]!;
    string azureOpenAIKey = configuration["AzureOpenAIAPIKey"]!;
   
    var mathKernel = Kernel.CreateBuilder()
                                        .AddAzureOpenAIChatCompletion(azureOpenAIChatCompletionDeployment, azureOpenAIEndpoint, azureOpenAIKey)
                                        .Build();
   
    mathKernel.ImportPluginFromObject(new MathPlugin());
   
    var mathematician = AssistantBuilder.FromTemplate("./Assistants/Mathematician.yaml")
                                        .WithKernel(mathKernel)
                                        .Build();
   

   var thread = mathematician.CreateThread();
   await thread.InvokeAsync("Your ask to the assistant.");
   

Bring you own model ?