AutoGen.SemanticKernel provides built-in support for ChatCompletionAgent via SemanticKernelChatCompletionAgent. By default the SemanticKernelChatCompletionAgent only supports the original ChatMessageContent type via IMessage<ChatMessageContent>. To support more AutoGen built-in message types like TextMessage, ImageMessage, MultiModalMessage, you can register the agent with SemanticKernelChatMessageContentConnector. The SemanticKernelChatMessageContentConnector will convert the message from AutoGen built-in message types to ChatMessageContent and vice versa.

The following step-by-step example shows how to create an SemanticKernelChatCompletionAgent and chat with it:

Note

You can find the complete sample code here.

Step 1: add using statement

using AutoGen.Core;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Agents;

Step 2: create kernel

var openAIKey = Environment.GetEnvironmentVariable("OPENAI_API_KEY") ?? throw new Exception("Please set OPENAI_API_KEY environment variable.");
var modelId = "gpt-3.5-turbo";
var kernel = Kernel.CreateBuilder()
    .AddOpenAIChatCompletion(modelId: modelId, apiKey: openAIKey)
    .Build();

Step 3: create ChatCompletionAgent

// The built-in ChatCompletionAgent from semantic kernel.
var chatAgent = new ChatCompletionAgent()
{
    Kernel = kernel,
    Name = "assistant",
    Description = "You are a helpful AI assistant",
};

Step 4: create SemanticKernelChatCompletionAgent

In this step, we create an SemanticKernelChatCompletionAgent and register it with SemanticKernelChatMessageContentConnector. The SemanticKernelChatMessageContentConnector will convert the message from AutoGen built-in message types to ChatMessageContent and vice versa.

var messageConnector = new SemanticKernelChatMessageContentConnector();
var skAgent = new SemanticKernelChatCompletionAgent(chatAgent)
    .RegisterMiddleware(messageConnector) // register message connector so it support AutoGen built-in message types like TextMessage.
    .RegisterPrintMessage(); // pretty print the message to the console

Step 5: chat with SemanticKernelChatCompletionAgent

await skAgent.SendAsync("Hey tell me a long tedious joke");