Using Semantic Kernel in an ABP Framework

This guide demonstrates how to integrate Microsoft's Semantic Kernel SDK with the ABP.io framework. It includes a wrapper implementation to help you create AI-powered plugins for your ABP application services in a clean, modular way.

🏗️ Architecture Overview

The solution consists of two primary modules:

1. WafiOpenAISemanticKernelModule
   Core module that provides integration with Semantic Kernel.

2. WafiSmartHRAIPluginModule
   Sample implementation module demonstrating how to create AI plugins for business entities.

⚙️ Implementation Steps

Follow these steps to integrate Semantic Kernel into your ABP application.

🔹 Step 1: Add the Package to Your HttpApi Project

1. Add the Semantic Kernel wrapper package:

dotnet add YourApp.HttpApi.csproj package Wafi.Abp.OpenAISemanticKernel

2. Add the module dependency and configure Semantic Kernel options:
   This connects your ABP module with the Semantic Kernel framework and sets up OpenAI credentials.

[DependsOn(
    // your existing dependencies
    typeof(WafiOpenAISemanticKernelModule)
)]
public class YourAppHttpApiModule : AbpModule
{
    public override void ConfigureServices(ServiceConfigurationContext context)
    {
        var configuration = context.Services.GetConfiguration();

        Configure<WafiOpenAISemanticKernelOptions>(options =>
        {
            options.ModelId = configuration.GetValue<string>("SemanticKernel:OpenAI:ModelId");
            options.ApiKey = configuration.GetValue<string>("SemanticKernel:OpenAI:ApiKey");
        });
    }
}

3. Add the following configuration to your appsettings.json:

{
  "SemanticKernel": {
    "OpenAI": {
      "ApiKey": "your-openai-key",
      "ModelId": "gpt-4"
    }
  }
}

✅ This exposes the /api/OpenAISemanticKernel/ai/ask endpoint automatically, which the chat interface will use to interact with Semantic Kernel.

🔹 Step 2: Implement a Chat Interface

In your Web project, create a chat interface to interact with the /ai/ask API. You can refer to the host/Wafi.SmartHR.Web project for a sample implementation.

🔹 Step 3: Build Plugins to Connect ABP Services with Semantic Kernel

1. Add dependencies to the plugin project:

dotnet add package Wafi.Abp.OpenAISemanticKernel
dotnet add reference ../YourApp.Application/YourApp.Application.csproj

Add this to your plugin module class:

[DependsOn(typeof(YourAppApplicationModule))]
public class YourAppAIPluginModule : AbpModule
{
    // ...
}

2. Create AI Plugins for Your Entities

Each plugin class should clearly describe what the AI will get when calling its methods.

📄 Example: YourEntityPlugin.cs

public class YourEntityPlugin : ApplicationService, ITransientDependency
{
    private readonly IYourEntityAppService _entityService;
    private readonly IAuthorizationService _authorizationService;

    public YourEntityPlugin(
        IYourEntityAppService entityService,
        IAuthorizationService authorizationService)
    {
        _entityService = entityService;
        _authorizationService = authorizationService;
    }

    [KernelFunction]
    [Description("Search for entities by name or any relevant property using a keyword or phrase")]
    public async Task<string> SearchEntitiesAsync(string filter)
    {
        if (!await _authorizationService.IsGrantedAsync(YourPermissions.Default))
        {
            return "You are not authorized to access this data";
        }

        var entities = await _entityService.GetListAsync(filter);
        
        if (entities is null || entities.Items.Count == 0)
        {
            return $"No entities found for filter '{filter}'";
        }

        return JsonSerializer.Serialize(entities);
    }
}

🧠 Be sure to give a meaningful [Description] so that the AI knows exactly what this method does.

3. Create a Plugin Provider

Create a provider class to register your plugin with Semantic Kernel. This is a crucial step that makes your plugin discoverable by the AI system.

using Wafi.Abp.OpenAISemanticKernel.Plugins;
using YourApp.AI.Plugin.YourEntities;

namespace YourApp.AI.Plugin
{
    public class YourEntityPluginProvider : SemanticPluginProviderBase<YourEntityPlugin>, IWafiPluginProvider
    {
        public YourEntityPluginProvider(YourEntityPlugin plugin) : base(plugin) { }

        public override string Name => "Your Plugin Name";
    }
}

🔐 Authorization

All requests are protected by ABP's permission system. If a user doesn't have the required permissions, they will receive an appropriate message:

⚠️ You are not authorized to access this data.

The authorization check is performed in each plugin method:

if (!await _authorizationService.IsGrantedAsync(YourPermissions.Default))
{
    return "You are not authorized to access this data";
}

✅ Final Outcome

After completing these steps, your ABP application will have a fully functioning AI assistant powered by Semantic Kernel.

The AI interface can be accessed through the /askai endpoint in your web application. When users send questions to this endpoint, the system:

1. Processes the natural language query
2. Determines which registered plugins can fulfill the request
3. Calls the appropriate application services through the plugins
4. Returns formatted results to the user

Example 1: Employee List Query

Request:

POST /askai
Content-Type: application/json

{
    "question": "Give me the list of employees"
}

Response:

Here is the list of employees: John Doe - Email: john.doe@example.com - Phone: 1234567890 - Date of Birth: January 1, 1990 - Joining Date: January 1, 2020 - Total Leave Days: 20 - Remaining Leave Days: 20. Jane Smith - Email: jane.smith@example.com - Phone: 0987654321 - Date of Birth: February 1, 1991 - Joining Date: February 1, 2021 - Total Leave Days: 20 - Remaining Leave Days: 15

Example 2: Leave Records Query

Request:

POST /askai
Content-Type: application/json

{
    "question": "Give me the list of employee leave records"
}

Response:

Here is the list of employee leave records: John Doe - Annual leave from January 1 to January 5, 2025 (5 days) - Sick leave from March 15 to March 16, 2025 (2 days)\n - Personal leave from June 1 to June 3, 2025 (3 days) Jane Smith - Sick leave from February 1 to February 3, 2025 (3 days) - Annual leave from July 1 to July 10, 2025 (10 days)"

📚 Additional Resources

For a deeper understanding of the module's implementation, architecture decisions, and advanced customization options, see:

• Microsoft Semantic Kernel Documentation
• ABP Framework Documentation
• Full Module Source Code