Extending JD.AI

JD.AI is built on Microsoft Semantic Kernel, making it extensible through standard SK patterns.

Writing custom tools

Tools are Semantic Kernel plugins with [KernelFunction] attributes:

public class MyCustomTools
{
    [KernelFunction("my_tool")]
    [Description("Description of what my tool does")]
    public string MyTool(
        [Description("Parameter description")] string input)
    {
        return $"Processed: {input}";
    }
}

kernel.Plugins.AddFromObject(new MyCustomTools(), "MyTools");

Adding providers

Implement IProviderDetector:

public class MyProviderDetector : IProviderDetector
{
    public string Name => "MyProvider";
    
    public async Task<ProviderInfo> DetectAsync(CancellationToken ct)
    {
        // Detection logic
        return new ProviderInfo(Name, isAvailable: true, ...);
    }
    
    public IKernelBuilder ConfigureKernel(
        IKernelBuilder builder, 
        ProviderModelInfo model)
    {
        // Configure the kernel with your provider
        return builder;
    }
}

Using Claude Code skills

Create SKILL.md files in ~/.claude/skills/ or .claude/skills/:

---
name: my-skill
description: What this skill does
---

Instructions for the AI agent when this skill is loaded.

Architecture overview

Program.cs
├── ProviderRegistry (detect + build kernels)
├── AgentSession (kernel, chat history, tools)
├── AgentLoop (streaming + tool calling loop)
├── SlashCommandRouter (25+ commands)
├── ChatRenderer (Spectre.Console TUI)
├── InteractiveInput (readline + completions)
├── SessionStore (SQLite persistence)
└── Tools/ (8 plugin categories)

Example: LocalModelDetector

JD.AI's built-in LocalModelDetector is a real-world example of a custom provider. It scans for .gguf model files and wraps them via LLamaSharp:

public class LocalModelDetector : IProviderDetector
{
    public string Name => "Local";

    public async Task<ProviderInfo> DetectAsync(CancellationToken ct)
    {
        var registry = new LocalModelRegistry(modelsDir);
        await registry.LoadAsync(ct);

        var models = registry.GetAll()
            .Select(m => new ProviderModelInfo(m.Id, m.DisplayName))
            .ToList();

        var gpu = GpuDetector.Detect();

        return new ProviderInfo(
            Name,
            isAvailable: models.Count > 0,
            models: models,
            statusDetail: $"{models.Count} model(s) [{gpu.Backend}]");
    }

    public IKernelBuilder ConfigureKernel(
        IKernelBuilder builder,
        ProviderModelInfo model)
    {
        builder.Services.AddSingleton<IChatCompletionService>(
            new LlamaInferenceEngine(model.Id, _registry));
        return builder;
    }
}

See Local Models for the full guide.

Package	NuGet	Description
`JD.SemanticKernel.Extensions`		Skills, hooks, plugins bridge
`JD.SemanticKernel.Connectors.ClaudeCode`		Claude Code auth provider
`JD.SemanticKernel.Connectors.GitHubCopilot`		Copilot auth provider
`JD.SemanticKernel.Connectors.OpenAICodex`		OpenAI Codex auth provider

Contributing

See CONTRIBUTING.md for development setup and guidelines.

Table of Contents