DotKernel

Lightweight, Native AOT–friendly AI kernel for .NET. Register plugins, prompts, and filters with attributes — a simpler alternative to Microsoft Semantic Kernel.

• Docs: https://0use.net/DotKernel/
• Live demo: https://0use.net/DotKernel/demo/
• Repository: https://github.com/0use-TE/DotKernel

Features

• Attribute-driven plugins ([KernelPlugin], [KernelFunction], [KernelPrompt], [KernelProperty])
• Live context: marked properties injected into each model request
• Tool-call filter pipeline with auto / manual approval
• Streaming via InvokeStreamingAsync
• Source generator for trim-safe static registration
• Avalonia cross-platform demo (Desktop + WebAssembly) with digital twin UI

Quick start

git clone https://github.com/0use-TE/DotKernel.git
cd DotKernel
dotnet test
dotnet run --project examples/DotKernel.AvaExample.Desktop

Docs version: v1.0.1 · NuGet: DotKernel 1.0.1

Install

dotnet add package DotKernel --version 1.0.1
dotnet add package Microsoft.Extensions.AI.OpenAI

Use 1.0.1+ (1.0.0 could fail compiling sync [KernelFunction] methods without CancellationToken). See release notes.

Minimal kernel (OpenAI)

using DotKernel;
using Microsoft.Extensions.AI;
using OpenAI;

IChatClient chatClient = new OpenAIClient("sk-...")
    .GetChatClient("gpt-4o-mini")
    .AsIChatClient();

var kernel = KernelBuilder.Create()
    .AddChatClient(chatClient)
    .AddPlugin(new WeatherPlugin())
    .Build();

var answer = await kernel.InvokeAsync("What's the weather in Seattle?");

Full walkthrough: Quick Start.

Configure DeepSeek (optional)

In the Avalonia demo, open API in the header and set Endpoint / Model / API key, then Apply.

Or set user-secrets / env vars (no committed appsettings.json):

dotnet user-secrets set "DeepSeek:ApiKey" "your-key" --project examples/DotKernel.AvaExample
# or: DOTKERNEL_DeepSeek__ApiKey=your-key

Without an API key, the demo uses a built-in Echo client that runs a scripted multi-step tool chain.

Solution layout

src/DotKernel/              Core kernel
src/DotKernel.Generators/   AOT source generator
tests/DotKernel.Tests/      Unit tests
examples/
  DotKernel.Example/        Console demo
  DotKernel.AvaExample/     Shared Avalonia UI + twin plugin
  DotKernel.AvaExample.Desktop/
  DotKernel.AvaExample.Browser/
docs/                       DocFX documentation (v1.0 + zh-CN)

Avalonia demo

Project	Purpose
DotKernel.AvaExample	Shared UI, BuildingTwinPlugin, twin state
DotKernel.AvaExample.Desktop	Desktop app (PublishAot in Release)
DotKernel.AvaExample.Browser	WebAssembly for GitHub Pages

# Desktop
dotnet run --project examples/DotKernel.AvaExample.Desktop

# Browser (local)
dotnet run --project examples/DotKernel.AvaExample.Browser

UI

• Left: streaming chat with Markdown replies; API panel for Endpoint / Model / API key
• Right: digital twin (lighting, conveyor, AGV, robot, inventory, alerts)
• Tool calls are logged in the Call history panel (not in chat bubbles)
• Toggle Auto / Manual for tool approval (default: Auto)
• [KernelProperty] on the twin plugin injects live context each turn

Native AOT publish

.\scripts\publish-aot.ps1

Documentation

Build and preview locally (uses / base path so links work on localhost):

# One-off local preview (GitHub Pages uses /DotKernel/ in docfx.json)
(Get-Content docfx.json -Raw) -replace '"_appBasePath":\s*"/DotKernel/"', '"_appBasePath": "/"' | Set-Content docfx.local.json
docfx docfx.local.json
docfx serve _site --port 8080
# open http://localhost:8080/

dotnet tool update -g docfx
docfx docfx.json
docfx serve _site --port 8080

GitHub Actions publishes docs + WASM demo to GitHub Pages on push to main. Set repository secret DEEPSEEK_API_KEY for the live web demo.

NuGet pack / publish

dotnet pack src/DotKernel/DotKernel.csproj -c Release -o artifacts
# packages: artifacts/DotKernel.1.0.1.nupkg (+ .snupkg)

dotnet nuget push artifacts/DotKernel.1.0.1.nupkg --api-key <YOUR_KEY> --source https://api.nuget.org/v3/index.json

License

MIT (see repository for details).