ApertusSharp

ApertusSharp is a modern .NET client for Swis-AI's Apertus LLM — built on Microsoft.Extensions.AI with full support for Semantic Kernel and custom chat pipelines. It’s designed for clean integration, simple usage, and flexible composition in .NET applications.

✨ Features

• ✅ Built on Microsoft.Extensions.AI abstractions
• 🔄 Supports both streaming and non-streaming chat
• 🧠 Semantic Kernel compatible
• 🧩 Extensible for custom chat pipelines and agents
• 🧪 Minimal, testable, simple .NET code
• 🧰 Ready for DI registration and service composition

📦 Installation

Install via NuGet:

dotnet add package ApertusSharp

🚀 Quick Start

List available models:

var apertus = new ApertusClient(apiKey);
var models = await apertus.ListModelsAsync();
Console.WriteLine($"Available models: {string.Join(", ", models.Select(m => m.Id))}");

Create the apertus chat client:

var apertus = new ApertusClient(model:"swiss-ai/apertus-8b-instruct", apiKey:apiKey);

await foreach (var stream in apertus.GenerateAsync("How are you today?"))
	Console.Write(stream.Text);

Use it as service extension:

var services = new ServiceCollection();
services.AddApertusChatClient(apiKey: apiKey, model: "swiss-ai/apertus-8b-instruct");

await using var provider = services.BuildServiceProvider(new ServiceProviderOptions
{
	ValidateScopes = true,
	ValidateOnBuild = true
});
var apertus = provider.GetRequiredService<ApertusClient>();

var messages = new List<ChatMessage>
{
	new ChatMessage(ChatRole.User, "Hello from ServiceCollection!")
};

Console.Write("AI: ");
await foreach (var chunk in apertus.GetStreamingResponseAsync(messages))
{
	Console.Write(chunk);
}
Console.WriteLine("\n Streaming complete.");

🔌 Semantic kernel integration

ApertusSharp can be used as a custom IChatClient for Semantic Kernel:

// Create a Semantic Kernel builder
var builder = Kernel.CreateBuilder();

// Register Apertus as a chat client service
builder.Services.AddApertusChatClient(apiKey: apiKey, model: "swiss-ai/apertus-8b-instruct");

var kernel = builder.Build();

// Use the kernel to get a chat completion
var chat = kernel.GetRequiredService<IChatClient>();

var msg = "How does Semantic Kernel work with Apertus?";
Console.WriteLine("User: " + msg);

var history = new List<ChatMessage>
{
	new ChatMessage(ChatRole.User, msg)
};

var result = await chat.GetResponseAsync(history);

Console.WriteLine("AI: " + result.Text);

Usage in jupyter notebooks

Cell 1 - Install ApertusSharp package:

#r "nuget: ApertusSharp"

Cell 2 - Import namespaces:

using ApertusSharp;
using Microsoft.Extensions.AI;

Cell 3 - Initialize the client:

var apiKey = "your-api-key-here"; // Replace with your actual API key
var apertus = new ApertusClient(
    apiKey: apiKey,
    model: "swiss-ai/apertus-8b-instruct"
);

Cell 4 - Simple chat interaction:

var response = await apertus.GetResponseAsync("Explain quantum computing in simple terms");
Console.WriteLine(response);

Tips for jupyter jotebooks

• Set your API key as an environment variable for security: Environment.GetEnvironmentVariable("APERTUS_TOKEN")
• Use display() function to render rich outputs
• Break complex workflows into multiple cells for better interactivity
• Leverage async/await for responsive notebook experience

🧱 Architecture

• IChatClient and IChatCompletionService abstractions
• Streaming via IAsyncEnumerable<ChatResponseUpdate>
• Extensible options via ApertusChatOptions
• Designed for clean DI and modular composition

📚 Documentation

• PublicAI API
• .NET AI SDK
• Semantic Kernel

🤝 Contributing

Pull requests welcome! Please open an issue first for major changes.

📄 License

MIT — see LICENSE for details.