LM-Kit.NET Microsoft.Extensions.AI Integration

This package provides Microsoft.Extensions.AI integration for LM-Kit.NET, enabling local LLM inference through the standard IChatClient and IEmbeddingGenerator abstractions.

Installation

dotnet add package LM-Kit.NET.Integrations.ExtensionsAI

Quick Start

Chat Completion

using LMKit.Model;
using LMKit.Integrations.ExtensionsAI.ChatClient;
using Microsoft.Extensions.AI;

// Load a model
using var model = LM.LoadFromModelID("gemma3:4b");

// Create an IChatClient
IChatClient client = new LMKitChatClient(model);

// Use it
var response = await client.GetResponseAsync("What is the capital of France?");
Console.WriteLine(response.Text);

Streaming

await foreach (var update in client.GetStreamingResponseAsync("Tell me a story"))
{
    Console.Write(update.Text);
}

Embeddings

using LMKit.Integrations.ExtensionsAI.Embeddings;

using var embeddingModel = LM.LoadFromModelID("embeddinggemma-300m");
IEmbeddingGenerator<string, Embedding<float>> generator = new LMKitEmbeddingGenerator(embeddingModel);

var embeddings = await generator.GenerateAsync(["Hello world", "Goodbye world"]);

Dependency Injection

using LMKit.Integrations.ExtensionsAI;

builder.Services.AddLMKitChatClient(model);
builder.Services.AddLMKitEmbeddingGenerator(embeddingModel);

Tool Calling

var options = new ChatOptions
{
    Tools = [AIFunctionFactory.Create((string city) => $"Sunny, 22C in {city}", "get_weather", "Get weather for a city")]
};

var response = await client.GetResponseAsync("What is the weather in Paris?", options);

Supported Features

Feature	Support
Chat completion	Yes
Streaming	Yes
Embeddings	Yes
Tool/function calling	Yes
Temperature, TopP, TopK	Yes
Max output tokens	Yes
Stop sequences	Yes
Frequency/presence penalty	Yes
JSON response format	Yes
Token usage reporting	Yes
Dependency injection	Yes

License

Apache 2.0