Hearth documentation

Hearth brings local GGUF inference to the standard .NET AI abstractions, so you can register a model once and keep the rest of your app on IChatClient and IEmbeddingGenerator.

Start here

Need Page
Install the package and run your first prompt Getting started
Understand HearthOptions and model selection Configuration
Expose OpenAI-compatible endpoints in ASP.NET Core ASP.NET Core integration
Orchestrate the inference server with .NET Aspire Aspire integration
Add retrieval-augmented generation to your app RAG pipeline
Drop a streaming chat UI into a Blazor app Blazor chat component
Pick the right GPU backend package GPU backends
Explore the sample apps in this repo Samples
Browse the public API surface API reference
Swap an existing OpenAI app to local inference Replacing OpenAI with Hearth
Pick the right model and quantization for your hardware Choosing the right model
Define and invoke tools from .NET methods Tool calling
Docker, Kubernetes, performance tuning Deploying to production
Understand the trade-offs vs Ollama Hearth vs Ollama

What Hearth includes

  • Single-line registration with AddHearth(...)
  • Local chat and embeddings over Microsoft.Extensions.AI
  • OpenAI-compatible endpoints via MapHearth() (Hearth.AI.AspNetCore)
  • Aspire orchestration — container resource + client wiring (Hearth.AI.Aspire.Hosting + Hearth.AI.Aspire)
  • RAG pipeline — chunking, document loaders, in-memory and SQLite vector stores (Hearth.AI.Rag)
  • Blazor chat component — streaming UI with themes and Markdown rendering (Hearth.AI.Blazor)
  • Optional GPU backends for CUDA, Metal, and Vulkan

Core registration example

builder.Services.AddHearth(options =>
{
    options.Model = "./models/qwen2.5-7b-q4_k_m.gguf";
    options.ContextSize = 8192;
    options.GpuLayers = 35;
});

Once registered, inject IChatClient anywhere in your app and keep the rest of your code independent from the model host.

Edit this page