# llmdot

> Run local GGUF language models from .NET — one core package, one model format, one programming model. A managed-by-default .NET runtime for local inference, with optional Metal and Vulkan compute backends.

llmdot is a .NET-native runtime for local language model inference, built around the GGUF model format. It executes major decoder-only and hybrid architectures in the 1–8B parameter range — including multimodal variants — through four architecture-agnostic execution templates resolved from GGUF metadata at load time. The default install is pure managed .NET with zero native runtime dependencies; optional packages add Vulkan and Metal compute backends.

Status: pre-alpha. Targets net8.0, net9.0, net10.0. MIT-licensed.

## Site

- [Home](https://llmdot.cognisoc.com/): Project overview, hello-world example, supported architectures, packaging.
- [About](https://llmdot.cognisoc.com/about/): What llmdot is, what it is for, what it is not, supported runtimes, status.
- [Blog](https://llmdot.cognisoc.com/blog/): Design notes and architecture writing.
- [Compare](https://llmdot.cognisoc.com/compare/): Honest comparisons with adjacent .NET options.

## Blog

- [Why we picked GGUF as the ingestion format for .NET](https://llmdot.cognisoc.com/blog/why-gguf-native-for-dotnet/): Why GGUF, why managed-by-default, what this rules out.
- [Four execution templates for every 1–8B model we care about](https://llmdot.cognisoc.com/blog/four-execution-templates/): The LLaMA-like, GPT-NeoX-like, Gemma-like, and LFM2-like templates, and why the engine is parametric not architecture-specific.
- [Managed-by-default: why CPU is the headline path, not the fallback](https://llmdot.cognisoc.com/blog/managed-by-default-and-the-cpu-path/): The deployment story is the product. Optional GPU acceleration is additive.

## Compare

- [llmdot vs. LLamaSharp](https://llmdot.cognisoc.com/compare/llmdot-vs-llamasharp/): Both run local GGUF in .NET; opposite choices on the native vs managed axis.
- [llmdot vs. Microsoft.SemanticKernel](https://llmdot.cognisoc.com/compare/llmdot-vs-semantic-kernel/): Different layers — SK orchestrates; llmdot executes. Connection point is IChatClient.

## Resources

- [Docs](https://docs.cognisoc.com/llmdot/)
- [GitHub](https://github.com/cognisoc/llmdot)
- [RSS](https://llmdot.cognisoc.com/rss.xml)
- [Sitemap](https://llmdot.cognisoc.com/sitemap-index.xml)