Home AI Project
This project explores running local LLMs on my home infrastructure. The goal is to provide AI tools on my network to keep data local rather than sending them to cloud-based services.
The original design ran Ollama and Open-WebUI in containers behind an NGINX reverse proxy. Over time I found this inefficient because the GPU was dedicated to AI and I couldn’t use it for gaming at the same time. I moved my GPU and frontend to a Moonlight gaming PC that lives on my network. It made more sense to run models on a dual purpose system that can still be used for gaming as well.
Open-WebUI Frontend
Open-WebUI was used as the initial frontend for my local models. Running in a container alongside the Ollama runtime, it used an HTTP API and presented an interface for submitting prompts, managing conversations and switching between available models.
MYSTY Frontend (Moonlight)
The current setup runs models directly on a Moonlight gaming node where the GPU can be shared between gaming and AI workloads. The machine runs the Moonlight streaming service for remote access across the network, while the Windows frontend Mysty is used locally to interact with the models.