rulith-dev/rulith-inference ? reverse-engineered prompt

Build me a Windows app that makes it easy to run a big local AI model on an AMD Strix Halo machine.

I want one simple installer style experience where a user can download the app, point it at the model files, and start chatting with a local server right away. It should run a fast OpenAI compatible endpoint on localhost, let me change the port, optionally allow other devices on my home network, and support an API key if I want one. I also want a small desktop app for controlling it, seeing logs, and managing the model launch settings so I do not have to remember command line flags.

Please make it work with the Qwen3.8 Flash Next model files described in the docs, including the draft model and the image projector. If you need current details from upstream docs or releases, look them up online. Keep the experience focused on Windows and on making this feel like a polished local AI server I can actually use.