Ai, ai, ai!

Three PowerShell scripts to keep your sanity

You know the drill. You want to run ComfyUI on your AMD card, natively on Windows, with the ROCm stack. So you open a browser tab. Then another. Then a forum post from 2023. Then a GitHub issue that sort of helps. Three hours and twenty-one open browser tabs later, you have have installed Python seven times and you are not sure which one is the right one.

This bootstrap script is the shortcut you wish you had. Three PowerShell scripts, one folder structure, and you are up and running.

TXT
1-init.ps1    creates folders, detects your GPU, sets up symlinks
2-deps.ps1    installs Git, Python, Ollama, FFmpeg
3-apps.ps1    installs ComfyUI, Open Web UI, and configures them

After that, a single ai command manages everything: start, stop, restart, status, install, logs. It handles NVIDIA CUDA, AMD DirectML, AMD ROCm, and CPU fallback automatically. If you have an AMD card with driver 26.2.2 or newer, it offers native ROCm PyTorch for faster inference.

What you get

A clean folder layout where models live once in the vault and runtimes reference them through symlinks. Reinstall any tool without losing your data. GPU-aware backend selection. No manual config hunting. Port management if the defaults clash with something else.

The control panel (ai) will allow you to directly:

  • Start, stop and restart all services
  • Present a dashboard showing running services, ports, model counts
  • Show full system diagnostics
  • List installed models by category
  • Install or update the apps (idempotent)
  • Watch service logs for the individual apps
  • Change service ports
  • Check and fix environment variables
  • Free up disk space — pin models to AI_VAULT first (HF, PyTorch, ComfyUI temp, logs)

Why it exists

Because getting a native Windows AI workstation running should not require a PhD in driver archaeology. The project name says it all: Ai, ai, ai! That is the sound you make when you realise there is a better way.

MIT License · GitHub · Documentation