🔥 ROCm-WSL-AI Web UI

Make your AMD GPU sing inside WSL2. Use one Python CLI (rocmwsl, alias rocm-wsl-ai) and an optional Textual TUI to install, launch, update, and remove local AI tools — always ready for the latest PyTorch Nightly.

What you get

• Always latest ROCm (from AMD’s “latest” apt repo) + PyTorch Nightly matched to your installed ROCm series
• A modern, keyboard-driven TUI (Textual) with clear categories
• One place to install, start, update, and remove local AI tools (image gen + LLMs)
• Optional no-chmod Python CLI: install and run everything with a single command

Tools included (by category)

Image generation

• ComfyUI
• SD.Next
• Automatic1111 WebUI
• InvokeAI
• Fooocus
• SD WebUI Forge

LLMs

• Ollama (with a small model manager script)
• Text Generation WebUI

ROCm‑WSL‑AI Web UI

A modern, lightweight web interface to install, run, and manage local AI tools on WSL2 with AMD ROCm. It wraps the existing project features into a single browser-based control panel with live logs, jobs, and per‑tool settings.

Highlights

• One web dashboard for popular tools (ComfyUI, A1111/SD.Next/Forge, Fooocus, InvokeAI, SillyTavern, TextGen, llama.cpp, KoboldCpp, FastChat, Ollama)
• Start/stop, status, and interface links per tool
• Live logs via SSE or WebSocket with filter and colorized streams
• Job history with progress for installers and long‑running tasks
• Models: location overview, index, refresh, link, and curated preset downloads
• Wizard to set up base folders/venv and defaults for tool flags
• Per‑tool settings persist (URL, extra args), plus smart Host/Port helpers
• Clean, responsive UI (PicoCSS), with theme toggle and small toasts/dialogs

---

Quick start

1. Install inside WSL2 (recommended)

PyPI (preferred):

python3 -m venv ~/.venvs/rocmwsl
source ~/.venvs/rocmwsl/bin/activate
python -m pip install --upgrade pip
pip install rocm-wsl-ai

PyPI package: https://pypi.org/project/rocm-wsl-ai/

From source (alternative, if you cloned this repo on Windows):

cd /mnt/f/Coding/rocm-wsl-ai   # adjust path to your repo inside WSL
python3 -m venv ~/.venvs/rocmwsl
source ~/.venvs/rocmwsl/bin/activate
python -m pip install --upgrade pip
pip install -e .

From Windows PowerShell into WSL using your local checkout:

wsl -e bash -lc "cd /mnt/f/Coding/rocm-wsl-ai && python3 -m venv ~/.venvs/rocmwsl && source ~/.venvs/rocmwsl/bin/activate && python -m pip install --upgrade pip && pip install -e ."

2. Run the Web UI inside WSL:

export ROCMWSL_WEB_TOKEN="set-a-strong-token"   # optional but recommended on LAN
rocmwsl-web                                     # serves on 0.0.0.0:8000

From Windows, open http://localhost:8000 in your browser (WSL2 forwards localhost).

You can also launch it from PowerShell directly into WSL:

wsl -e bash -lc "export ROCMWSL_WEB_TOKEN='set-a-strong-token'; rocmwsl-web"

Tip: Change the port if needed: run(host='0.0.0.0', port=XXXX). Example from within WSL:

python -c "from rocm_wsl_ai.web.app import run; run(host='0.0.0.0', port=9000)"

---

Using the Web UI

Dashboard

• Cards show each tool’s status (running/stopped), PID, and actions.
• Click Install to run the installer as a background job.
• Start launches the tool (background when supported). Stop ends it.
• Logs opens a live log stream (switch between SSE/WS). Use the filter box for regex filtering; stderr/stdout are color‑coded.

Tools page

• Per‑tool settings:
  • Interface URL (used for the “Open interface” link in cards)
  • Host & Port helpers that auto‑compose common flags (e.g., --listen/--port)
  • Extra Args to pass on start (stored and reused)
• The UI keeps Host/Port and URL in sync for convenience.

Models page

• See where models are located for different categories.
• Build and refresh a searchable models index.
• Link your models into supported tools folders.
• Download preset model bundles (curated). Tasks run as jobs with progress.

Wizard

• Configure base directory, venv, and optional defaults for tool flags (host/port/flags).
• Saves defaults into a tools.json so starts can reuse them.

Help

• Quick tips and troubleshooting pointers integrated into the UI.

---

Updates

To update the package:

pip install -U rocm-wsl-ai

Your settings (tools.json), job history (jobs.json), and config live in the project’s config directory and are preserved across updates.

---

Security

• Optional token-based auth: set an environment variable before you start the server.

$env:ROCMWSL_WEB_TOKEN = "your-long-random-token"
rocmwsl-web

• With a token set, the UI redirects to a small login where you paste the token. APIs also accept the token via cookie, x-auth header, or token query parameter.
• If you expose the server on your LAN (host 0.0.0.0), use a token. For public networks, prefer a proper reverse proxy and TLS.

---

FAQs / Troubleshooting

• The tool doesn’t start or shows “stopped” quickly.
  • Open Logs to see errors in real‑time. Check Extra Args on Tools page. Verify the tool repository and dependencies are installed.
• Interface link opens the wrong port.
  • Edit the URL in the tool’s settings. Host/Port helpers can auto‑compose flags; the UI syncs URL and Host/Port.
• SillyTavern install requires Node.
  • The installer attempts to guide via nvm. If Node isn’t present, the job runs nvm install/use LTS and npm install in the SillyTavern folder.
• Where are my settings stored?
  • tools.json and jobs.json are saved next to your main config.toml (see Models page → Where for base folder hints).

---

Uninstall

pip uninstall rocm-wsl-ai

---

License

MIT

---

Releasing (maintainers)

PyPI:

python -m pip install --upgrade build twine
python -m build
twine check dist/*
twine upload dist/*

GitHub:

• Tag the release (e.g., v0.2.0) and push tags
• Create a GitHub Release with notes and attach wheels/sdist if desired

After PyPI release, you can simplify README install instructions to a single line:

pip install rocm-wsl-ai

The TUI (optional)

rocmwsl menu   # launches the Textual TUI

Use arrow keys/Enter. Install “base” first, then pick your tools. Launch and update from the TUI or via CLI.

Typical first run

1. Wizard: schnelle Ersteinrichtung (Base + optional ComfyUI)

rocmwsl wizard

oder manuell:

1. Installation → Base (ROCm & PyTorch Nightly)
2. Restart WSL if asked
3. Installation → Pick your tools (e.g., ComfyUI, A1111, Ollama)
4. Launch → Start your tools

CLI equivalent

rocmwsl wizard --base-dir "$HOME/AI" --venv-name genai_env
# oder manuell
rocmwsl install base
rocmwsl install comfyui
rocmwsl start comfyui

Upgrading

Upgrades:

• Update everything: rocmwsl update all
• Update a single tool (e.g., ComfyUI): rocmwsl update comfyui
• Update base (PyTorch Nightly): rocmwsl update base
• Self-update (CLI/TUI):
  • If installed via pipx: pipx upgrade rocm-wsl-ai or pipx upgrade rocmwsl
  • From CLI: rocmwsl update self

Useful tips

• Diagnose: rocmwsl doctor (prüft /dev/kfd, rocm-smi/rocminfo, Torch/HIP im venv)
• Konfiguration: ~/.config/rocm-wsl-ai/config.toml

   [paths]
   base_dir = "/home/<user>/AI"
  
   [python]
   venv_name = "genai_env"
  

• If the TUI looks very plain, install whiptail (see Requirements)
• If you changed groups during base install: restart WSL (wsl --shutdown from Windows)
• Ollama’s systemd user service may require systemd in WSL; if it doesn’t start, run it manually via the scripts
• For ROCm trouble, use the menu’s Driver Management and follow the prompts

License

MIT