Wigi LLM — AI skill for Claude Code

Wigi-LLM

A physical control surface for local LLM infrastructure, built on the G.SKILL WigiDash touchscreen panel.

The core of this project is **LLM Launcher** — a `buttons.json` config that turns the WigiDash into a one-touch model switcher for your llama.cpp router. No build step, no .NET, just JSON. It's what I use every day.

The repo also includes a suite of experimental C# widgets (model dashboards, brain monitors, a clipboard agent) that go further into custom UI territory. They work, but they're not as battle-tested as the launcher and I rarely reach for them. Treat them as reference implementations or starting points.

---

Table of Contents

• LLM Launcher (the main event)
• Scripts
• Architecture
• Prerequisites
• Quick Start
• Configuration Reference
• Experimental: C# Widgets
• Project Structure
• License

---

LLM Launcher (the main event)


*The launcher on the panel: a tap-to-switch grid of model buttons. Active models glow (Router ON, OmniCoder); the top bar shows the live loaded model, and a dedicated KILL button flushes VRAM.*

**Type:** C# WigiDash widget driven by a `buttons.json` config.

Architecturally a hybrid: a compiled C# widget DLL handles rendering, polling, and click dispatch; a JSON config file drives what each button does. End users typically only edit the JSON — the DLL stays as-is.

• One-touch model switching — tap a button, the model loads
• Radio group behavior — loading one model auto-unloads others in the same group, preventing VRAM overflow
• Health polling — buttons show active/inactive icons based on real-time /health checks
• Remote monitoring — poll models running on other machines across your network
• Kill switch — dedicated button to un