Checking Your System...

Detecting WebGPU capabilities and hardware...

How It Works

WebGPU

Next-generation graphics API that enables GPU-accelerated AI inference directly in your browser.

WebLLM

Runs Hermes 2 Mistral 7B locally. Your prompts never leave your device — complete privacy.

Local First

The AI model (~4GB) downloads once and runs entirely on your hardware. Low latency, no API calls.

Hardware Tiers

BEST

Optimal Performance

Smooth, responsive AI with no compromises

Apple Silicon

  • M1 Pro / M2 Pro / M3 or better
  • 16GB+ unified memory
  • Safari, Chrome, or Edge with WebGPU

Windows / Linux

  • 8+ core CPU
  • RTX 3060+ or equivalent (8GB+ VRAM)
  • 16GB+ system memory
  • Chrome, Edge, or Firefox with WebGPU
What to expect:Fast model loading, smooth inference, great multitasking
BETTER

Good Performance

Works well with some care

Examples

  • Base M1 / M2 Mac with 8GB RAM
  • Windows laptop with 16GB RAM + integrated graphics
  • Desktop GPU with 4-6GB VRAM

Tips for best results:

  • Close other heavy apps before using
  • Keep only one GhostNN tab open
  • Expect slightly slower startup
  • May experience occasional pauses on long sessions
What to expect:Usable but needs dedicated attention
GOOD

Limited Support

May struggle or not work

Not Recommended

  • 4GB RAM devices of any kind
  • Older Intel Macs with 8GB RAM
  • Chromebooks and tablets
  • Mobile devices (experimental)
Reality:Model may never load, frequent crashes, unusable latency

Quick Checklist

Answer yes to all three to run GhostNN smoothly:

1Can your device download ~4GB once for the AI model?
2Do you have 16GB+ unified memory, or 8GB+ GPU VRAM + 16GB system RAM?
3Are you on a modern browser with WebGPU enabled?

Browser Support

Chrome
ChromeRecommended
Edge
EdgeRecommended
Safari
SafariApple Silicon
Firefox
FirefoxMay need flag

Ready to Start?

If your system meets the requirements, you're all set.

Launch GhostNN

GhostNN runs Hermes 2 Mistral 7B via WebLLM & WebGPU