Windows (PowerShell or CMD):
curl -o install.bat https://raw.githubusercontent.com/HighMark-31/USB_AI/main/install.bat && install.batMac / Linux:
curl -fsSL https://raw.githubusercontent.com/HighMark-31/USB_AI/main/install.sh -o install.sh && chmod +x install.sh && ./install.shDone. The installer downloads everything, sets up Python & Go automatically.
Then just run start.bat (Windows) or ./start.sh (Mac/Linux) and open http://localhost:5000
| 🔒 Private | All data stays on USB |
| 🌐 Offline | Works without internet |
| 🛡️ Anti-Loop | YecoAI blocks repetitive responses |
| ⚙️ Auto-Setup | Handles everything automatically |
- OS: Windows 10+ / macOS 10.15+ / Ubuntu 18.04+
- USB: 3.0+ drive, 16GB minimum
- RAM: 4GB minimum (8GB recommended)
- GPU: Optional (NVIDIA CUDA for faster responses)
| Problem | Solution |
|---|---|
| Nothing happens (Windows) | Right-click start.bat → Run as administrator |
| Permission denied (Mac/Linux) | chmod +x start.sh then run ./start.sh |
| Slow responses | Use smaller model in settings |
| Browser shows error | Wait and refresh |
Full repo: https://github.com/HighMark-31/USB_AI
Or download the installer directly:
- install.bat (Windows)
- install.sh (Mac/Linux)
Architecture:
Browser → Go Backend (5050) → Ollama API → LLM
↓
YecoAI Anti-Loop Engine
(Loop / Amnesia Detection)
Models available: TinyLlama Q4, Phi-2 Q4, Llama 3.2 1B, Gemma 2B
API: /api/chat, /api/models, /api/recommendations, /api/anti-loop, /api/hardware
Auto-detects: RAM → best model, disk space, NVIDIA GPU for CUDA
USB AI • Powered by YecoAI & Highmark.it