Hermes Portable Rescue USB
===========================
Multi-boot IT rescue toolkit — boot from ISO or use on a running Windows PC.

🥾 BOOT MODE (Ventoy)
   Restart PC -> F12/F2/ESC -> select USB -> pick ISO/hermes-rescue.iso
   Boots Xubuntu with auto-launching hardware diagnostics + Hermes AI.

🪟 WINDOWS PORTABLE MODE
   Plug USB into any Windows PC, open windows-hermes/, double-click:
     run-hermes.bat

   MENU:
     1) Hermes CLI  — auto: online (opencode-go), falls back to local LLM
     2) Hermes CLI  — force offline (pre-starts local llama.cpp)
     3) Web Dashboard + AI Chat  — http://rescue.local:8080 / http://localhost:9090
     4) Hardware Diagnostics  — CPU, RAM, disks, GPU, BSOD, drivers
     5) Disk Tools  — CrystalDiskInfo (SMART health check on Discrete)
     6) Windows MCP Server  — UI Automation (click, type, screenshot)
     7) Exit

📂 USB STRUCTURE
   /ISO/                       — Bootable ISO images (Ventoy)
   /hermes/                    — Config, scripts, auth, web dashboard
     config.yaml               — Online + offline providers (auto-fallback)
     auth.json                 — Your opencode-go credentials
     scripts/diagnose_windows.py  — Windows-native hardware diagnostics
     web/server.py             — Hardware dashboard on :8080
     tools/disk-info/          — CrystalDiskInfo portable
   /windows-hermes/            — Windows portable agent (embedded Python 3.13)
     python/                   — Embedded Python 3.13 + Hermes Agent 0.15.x
     bin/
       wheels/                 — 47 pre-downloaded Hermes Agent .whl files
       wheels-windows-mcp/     — 95 pre-downloaded Windows-MCP .whl files
       llama/                  — llama-server.exe + CPU-optimized builds
       ollama/                 — Qwen2.5-Coder-1.5B GGUF model (1,117 MB)
     run-hermes.bat            — Main launcher (double-click)
     start-llama-server.bat    — Starts local LLM (auto-detects GPU)
     start-windows-mcp.bat     — Starts Windows-MCP UI Automation
     start-hermes-dashboard.bat — Starts Hermes AI Chat on :9090
   /ventoy/                    — Ventoy bootloader config

🔑 NOTES
   - Online mode: opencode-go (deepseek-v4-flash) — uses your credentials
   - Offline mode: llama.cpp (Qwen2.5-Coder-1.5B Q4_K_M) — auto GPU detection, ~2 GB RAM
   - All modes share the same memory/session data
   - Diagnostics and dashboard work fully offline
   - CrystalDiskInfo requires admin for SMART data on some systems
   - Windows-MCP (UI Automation) requires admin for screenshots/clicking
   - Hermes AI Chat at http://localhost:9090 (requires llama-server running)