How g-h-0-s-t/ai-pendrive works
AI Pendrive is a portable Windows launcher for a local llamafile model server, with tier-specific model profiles and a separate live status dashboard. The main workflow is selecting a launcher, loading its configured model in llamafile, and using the local browser UI; monitoring is a parallel workflow. The README documents metrics, agent tools, and server configuration, but no source excerpts were available, so internal wiring is limited to documented behavior.
Main entry points: server Llamafile Server · start_4gb_gpu.bat 4 GB Launcher · start_6gb_gpu.bat 6 GB Launcher · start_8gb_gpu.bat 8 GB Launcher
Components 14
Launch Profiles
- 4 GB Launcher
start_4gb_gpu.bat - 6 GB Launcher
start_6gb_gpu.bat - 8 GB Launcher
start_8gb_gpu.bat - 4 GB Model
models/4GB - 6 GB Model
models/6GB - 8 GB Model
models/8GB
Local Inference
- Llamafile Server
server - Browser UI
server - Agent Tools
server
Monitoring
- Status Dashboard
status.bat
Other components
- User
- Browser
- Windows Metrics
- NVIDIA GPU Data
Connections 23
- User → 4 GB Launcher: selects
- User → 6 GB Launcher: selects
- User → 8 GB Launcher: selects
- 4 GB Launcher → 4 GB Model: selects path
- 6 GB Launcher → 6 GB Model: selects path
- 8 GB Launcher → 8 GB Model: selects path
- 4 GB Launcher → Llamafile Server: starts
- 6 GB Launcher → Llamafile Server: starts
- 8 GB Launcher → Llamafile Server: starts
- Llamafile Server → 4 GB Model: loads selected
- Llamafile Server → 6 GB Model: loads selected
- Llamafile Server → 8 GB Model: loads selected
- 4 GB Launcher → Browser: opens when ready
- 6 GB Launcher → Browser: opens when ready
- 8 GB Launcher → Browser: opens when ready
- Browser → Browser UI: uses localhost UI
- Browser UI → Llamafile Server: requests inference
- Llamafile Server → Agent Tools: enables tools
- Agent Tools → Windows Metrics: runs shell and file tools
- User → Status Dashboard: starts dashboard
- Status Dashboard → Llamafile Server: polls metrics
- Status Dashboard → Windows Metrics: reads metrics
- Status Dashboard → NVIDIA GPU Data: reads GPU data