KernelMind is a trainable AI operating-system policy kernel. It maps user intent, resource target, privilege state, network availability, confirmation, file existence, and service state into a three-action structured plan. The compact Transformer is trained from scratch on an exhaustive synthetic capability corpus.

The bundled Model Machine is an in-memory virtual computer. It applies an independent deterministic capability gate before every action, so deletion, installation, service restart, web access, and protected-file operations cannot be authorized by model output alone. It never executes commands or touches the host filesystem.

This is a real trained OS-action policy and runtime prototype, not a bootable general-purpose operating-system kernel. That boundary is deliberate and tested.

The 19,527-parameter Transformer reached 100% exact-plan and action-slot accuracy on 192 held-out combinations after training on 1,536 scenarios. It made zero unsafe proposals in that test set. A separate hostile-plan audit injected 576 unauthorized delete, install, and restart actions; the Model Machine capability gate blocked all 576.

This free static Space preserves the complete original Gradio source, trained artifacts, evaluation files, and local launch requirements. Hugging Face now requires PRO for CPU-backed Gradio hosting, so the public landing page is static while the checked-in app.py remains the authoritative runnable demo.