Local AI (lightweight or expanded LLM)

Lightweight (default, no LLM): Pure shell/PowerShell, uses negligible resources. Good for conversations and toolkit help.

Expanded (optional): Small local LLM via llama.cpp + model (user adds, total added <1.5GB recommended).

When using expanded:
- Launchers apply limits silently: ~20% CPU (low prio + 2 threads), GPU off (0%, edit to low ngl for ~40% if wanted), RAM limited to ~30% of system RAM (Mac ulimit, Windows via small context + model choice).

See main README for menu usage. Choose version in option 12.
No resource usage is printed at start.
