DeepSeek OG researcher just dropped a personal AI agent that runs a 125B model on a 12GB GPU.

Engine splits the model → hot parts on GPU, rest in RAM. Qwen3.8-Flash-Next runs on 12GB card + 64GB RAM.

One-command install auto-detects your hardware and picks the best model.

Scans your inbox, calendar, files → builds a searchable wiki overnight.

Plugs into your Chrome session. Asks permission before sending anything.

Zero token fees on local models.

./scripts/install/install.sh

This is how you run agents without bleeding money to APIs. Local inference is eating SaaS alive.