
Private, always-on AI on your own PC — and the mistakes that cost us weeks so they don't cost you.
These are real field notes from running local AI models and agents every day on one home PC with an 8 GB graphics card, for a real small business. Every rule was learned the hard way and written down only after the fix was proven.
Inside (A5 PDF, 11 chapters + checklist):
• Why your graphics card is the budget, and how to check the model is actually using it
• Installing properly (and the older-AMD-card Vulkan fix that took us from 2.5 to 20 tokens a second)
• The context-window trap that crashes home PCs, and the one setting that prevents it
• One machine, one of everything: duplicate servers, orphan processes, clean restarts
• Giving each model a job: tool worker, quick chatter, cleaner
• Slow is fine, stuck is not: jobs, timeouts, flood control
• Protecting your models from deletion, plus backups and drive speed
• Proving it works, filtering untrusted input, and writing playbooks smaller models can follow
• A one-page checklist
Plain English, no fluff. Instant PDF download. Companion to Agents That Actually Work.