A chatbot once sold a Chevy Tahoe for a dollar because someone asked nicely. Here is what it took to put a real prompt-injection filter in front of my local LLM stack — and the one-character trick that walks straight past it.
My ISP blocks inbound ports, so I stopped trying to let the internet in and had my server reach out instead. A full Cloudflare Tunnel setup with nginx and Docker — no port forwarding, no public IP, nothing to scan.
“One more model won't hurt.” That sentence has probably caused more home-lab incidents than expired certificates, broken Docker volumes, and mysterious YAML indentation combined. It certainly caused mine. My NVIDIA DGX Spark was already busy. It was running chat models, an embedding service, speech-to-text, text-to-speech, vector search, automation
Ever since I watched Iron Man, I have wanted my own JARVIS. Not necessarily the flying armor. Although, to be clear, I would accept the flying armor. What I really wanted was the assistant: something that could listen to me, understand my occasionally drunk-sounding broken English, remember previous conversations, search
