Infrastructure engineer - UK. Twenty-odd years of building the plumbing.
Through 2026 I spent a few months asking teams operating LLMs in production (especially in regulated environments) how they actually handle runaway spend and "prompt data can't leave our environment" constraints. Twenty-minute conversations, nothing for sale, patterns shared back with everyone who took part.
The write-up is here: Everybody watches. Almost nobody stops.
If you're living this problem, or you think I've got something wrong, I'd still like to hear it: thedanielanthony@gmail.com
The reason I was asking. Fusewire is a self-hosted governance proxy that hard-stops LLM API overspend before a request ever leaves your network. It runs in your VPC, never holds your keys and never touches the payload.
It's built and it works. It's also parked - I never reached enough of the right people to find out whether there was a business in it, and I'd rather stop than keep guessing. If you're in that position and want to talk about it, the door's open.
I'm a 20-year software veteran and startup CTO who spent six years owning cloud spend and production infrastructure for live AI and gaming platforms - including a Kubernetes-based MMORPG backend where every millisecond of latency and every dollar of GCP bill was my problem. I've been building with LLMs since before "RAG" had a name and Fusewire exists because I needed it: real-time budget enforcement that never sees your prompts, never holds your keys and never rewrites a byte.