I built this inside KOVAL Distillery in Chicago, where I work, starting in January 2026. It ran there four months as an internal system before any of it was packaged. That packaging began in May 2026, and it has never been installed by anyone other than me.
Separating those two halves is the only reason the first is worth anything. What ran at the distillery is not what you would install. What you would install has no customer yet.
What that does prove
The problems are real ones. Every part of this was written because something in a working business was not getting done: mail nobody answered, a phone nobody picked up, stock counts nobody updated. Four months of daily use is four months of things breaking in ways nobody plans for, and that is where most of the design came from.
What it does not prove
That it installs cleanly for someone else. That it survives a business I do not work at. Whether the setup work generalises or has to be redone each time. The first customer finds that out alongside me.
What you can check today
- A live demo node, running the distillery scenario end to end, including a phone call that reaches an agent and comes back answered. The business data is seeded and labelled as seeded: the mechanism is real, the distillery in it is a stand-in.
- The system, on a call. The event log and the composed prompt of a live agent are each one request away, so nothing here needs taking on trust.
- The release history, dated and complete, with nothing left out because it looked bad.
- Limits, which is the page I would read first if I were you.
The offer
A pilot on your real workflow, where it either earns a place or it does not, and you find out early. For the first customers I am the engineer, not the vendor: you get the person who wrote it, and I get the one thing this does not have yet, a business that is not mine.