Synchronos
A machine narrow enough that a person can stand behind its work. Built with one profession at a time, and left in your building.
Engagements begin 2027 · shown by invitation
A computer in a locked room in your building. It has been taught to read documents. You load documents onto it; it reads them and writes down what it found, with the page each fact came from. A person checks the work. It has no internet connection and cannot be reached from outside.
There are three other kinds of machine being sold to you.
ChatGPT, Claude, Gemini. Answers anything, runs on that company’s computers far away. Whatever you type into it is sent there.
The big system built for your sector. Online, shared with other institutions, your staff logs in and learns it. Set up for you by the vendor.
The newest kind. Given tasks and left to complete them on its own. Someone has to check what it did.
A box in your building that reads only what you give it, shows where every number came from, decides nothing, and stops.
The machine is the same everywhere. What differs is what it has been taught to read and whose rules it follows. Those are written down with the people who do the work, and signed by them.
A week finding where the hours actually go. Then we take on one job and do it properly.
The machine is installed in your building in the first week. Nothing else runs on it. We work on site, in the room with your people.
Your people learn what it does, what it refuses to do, and how to check its work.
The machine never moved. Take title whenever you want it. We stay for the yearly work, or we don’t.
Community banks and credit unions. Underwriting, loan review, portfolio monitoring. smallbanc.com
Wherever the friction is worst, led by someone with real years in that work.
Everyone who sits with a client has done that work themselves, for years. That is why this does not grow like software.
Nobody — not even the people who built it — can look inside one of these models and know for certain how it will behave. So everything around it is built on the assumption that it might be wrong. The model reads. Everything else is ordinary software and a person.
No internet, no outside library, no other client’s files.
No passwords to your systems. We can connect only when someone in your office lets us in, and only while they watch.
Form and line, page, cell. Anything without a page reference is thrown out, and we keep a record of what was thrown out.
No score, grade, rating or recommendation. No approve or decline. There is nowhere in the system to put one, and a check runs before anything is released and blocks it if one slips in.
The arithmetic happens afterwards, using a method you wrote down and signed.
Two possible answers, an unclear label, a smudged number — flagged for a person, not guessed.
A number that has not been checked cannot be used in a calculation. The system will not allow it.
A correction is added; the original stays with its date. Your files are never used to improve the product.
If the model gets something wrong, it produces a wrong reading and a person catches it. It cannot send your files anywhere, cannot put its own opinion into the work, and cannot get a number into a calculation without someone checking it first.
Who can read them, how long they keep them, under what terms — in writing.
Anything other than “it stops and says so” needs a second question.
If it cannot be opened in seconds, it may not exist.
Written down, dated, signed — and yours.
The honest answer does not require the vendor to survive.
A small number of engagements beginning in 2027, one profession and one region at a time.
Write to us