Synchronos
Owned intelligence for work that has to be defended — built with one profession at a time, and left in the building.
Engagements begin 2027 · shown by invitation
Everything being sold as artificial intelligence right now is sold on more. More autonomy. More speed. More decisions made without you. We build the opposite, and not as a compromise.
A machine narrow enough that a person can stand behind its output and mean it.
Each of those is why the work can be trusted. A machine that could do more would produce output nobody could check — which is the position most professionals now find themselves in, holding a confident answer with no way to verify it.
Some work has to survive being questioned. A lawyer answers to a judge, a lender to an examiner, an accountant to a reviewer, a clerk to the public record. In that work, an answer nobody can trace is not an answer.
These are also the professions least able to use what is being sold to them. Client files, member records, privileged material and case documents cannot be typed into a service somewhere else — and the people doing it know that. They do it anyway, because the tool is open and the deadline is Friday.
The problem is not that they do not understand the risk. It is that no one has offered them the other thing.
What every one of them shares: documents arriving on a schedule, checked against written rules, producing a record someone signs. That is one machine pointed at different paper.
The machine is the same everywhere. What changes is what it has been taught to read and whose method it applies — and that part is written down with the people who do the work, and signed by them.
A week of looking rather than a proposal. We sit with the people doing the work and find where the hours actually go, which is rarely where anyone expects. Then we take on the single job that gives back the most for the least effort.
A machine is installed in your building in the first week and nothing else ever runs on it. Onto it goes your setup: what your documents require, how you calculate things, what a complete file looks like. We work on site.
Your people learn what the machine does, what it refuses to do, and how to check its work in seconds. An institution that finishes this can evaluate any AI vendor on earth, including us.
Which is not an event, because the machine never moved. You may take title whenever you want it. We stay for the recurring work, or we don’t — and either way you keep the machine and everything on it.
We hold no client files, ever. A room containing many institutions’ records would be a target that did not previously exist. We decline to build one.
Most of what makes this work is not ours, and saying so is the point. Two layers are bought. Two are built once and used everywhere. One is built with each profession. And the last one is not a component at all.
Bought, not built
A box sized for the job it does, not for the job it might one day do. Deliberately unfashionable, bought a generation behind, so an identical one can still be found in five years.
Open weights, running locally, swapped on a schedule. Anyone can download the same one we use. Nothing about the model is what makes this different.
Built once, used everywhere
Page in, findings out. Every finding carries the place it came from — a form and line, a page, a cell. A finding without one is discarded.
What the model is allowed to see, what shape its answer must take, and what it cannot produce at all. There is no field for a rating, so nothing can put one there. This is the part with our judgment in it.
Built with each profession
Its own reference material, its own method, its own name, and people who have done that work themselves. smallbanc is the first.
Not a component
Confirms what the machine found, adopts the method it applies, and puts their name on the result. Everything above exists to make that signature mean something.
The two marked layers are the whole company. The hardware and the model can be purchased by anyone. The vertical is a relationship. The signature belongs to the client. What we own is the reader and the rule book — and that is what makes the second profession cost a fraction of the first.
The method does not transfer by being described. It transfers by being done. So each profession gets its own practice, its own reference material, and people who have actually done the work in it.
Community banks and credit unions. Underwriting, loan review, portfolio monitoring, and the recurring paperwork that arrives whether or not anyone has time. smallbanc.com
Chosen by where the friction is worst and where someone with real years in the work will lead it. Not before.
Practitioners must have practiced. Everyone who sits with a client has done the work themselves, in the profession, for years. It is why the first week produces answers rather than a questionnaire, and it is why this does not scale like software. We accept that.
Named in advance, before any demonstration: it does not decide anything a professional is accountable for. No score, no grade, no rating, no ranking, no recommendation, no approval or refusal, no evaluative characterisation of any kind.
This is not a setting we chose and could change. There is no field in the system for a rating, so nothing can put one there. An automated test reads everything the machine writes and stops the build if an evaluative word appears.
The work may state that a figure computes a certain way under your signed method, with citations. What that means is yours to say.
In writing, before anything is sold — because these are the parts that would be easiest to quietly abandon later. They hold in every profession we work in.
Not for the engagement, not for the recurring work, not ever. Nothing is typed into a rented service, transits a third party, or trains anyone’s system. We hold no copy anywhere, and there is no standing connection into your machine.
A page, a line, a cell. A number that cannot be checked in seconds is not a deliverable. Verifiable is not verified — the citation makes checking fast, but a person still checks.
How things are calculated, what a complete file contains, which rules apply and in which edition — assembled with you and signed by you before the first pass runs.
Checking happens against the original document and the published rule, by plain machinery or by a person. Nothing grades its own work, and every number in a calculation is arithmetic rather than a prediction.
Your files are not our laboratory. Development happens on invented data.
The machine is already in your building. The setup, the record and the citations stay with you and stand on their own.
Not only us. Ask these of every vendor who brings artificial intelligence into your practice.
If the answer involves anyone else’s computers, ask who can read them, how long they stay, and under what terms — in writing, before the work starts.
If the answer is anything other than “it stops and says so,” ask what it does instead, and how you would know.
Every figure and every citation should point somewhere a person can open in seconds. A source that cannot be opened may not exist.
Ask to see it in writing, with a date and a signature. The answer should be a method you adopted — not a model’s judgment.
A small vendor is not a disqualification, but the honest ones have an answer that does not require them to survive.
We are taking a small number of engagements beginning in 2027, one profession and one region at a time.
Every engagement is a field test. The record the work produces is built to stand in front of whoever asks.
Write to us