What we do, and how we ended up doing it.

Why it exists

More and more AI writing now carries a mark of some kind. A watermark from the company whose model produced it, or a signed record travelling with the file. Those marks are real evidence, and hardly anyone reads them.

Meanwhile schools, journals and employers reach for tools that guess from writing style, tools wrong often enough to ruin somebody's year, and then treat the output as a verdict. The gap between the evidence that exists and the evidence people actually use is where the damage happens. We built this to close that gap, and to be straight about the part of it that cannot be closed.

What we will not do

This started as a different product. The original idea was a tool for stripping AI watermarks out of text and getting past detectors. We did not build it, and that decision is the whole company: a business that helps content hide where it came from cannot also be trusted to tell you where content came from. You have to pick one.

Two commitments are what remain of that choice. We will not build a watermark remover, or a detector-evasion tool, or anything meant to make AI writing pass as human. And we will not claim anything that cannot be verified, not in the product, not on this website, not in a sales conversation.

That second one costs us more than it sounds. It is why this site tells you which of our checks do not work yet. The full list of lines we will not cross, and the numbers we publish about our own limits.

Where we have got to

Private beta, with a small number of design partners who are helping us find where the method breaks before anyone has to rely on it. There is no product you can log into yet. When there is, this page will say so.