One person. A fleet.
Nomos Logos is a small program: one human, and a crew of AI agents that do the research, the building, and the measurement. The point isn't to work alone. It's to prove that when you can trust AI, you don't have to work at scale to do serious work.
The human steers. The fleet does the work.
One person sets the direction: the questions, the taste, the judgment about what actually matters. A fleet of AI agents does the rest: reading the literature, writing the kernels, running the experiments, measuring the results, drafting the findings. In parallel, continuously, through the night.
It isn't one person doing everything. It's one person who never has to do the parts a machine does better, so the human hours land where only judgment counts. A research team's worth of work, steered by a single pair of hands.
Always on.
One human on direction; a fleet of agents carrying the load, each with a lane.
The agents run on the same principles as the model they study: their work is reproducible because the stack is deterministic, and legible because it's built to be read. That's what makes it safe to hand them real work.
A lab that studies thinking, run by things that think.
There's a reason the program can run this way, and it's the same reason the program exists. You can only hand real work to an agent you can trust, and you can only trust what you can verify and read. Determinism makes the fleet's work reproducible; interpretability makes it legible. The two pillars aren't just what we study. They're what let the study happen.
So the lab isn't a side effect of the research. It is the research, proven: the tools we build to read a model are the same tools that let a model do the work. It's recursive on purpose: a program about how models think, carried out by models, each result sharpening the next instrument that finds the one after it.
No "trust us."
Everything here is meant to be checked. The research is public. The kernel and the training pipeline are built to be released. The reasoning is readable by construction. We're not asking you to take a lab's word for a result. We're building the kind of AI whose work you can inspect, and then doing our own work in front of you with it.
If the whole argument is don't trust the people holding the technology — check the law, and read the reason, then the lab has to hold itself to it first.
The lab is the argument: build AI you can trust, and one person can do the work of a team — with the work shown.— The method, in one line