How Ledatic works, what it measures, and what it wants from the work.
Ledatic builds software that runs itself. That sentence took longer to find than the software took to build, and that says most of what there is to say about this company.
We build before we explain. Ideas here start as working systems, and the understanding lives in the thing itself before it lives in a sentence. We have made peace with that order. A system that works but is poorly described can be fixed with a better sentence. A system that is beautifully described and does not work can only be fixed by doing the work.
We own the foundation. Ledatic wrote its own programming language from scratch: Rail, a compiler that compiles itself, from a seed binary of 1.1 MB that links nothing beyond the system library. Nobody required this, and existing languages would have been faster. But a company whose purpose is software that operates without supervision should understand its tools all the way down, and the surest way to understand something is to build it.
Today we train models against that compiler, with a compile as the checker, and our own 138M-parameter model replays its training steps to the same bytes on chips from three makers. Where we still lean on someone else's runtime, we say so: a Cloudflare Worker serves this page, and Python signs the beacon's pulses. Every layer we wrote is one we can open and read.
We measure by autonomy. The test we apply to anything we make is simple: does it keep working when nobody is watching?
An entropy beacon that signs a pulse every 2 s, each one chained to the last, more than ten million of them so far. A site that rebuilds, re-signs and redeploys itself every morning. A data stall that takes payment from software agents and hands back a signed receipt with no hand on it. A trading system that ran on its own until we paused it. A reasoning loop built to tune itself, which taught us the most on the day its gains turned out to be noise.
These look like unrelated projects, and for a long time we called ourselves "multi-faceted" because we could not see past that. They are the same project. Each one answers the question of how much a piece of software can be relied on to do alone.
We want three things from the work: knowledge, data, and income. We say so plainly because pretending otherwise would be dishonest. Knowledge, because every system we build teaches us something we could not have learned by reading. Data, because systems that run themselves produce a record that is valuable in its own right. Income, because a company that earns can keep building on its own terms.
Income is the one still unproven. The stall takes payment and delivers by itself, and as of November 2026 no stranger has bought from it. When one does, that will be the proof that matters: something we built runs well enough alone that people pay for what it produces.
We would rather be accurate than impressive. It is easy to describe technology in a way that sounds good to a room, and harder to describe it in a way that is true. Builders know every exception, and the temptation is either to bury people in detail or to give up and say "I don't even know what I'm building." Neither is honest. Our commitment is to say the simple true thing first, then go as deep as anyone asks. It is why the paragraph above says what has not happened yet, and why this site starts at sea level and climbs.
What Ledatic has is a habit: make the thing, make it stand on its own, then learn to say what it is. The software runs itself. We are still learning to speak for it.