The Obvious Case for Open and Local AI

We are the generation that learned to build with whatever was in front of us. We grew up alongside the technology, from dial-up to the supercomputer in our pockets, and most of us came out of school into a bad economy with debt and little else. So we hustled. We are used to turning very little into something.

So when a tool comes along that turns very little into a lot, we know exactly what to do with it. AI is a step change in what one person can do. Work that used to take weeks takes minutes. We are moving at the speed of ideas. That is the part worth being excited about. It is not the whole picture.

But AI as it is being built today does not naturally reach all of us. It concentrates more than it spreads. The problem is not that AI is powerful. The problem is that the power is being rented back to us. A few companies hold the models and decide who gets access, and the value flows back to them.

The structure shapes the conduct. When a handful of people hold the models and set the terms, the priorities of the whole field tend to follow theirs. A technology this consequential deserves to be carried with care, with the pace set by what is possible rather than by who is competing with whom. The people holding it are gatekeepers whether they want the title or not, and that is a role worth carrying with discipline.

The harder question sits underneath it. If one of these companies reaches what it keeps promising, the decision of what to do with it rests with the same small group. Open weights on your own machine do not change that. Frontier capability will stay expensive and largely closed for a while yet. What open and local AI changes is the other side of it. Not who controls the frontier, but whether the rest of us depend on that group for ordinary intelligence, or stand on something we own. We cannot promise you a say in how superintelligence gets built. We can make sure you are not starting from zero if it does.

There is another way to build this, and we have done something like it before.

Open the models. Fund the teams shipping open weights for everyone, not only the ones running closed systems behind a meter. Reduce the dependency on someone else's tokens, someone else's pricing, someone else's permission. Intelligence should work the way the internet does: open, available, running on infrastructure you control.

This is not a new argument. The cypherpunks made a version of it a generation ago. They decided privacy and individual freedom were worth building in the open, against people who wanted them locked down. They did not win by asking permission. They won by shipping protocols and tools people could actually use. We are living in the world they chose. The same choice is in front of us now.

We think it is the obvious one. Intelligence you can run, inspect, extend, and keep. That no one can revoke or price out from under you. That reaches all of us, not only the people building it. That is the version we want to live in, so it is the one we are building.

That is why we are building Osaurus.