Intelligence should choose its body.

Computers became universal when one machine could run any program. Robots have not had that moment yet. Each one is still built around a single body, and that body decides where it can go, what it can reach and how hard it can push.

Robot brains are getting better every month. The body has become the bottleneck.

Proteus makes the body programmable.

Where this goes

Picture a pool of thousands of small modules. Give it a task, and it becomes the machine that task needs. Give it two, and it splits into two machines that work side by side, then merges back into one when the job needs strength. The body is never fixed.

A humanoid has one body. A pool of modules has as many as the work requires, on a factory floor, in a home or on a space station. We think that pool will one day be as ordinary as a toolbox.

That is the embodied AI we are building: one that learns not only what to do, but which body, and how many bodies, to do it with.

It also changes the economics of robots. A humanoid is built from many different parts. Proteus is built around one part, and one part can be made by the million. One pool can do the work of a fleet of specialized machines, because it never sits idle waiting for the one job it was built for. When a module breaks, you swap the module, not the machine. And capacity grows the way compute does: add modules.

Why small matters

Size is the lever. Halve a module, and the same space holds eight times as many: finer structures, narrower openings, more machines at once. Small and numerous, modules stop behaving like one robot and start working like a team.

Each step down takes new actuators, docking mechanisms and manufacturing methods. That is half of our deep-tech core. The other half is the intelligence that lets thousands of modules act as one.

How we get there

Modular robots have lived in research labs for decades. Most were built to prove that reconfiguration works, not to do work with force. Ours is built around load-bearing docks from the first module on, because real work pushes back.

Every generation goes to work as soon as it can do real work. Today’s modules are sized for industrial equipment, so we start in energy and industrial plants, where people still have to climb into vessels and tanks for inspection and maintenance. Proteus goes in through the nozzle or manway instead. Rescue comes next: collapsed structures and voids, where the first look costs a module, not a rescuer.

Our design partners will get a robot that goes in first. We will get the field experience that makes every generation better than the last: smaller, stronger and ready for new kinds of work.

Full stack: the bodies and the intelligence

Ask for a task, “place this sensor behind that opening”, and the system decides which machines to become, how many, how to assemble them and how to move them as one team. That intelligence is ours to build. Our planner already works out structures and docking sequences on the real geometry of our modules, before a single module moves.

We build the full stack: the modules, the swarm intelligence that coordinates them, and the real-world experience that trains it. General models will keep getting better at seeing and reasoning. Choosing a body, splitting into a team and moving thousands of modules as one is the problem we solve. Every deployment teaches our system which dock holds which load, which transition fails and how to recover. No lab produces that data.

The plan

1. Build a module that docks, carries load and moves under its own power.
2. Take it into the field where people still have to climb in.Inspection and maintenance in industrial plants first, then rescue, with design partners.
3. Shrink it, generation by generation.Each size opens new kinds of work.
4. Let intelligence choose the body, and how many bodies.Our own swarm intelligence on our own modules: one platform for physical work, from industry to everyday life and beyond.

Every task should get the machine it needs. We are building that future, one module at a time.