Careers
Build the gym.
We are a small crew building the training loop for RL agents — the environments, the verifiers, and the operating system they run in. The work is forward-deployed, cross-substrate, and judged on evidence.
01
Forward-deployed
We build inside real workloads, next to the teams that own them. The product is shaped by what survives contact with a customer's stack.
02
Evidence over opinion
Claims come with the run that produced them. Our own agents work this way — a conclusion that cannot name its evidence does not get stored — and neither do we.
03
Every substrate
Robots one week, language agents the next, diffusion models after that. The loop is the same; the engineering never is.
04
Small, senior, accountable
A small crew where each person owns a whole surface — the simulator integration, the verifier layer, the desktop runtime — end to end.
Open roles
Nothing posted right now — we hire when we meet someone exceptional. If the problems above read like your problems, write to us with something you have built.
Work with us
Show us something you built.
The strongest application is a thing that runs: an environment, a verifier, a harness, a robot that does the task.