We measure the world for a living.

Ommny began in 2024 with a shared irritation. The three of us came from speech, vision, and robotics teams that kept meeting at the same bottleneck: the senses do not arrive in separate requests. A person hearing a page turn has already expected the shadow. Every model we had was born deaf to that expectation, and no amount of wiring models together after the fact seemed to teach it.

So we stopped building six kinds of model and started building one corpus. The omnibus corpus records the world the way witnesses experience it, every channel at once, with provenance attached. The models that train on it, O-1 and its distillations, inherit the expectation we could never engineer: a strawberry looks, sounds, and feels like one thing.

Today the lab is thirty-one people across six disciplines, including two perfumers, an audiologist, and a former line cook, all of whom appear in the acknowledgments of our reports because their senses are in the corpus too. We remain small on purpose. Perception research is a craft of measurement, and measurement does not scale by headcount.