In robotics, the sim-to-real gap is a hard problem: policies learned in simulation fail when they meet the friction of the physical world. Global manufacturing suffers from the same pathology at an industrial scale. Creative intent is the desired state; the supply chain is responsible for realizing it. Instead, we get 12-month latency, signal degradation at every handoff, and output that approximates demand only statistically.

This is an open-loop system. No feedback. No correction. No learning.

We built ONE to close that loop.

**Intent → Code → Atoms → Customer → Learning**

Intent gets expressed once, as structured data. The control plane compiles it into machine-executable code, orchestrates a distributed network of factories, observes outcomes at the customer boundary, and updates its models. Each production episode tightens the mapping. The distance between what was imagined and what gets made shrinks with every cycle.

At the foundation is MetaOne, a knowledge graph encoding the physics of production—trained on eight years of ground truth across hundreds of thousands of garments, materials, bodies, and factory outcomes. The system predicts how a fabric will behave on a machine for a body before the material is even created. Agents swarm for construction, nesting, and quality control, reasoning and arbitrating over this shared ontology.

CreateOne is our Generative Intent Model: a learned mapping of the translation of ideas to what a creator intends to offer. This design intent is compiled into site-aware instructions via a typed intermediate representation. Factory drivers execute locally. The compiler doesn’t just emit instructions—it embeds machine-enforced Quality Contracts. If the physical state fails the contract, the step cannot be completed. In ONE, quality becomes code.

External actors connect through Edge Nodes: creators, factories, commerce, marketplaces. A single transaction might span multiple creators, licensed artwork, pattern suppliers, enterprise brands, and factories across continents. ONE treats that as one execution graph—one intent, one plan, one Proof-of-Production, one settlement. Payments clear only when provenance and quality thresholds are met. No provenance, no payout.

The physical world is modeled as a production graph. Each factory runs a node agent with a learned policy over local state—capacity, materials, machine health. Global orchestration routes micro-orders to minimize cost across time, quality, and carbon. Heterogeneous SKUs aggregate dynamically into efficient runs: mass production to the machines, on-demand to the customer.

Every trajectory feeds the learning engine. The system optimizes against what matters: time-to-instructions, material efficiency, defect rate, fit-driven returns. Each loop tightens the mapping from intent to artifact. The gap closes.

The architecture is portable. New domain, four substitutions: intent schema, codegen, node types, quality contract. Everything else stays invariant.

ONE is not an app. Not a factory. Not a marketplace. It is the operating system for programmable physical reality.

CreateOne is the first instantiation—the operating system for clothing. Nearly 40,000 brands in nine months, no sales team. Physical AI is not about robots in isolation. It is about algorithmic orchestration of entire value chains.

We built the control plane. We invite you to build on it.