We build the data, agents and simulation that let AI work in real operations.
Frontier models can reason. Whether they work inside an operation depends on what surrounds them: the data under them, the harness around them, the fleet they direct and the site they are tried in first. We build those for operators. For teams that build their own agents, we attack those agents and measure each one against its goal.
AiNOS One already runs in two industries: AiNOS Cover in insurance and AiNOS Hub in supply chain.
For the physical world, our next research focus is the behaviour inside the twin: how people work, how machines interact with them, and how changes can be tested before they reach the site. See the research
Overview
What AiNOS builds, and for whom.
From digital twins to physical intelligence.
We are developing realistic human assets and reusable motion primitives to bring task-aware human behaviour into industrial digital twins. The goal is to reuse this foundation across safety training and operational simulation and, with additional validation, extend it to robot evaluation and learning. Each new site should build on shared components, while its procedures, operating conditions and risks are adapted and validated locally.
For robots and automation the same twin is the gate and the loop: a new fleet is tested there around people before day one, and after it the real site keeps feeding the twin, so each improvement to the plan is tried there, approved by a person and deployed back.
On-screen textevery line of the film, in order
- Industrial AI needs more than a model.
- It needs a world it understands, and a place to test.
- [AiNOS today] AiNOS links the site's records, rules and agents.
- One twin of the operation. That is today.
- [Research visualisation] Our next research focus: the people in it.
- Coherent body, motion and contact.
- Reusable movements, composed by the task.
- Interrupted. Then resumed.
- A new target. Re-planned, not re-animated.
- Another site, the same movements. Its own procedure.
- Rehearse the work. Compare the change.
- [Future direction] Before day one: robots tested in the twin, around people.
- Cleared by a person. Then the real site.
- After day one, the real site feeds the twin.
- Improved in the twin. Approved, then deployed.
- Operation and simulation, one loop.
- AiNOS. From digital twins to physical intelligence.
- Research now
- Human assets
- Motion primitives
- Task behaviour
- Needs validation first
- Robot evaluation
- Robot learning
Bring one real workflow. Let's define what the simulation must reproduce and how to validate it.
Bring a workflow