Read cofounder Muhammad Umar's memo, Compression and Reinflation
CodeNinja AtomsUnited StatesSaudi ArabiaPakistan

Physical AI

Physical AI

Physical AI is AI that senses the physical world, reasons about it and acts in it. Cameras, instruments and machines feed it; a typed model of the operation (an ontology) holds what it knows; open-weight models reason over that model; and a named person approves every action that changes the physical world. Where generative AI works on text, physical AI works on plants, fleets, grids, ports, mines and fields. CodeNinja Atoms has published 44 open physical AI designs for real operations across 12 sectors in the United States (19), Saudi Arabia (13), Pakistan (11), Germany (1), each free to reuse under CC BY 4.0.

44open designs12sectors4countries3compute patterns

What physical AI does, sector by sector

Each row is a sector with at least one published design. Every design is one real operation, newest first; open it for what it senses, the models it runs and what it costs.

SectorCountriesDesigns
Energy and utilitiesPakistan, Saudi Arabia, United StatesA Self-Improving AI System for OT Vulnerability Management at a Utility in the United States, Sovereign AI Architecture for Distribution Loss and Theft Reduction in Pakistan, Industrial Data Ontology for a Water Utility in the United States, Sovereign AI Architecture for a Grid Control Room in Saudi Arabia, Turbine Island Watch, Reliability Atlas, Grid Context Watch, Feeder Firewatch
Heavy industry and constructionPakistan, Saudi ArabiaIndustrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan, A Self-Improving AI System for Cement Kiln Control in Saudi Arabia, Structure Phase Watch, Steel Count Ledger, Factory Fire Watch
Agriculture and earth observationPakistan, Saudi Arabia, United StatesA Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan, Baseline, Fodder Watch, Field Ledger
Maritime and portsGermany, Saudi Arabia, United StatesIndustrial Ontology Examples, Port Drone Watch, Port Twin, Terminal Pulse
Oil and gasPakistan, United StatesPipeline Integrity Watch, Corridor Guard, Loop Integrity Watch, Sovereign HSE Watch
Aviation and aerospace manufacturingSaudi Arabia, United StatesIndustrial AI Reference Architecture for an Engine MRO Shop Floor in the United States, Physical AI System Design for Composite Part Inspection in an Aviation MRO in Saudi Arabia, Hangar Form Cell
Discrete manufacturing and automotivePakistan, United StatesPhysical AI System Design for Weld and Assembly Quality in a Vehicle Plant in Pakistan, Depot Operating Object Model, Pit Camera Watch
MiningPakistan, Saudi Arabia, United StatesIndustrial AI Reference Architecture for Tailings Storage Facility Monitoring in the United States, Sovereign AI Reference Architectures for a Remote Mine on a Satellite Link in Pakistan, Industrial AI for an Open Pit Mine Haul Fleet in Saudi Arabia
Rail and transportationSaudi Arabia, United StatesSovereign AI Reference Architectures for Passenger Rail Fleet Maintenance in Saudi Arabia, Wayside Trend Watch, Consist Watch
Supply chain and logisticsSaudi Arabia, United StatesSovereign AI Architecture for a Freight Corridor and Customs Clearance in Saudi Arabia, Industrial Data Ontology for a Cold Chain From Plant to Store in the United States, A Self-Improving AI System for Carrier Exception Handling in the United States
SemiconductorsSaudi Arabia, United StatesPhysical AI System Design for Wafer Defect Inspection in the United States, Silicon Ledger
Warehousing and intralogisticsSaudi Arabia, United StatesPhysical AI System Design for a Mobile Robot Fulfilment Centre in Saudi Arabia, Warehouse Twin Command

How a physical AI system is designed

Every published design makes the same decisions in the same order: what to sense, where inference runs, the object model, the open-weight models, the compute sized from their weights, and the write paths back into the operator's systems with a person in control.

Where physical AI runs

Physical AI runs next to the operation, not in a distant cloud: on the edge, on the site or on one node in the operator's own country. The 44 designs fall into 3 compute patterns.

PatternWhat runs whereDesigns
Frontier reasoning nodeeight 141 GB HBM-class GPUs holding an open-weight frontier model at FP8, owned or hosted in the country28
One inference card48 GB PCIe inference cards, from one card to one eight-card server, at the site or in the country; no frontier model7
No new computethe design runs on servers or software the operator already owns; nothing to buy9

Sovereign physical AI

In every published design the models are open weights, self hosted in the operator's country, and no frame or reading leaves it. That is what makes physical AI sovereign AI: the operator owns the models, the data and the decisions.

Built from the published designs only: each paper's At a glance table, its object model and the dataset. Designed on Praxis; object models load into Hyper Ontology. Also: Industrial AI · Ontologies · Self-improving systems · sovereign AI. Updated 2026-11-14.

Want one of these designs for your own operation? Talk to us.