CrossnoKaye brings physical AI into industrial control with AWS AgentCore
CrossnoKaye is using Amazon Bedrock AgentCore to connect its AURA AI system with live industrial plant data, engineering logic and closed-loop equipment control. The company says its platform now supports more than 100 facilities and 13,000 controlled assets across heavy industry.
- AURA uses Amazon Bedrock AgentCore within CrossnoKaye’s ATLAS platform.
- The system supports more than 100 facilities and 13,000 controlled assets.
- AURA runs Python workloads in isolated AWS-managed microVM sandboxes.
- ATLAS keeps base control and safety logic running locally at the plant edge.
- CrossnoKaye reports 650,000 closed-loop control actions per day.
A different approach to industrial AI
Industrial facilities typically divide responsibility among programmable logic controllers, distributed control systems, historians, supervisory control and data acquisition systems, and separate optimization or alarm tools. According to AWS, this makes it difficult for an external AI agent to understand both the plant’s data and the control logic governing its equipment.
CrossnoKaye’s approach is to keep the AI within its software-defined automation platform, ATLAS. The platform connects live plant data and AI reasoning to equipment control in food and beverage plants, industrial refrigeration facilities, aggregates quarries and manufacturing sites. ATLAS runs on AWS and combines the facility model, base control software, historian and optimization applications within one security boundary.
CrossnoKaye’s AURA agentic intelligence system can therefore access the operational state of a plant through the platform’s internal data model. AWS says the system stores more than 22 trillion machine data points, ingests more than 50 billion new readings every day and runs over 1,000 live advanced process control applications.
Control remains local while AI uses the wider plant model
ATLAS operates across existing PLC and DCS infrastructure, including equipment such as kilns, crushers, drying rooms, ovens, spiral freezers, chillers, turbines, valves and compressors. Customers do not need to replace their plant hardware, according to AWS.
Each facility has an on-premises ATLAS control panel with redundant computers, power supplies and network switches. A local Kubernetes cluster runs the control services and executes versioned control narratives that define equipment state machines, sequences, operating bounds, interlocks and alarm rules.
The edge appliances act as outbound-only IT and OT gateways, with zero inbound firewall ports inside the plant. If cloud connectivity drops, the plant continues operating on the local control narratives. CrossnoKaye reports more than 9,000 control software deployments per year and more than 1 million run hours with zero safety incidents.
Cloud-based systems add derived engineering signals and optimization. These can include heat-flow distributions, asset health indices, energy intensity and mass-balance deviations. Optimization applications combine those signals with physics models, model predictive control, reinforcement learning, utility tariffs and weather forecasts to calculate setpoint schedules. CrossnoKaye reports that its closed-loop system runs 650,000 control actions per day.
AgentCore supplies computation for AURA investigations
Large language models can interpret engineering context and plan an investigation, but they cannot reliably calculate statistics or physical balances from millions of sensor readings supplied directly in a prompt. AURA addresses this through the code interpreter capability in Amazon Bedrock AgentCore.
When numerical analysis is required, AURA writes and runs Python workloads at runtime inside isolated, AWS-managed microVM sandboxes. The system uses these environments for statistical, thermodynamic and multi-dataset calculations, while Amazon Bedrock Guardrails and local edge safety controls provide additional safeguards.
AURA runs as microservices and durable workflows inside the platform boundary. Under a user’s enterprise permissions, its agents can read equipment topology, deployed control narratives and interlocks, physical and synthetic readings, alarm and setting-change histories, optimization schedules, and plant documentation indexed in Amazon Bedrock Knowledge Bases.
AWS describes a quarry investigation in which the back half of a crushing plant tripped twice during one shift. Historian timestamps suggested that a secondary crusher was at fault. By also examining the conveyor layout and control interlocks, AURA traced the problem to a slipping recirculating belt beneath the tertiary crusher. The belt triggered a tertiary-crusher overload interlock, which then halted the secondary crusher. An operator confirmed and fixed the belt problem within minutes.
Why the architecture matters for industrial users
CrossnoKaye reports that its advanced process control applications improve production throughput by 11, 17 percent while reducing total facility energy costs by an average of 21 percent. The company’s architecture is designed to let AI use the full operational context without distributing credentials across multiple plant systems or placing core safety control in an external agent.
For industrial users, the example shows how AI can support troubleshooting while remaining connected to existing control infrastructure. Local control narratives continue to govern equipment at the edge, while AURA can investigate events using plant topology, interlocks and sensor history together. Amazon Bedrock AgentCore adds runtime Python execution for calculations that require more than language-model reasoning.
