How CrossnoKaye grounds physical AI agents in closed-loop industrial control with Bedrock AgentCore

AWS ML Blog · rss · 2026-11-10

AWS ML Blog details CrossnoKaye's physical AI stack on Amazon Bedrock AgentCore:

Problem: PLCs/DCS, historians and SCADA are siloed; control logic is locked in proprietary controller code, and bolting an external agent onto OT systems means spreading credentials.
Architecture: The ATLAS platform software-defines the plant—edge Kubernetes clusters run versioned control narratives locally (9000+ deployments/year, zero safety incidents), covering 100+ facilities, 13,000 assets, 50B readings/day.
Closed-loop optimization: Physics models, MPC, RL, tariffs and weather compute setpoint schedules—650,000 control actions/day; reported throughput gains of 11–17% and 21% average energy cost reduction.
AURA agent: Microservices + Temporal workflows on EKS; LLMs plan investigations while Bedrock AgentCore's code interpreter runs Python in isolated microVM sandboxes for real math, with Guardrails and edge safety controls.
Case study: Repeated crusher trips at a quarry looked like a crusher fault; AURA traced the actual interlock logic to a slipping recirculating belt, fixed in minutes.

More from coding & agent