A Hybrid Edge-Native Architecture for Autonomous eCommerce Agents

This paper presents a hybrid AI-based autonomous-agent architecture applicable across device, edge, and cloud layers. The proposed system achieves decision latency below 100 milliseconds and reduces communication overhead by over 50% through selective state migration and compressed world-model reasoning, while maintaining decision-making accuracy above 90%. Security and explainability are enhanced by on-device memory with encrypted logs, resulting in more than an 85% reduction in exposure of sensitive data. Multi-agent coordination is accelerated and stabilized, with the design optimized for low-bandwidth environments. Cross-tier continuity ensures consistent agent behavior across network conditions. The architecture demonstrates high performance in real eCommerce applications across speed, privacy, safety, and scalability.