HPE Zerto Develops On-Premises Agentic Troubleshooting System Using Amazon Bedrock

HPE Zerto, in collaboration with AWS, has introduced an innovative agentic troubleshooting system that operates within customer environments to improve operational efficiency and recovery readiness. This system leverages Amazon Bedrock's capabilities to support secure, flexible deployment of foundation models, enabling natural language interactions grounded in live operational data.

The system architecture integrates multiple components, including a UI layer embedded in the existing Zerto interface, an agentic layer with specialized sub-agents built using the Strands Agents framework, and an intelligence layer that manages inference, security, and operational controls. The deployment is designed to run entirely on-premises, with only inference requests and knowledge base queries transmitted securely to AWS.

Key challenges addressed during development included model selection, evaluation methodologies, and ensuring the system's operation within strict on-premises constraints. The architecture emphasizes reliability through multi-agent decomposition, with a hub-and-spoke topology that centralizes control and simplifies debugging. Real-time streaming of investigation progress enhances user trust and experience.

Since its deployment in Q2 2026, over 20% of Zerto customers have adopted the system, resulting in a 10% reduction in support cases and faster troubleshooting. The system's grounding in live data and trusted documentation allows operators to make more informed decisions, reducing operational friction and enhancing disaster recovery capabilities.

Lessons learned from this project highlight the importance of on-premises deployment considerations, agent decomposition for reliability, and the value of real-time feedback in user interactions. Overall, the integration of AI with operational resilience tools marks a significant step forward in enterprise disaster recovery strategies.