HPE Zerto has embedded an AI troubleshooting system in its disaster-recovery software, giving administrators a natural-language way to investigate alerts, configuration problems and service health. The application runs as a pod inside the customer’s on-premises environment and connects to live Zerto data through a local Model Context Protocol server.
An orchestrator answers routine questions directly and sends deeper investigations to specialized sub-agents. Separate agents handle Zerto Manager and replication-engine issues, each with dedicated tools, prompts and context. They return compact reports rather than sending raw logs into the parent conversation.
Amazon Bedrock supplies access to multiple foundation models, while Bedrock Guardrails checks prompts and outputs against policy. Public documentation and runbooks are retrieved through a knowledge base, and tenant-specific quotas are enforced with AWS services. Zerto says secrets and raw logs are removed before the final response reaches the user.
The architecture is designed to reduce time spent moving between dashboards, events and documentation, especially during outages. It remains an account from HPE Zerto and AWS rather than an independent performance evaluation; customers will still need controls for model accuracy, permissions and safe remediation in their own environments.