Monitoring and reliability layer for AI agents that surfaces patterns across hallucinations, tool call failures, and user frustration.