Monitoring tracks predefined system signals and alerts when they cross thresholds. It detects potential problems; observability uses more granular evidence to explain unexpected behavior. section 8.3.1
For agent systems, monitoring extends beyond infrastructure metrics, logs, and traces to capture agent activity and state. These records help track long-running tasks and investigate failures across tools, memory, and agent handoffs. section 9.3.6
Signals may include:
- System measures such as uptime, error rate, latency, and resource utilization. section 8.3.1
- Agent events such as prompts, retrieved memory, tool calls, intermediate plans, approvals, and environment state; checkpoints can support reproducibility and rollback. section 9.3.6


















































