Your agents are running. Do you know what they did last night? Midlantics A2A is a governance layer for multi-agent AI systems — four modules that work independently or together: → Observe — tracing for agent action. Open source. → Policy — YAML rules that block or flag agent. → Approval — human gate for high-risk actions. Notified email. → Firewall — scans every message for prompt injection and PII before it reaches your LLM. Under 2ms, runs locally. VPC-native. No data leaves your perimeter.