LLM observability

Cluster: Supports the pillar AI governance.

Track which models were called, with what inputs (redacted if needed), and what was returned. Essential for audit and debugging. Pair with secure prompting and prompt injection defenses.