Creating Policies
Define static rules, semantic guardrails, and approval workflows for your AI agents.
Static Rules​
Evaluated in <10ms using JSONPath matching. Ideal for budget caps, regex patterns, and field validation.
{
"name": "Block High Value Transfers",
"type": "static",
"rule": {
"field": "content.amount",
"operator": "gt",
"value": 5000
},
"action": "block",
"message": "Transfers over $5,000 require approval"
}
Semantic Rules​
LLM-based evaluation for toxicity, PII, and context-sensitive governance (~100ms latency).
Policy Actions​
| Action | Behavior | Use Case |
|---|---|---|
block | Prevents action entirely | Hard budget limits, banned operations |
warn | Flags but allows | Soft limits, monitoring thresholds |
require_approval | Pauses for human review | High-value decisions, edge cases |