Governance

Define and enforce safety policies, tool restrictions, and human-in-the-loop approval gates for your agents. Ensure AI behavior stays within boundaries.

What is Governance?

Governance policies are rules that control what agents can and cannot do. They are enforced automatically before every tool call and LLM interaction, ensuring your agents stay safe in production.

Create a Policy

Create Governance Policybash

curl -X POST http://localhost:8000/api/v1/governance/policies \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "name": "production-safety",
    "agent_id": "AGENT_ID",
    "rules": [
      { "type": "token_limit", "max_tokens_per_run": 5000 },
      { "type": "tool_restriction", "blocked_tools": ["file-delete", "email-send"] },
      { "type": "approval_gate", "requires_approval_for": ["dangerous"] },
      { "type": "content_filter", "block_patterns": ["credit card", "SSN"] }
    ],
    "enabled": true
  }'

Policy Types

Type	Description
token_limit	Max tokens per run to control costs and prevent runaway generations
tool_restriction	Block specific tools from being called
approval_gate	Require human approval before certain actions
content_filter	Block outputs containing sensitive patterns
escalation	Auto-escalate to a human agent under certain conditions

Always add governance for production agents!

At minimum, set a token limit and block dangerous tools. This prevents unexpected costs and unsafe behavior.

RBAC

Event Bus