Skip to main content
Agent Sentinel includes foundational hooks for EU AI Act–style logging and human oversight.
In v0.1 these features are foundational stubs and may be gated to Enterprise tiers in production deployments.

​
Human approval for actions

Mark an action as requiring approval: 
from agent_sentinel import guarded_action

@guarded_action(
    name="transfer_funds",
    cost_usd=0.10,
    requires_human_approval=True,
    approval_description="Transfer funds between accounts",
)
def transfer_funds(from_acct: str, to_acct: str, amount: float) -> None:
    ...
If no approval handler is configured, the action is blocked via PolicyViolationError.

​
Register an approval handler

from agent_sentinel import HumanApprovalHandler, ApprovalResponse, ApprovalStatus, ApprovalRequest

def approve(request: ApprovalRequest) -> ApprovalResponse:
    # Replace with your UI/workflow (Slack, web console, etc.)
    return ApprovalResponse(
        request_id=request.request_id,
        status=ApprovalStatus.APPROVED,
        approver_email="security@example.com",
        notes="Approved for test",
    )

HumanApprovalHandler.set_approval_handler(approve)

​
Platform compliance endpoints

The platform provides endpoints like:
  • GET /api/v1/compliance/health
  • POST /api/v1/compliance/export
  • GET /api/v1/compliance/oversight
See Platform compliance.