Skip to Content
ExamplesApproval policies

✅ Enforce approval policies org-wide

Stack — Policy: action + threshold → required approver · Pause: request_approval · Queue: Approvals page · Output: consistent, provable sign-off

Who it’s for — teams that can’t leave “should a human approve this?” to each agent’s prompt. What it does — defines approval as org policy, not per-agent instruction: rules map an action and threshold to the role that must sign off. Any agent hitting a gated action pauses for the right approver — the same way, every time, with a record.

Build it

  1. Write the policies — e.g. refunds over $500 → finance lead, any external send → manager, production config change → admin. Rules live at the org level, above any single agent.

  2. Reference the gate in agents — the agent calls request_approval for the action; the policy decides who must approve and when.

  3. Route to the queue — the run pauses at waiting_approval and the request lands on the Approvals page with full context. Here the demo’s support agent hit its refund rule on a real $820 Brightway Logistics claim and stopped for a human:

    A real refund escalation waiting for approval

  4. Approve or reject — the run resumes or stops on their decision; both outcomes are recorded.

  5. Prove it — every approval and its decider is in the audit trail (see Stream your audit trail to a SIEM).

Make it yours

  • Escalate by amount: auto-approve under $50, one approver under $5k, two above.
  • Different approvers per team via RBAC.
  • Apply the same policy across every agent so a new agent inherits the guardrails.

Uses — Enterprise · Approvals · Governance

Govern AI spendMulti-team RBAC