Use case · CrewAI
Human-in-the-loop approvals for CrewAI crews
Give a CrewAI agent an off-switch on its riskiest tasks. A single gate.require() routes the action to a human, blocks until the verdict, and logs it — so autonomous crews can act freely while irreversible steps still need a person.
await gate.require("database.drop_table", { table });
// blocks the crew task until approved, else fails closedWhere it fits
- ›Approve destructive or external actions before a crew executes them
- ›Set per-agent policies so only sensitive roles require sign-off
- ›Audit every human decision across the crew
Frequently asked questions
How do I require human approval in a CrewAI crew?+
Call gate.require() inside the tool a task uses. The crew task blocks until a human approves, so autonomous agents keep running while only the irreversible steps wait for sign-off — each decision recorded in a tamper-evident audit trail.
Can only some CrewAI agents require approval?+
Yes. Policy is by action, amount, scope, and agent, so you can require sign-off only for sensitive roles or high-risk actions while the rest of the crew proceeds automatically.