Use case · AutoGPT
Human-in-the-loop approvals for AutoGPT agents
AutoGPT runs long, self-directed loops that can execute real commands. Seal puts a human off-switch on the dangerous ones: wrap the command in gate.require() and the loop pauses for approval before acting, with every decision on a tamper-evident record.
await gate.require("shell.exec", { command });
// the autonomous loop waits — or fails closedWhere it fits
- ›Require sign-off before the agent runs a shell command or writes files
- ›Gate outbound actions like sending email or spending money
- ›Keep an auditable trail of a long autonomous run
Frequently asked questions
How do I stop an AutoGPT agent from taking dangerous actions?+
Wrap the risky command in gate.require(). The autonomous loop blocks until a human approves, so a self-directed agent can keep planning while irreversible steps still need explicit sign-off — recorded in a tamper-evident audit trail.
Does this slow down the whole agent?+
No. Only actions that match your policy pause for approval; everything else runs at full speed. Approval by exception keeps the loop fast while guarding what matters.