When your AI agent is about to do something risky — move money, issue a refund, change an account — Approv gets a real human's yes/no on WhatsApp, and keeps a tamper-proof record of every decision.
Free during early access · WhatsApp & SMS · EU AI Act–ready audit trail
// Agent is about to refund $4,200 — pause and ask a human first. const { id } = await approv.request({ action: 'Refund $4,200.00 to order #8821', approver: 'appr_123', }); const result = await approv.wait(id); // human taps Approve on WhatsApp if (result.status === 'approved') { // ✓ safe to act — and it's signed & auditable }
You decide when an action is risky. Approv handles reaching the human and recording the outcome.
One API call when it hits a high-stakes action. Nothing runs until a human decides.
The right person gets a clear Approve / Reject on WhatsApp — delivered reliably, with SMS fallback.
The decision returns to your agent, and every step is written to a tamper-proof, exportable audit trail.
AI agents are starting to take real actions. When they get it wrong, it costs real money — and soon, real fines (EU AI Act, Aug 2026, requires human oversight + audit trails).
Most teams bolt on a Slack ping or an email. Those get lost, ignored, and leave no record you can defend.
Reliable delivery and a trustworthy record are the whole job. That's what we do.
WhatsApp first, automatic SMS fallback, retries, and a live delivery trail. The message reaches the human — or you know exactly why not.
Every step is hash-chained and signed. Anyone can independently verify a decision wasn't altered. Export-ready for auditors.
Built for the channels people actually reply on, with first-class Arabic and MENA delivery — where other tools fall down.
The human-oversight step and audit trail the EU AI Act asks for — as a few lines of code, not a project.
We're onboarding a handful of teams building AI agents. Tell us what you're building and we'll reach out.