Human approval before your agent does anything risky.

Publishing, paying, deleting, writing to a client: some steps need a person's yes. Your agent asks for it, waits, and only carries on once someone allowed to decide has said so.

How it works

Three steps, and the agent carries on.

1

The agent asks for a sign-off

It describes the action, how risky it is and who should decide. It can go ahead on the first yes, or wait until every approver agrees.

2

A person decides on a page

The approver opens a private page that shows exactly what will happen. To approve or reject, they enter a one-time code we email them at that moment, or use their passkey.

3

The agent carries on, or stops

A yes lets the agent continue; a no stops it. Every decision is kept with who decided, how they proved it and their note.

Example

A release that waits for its lead

A release agent in Claude Code gets version 2.4.0 ready and asks the tech lead before it ships. Sara approves from her phone with a one-time code, and the agent publishes by itself.

Say it to your agent
  • “Before you publish to production, get Sara's approval.”
  • “Ask finance to approve any payment over 500 euros before you send it.”
  • “Don't delete anything until both Luca and Anna have approved.”
Release agentClaude Code· running on its own
request_approval(
  approvers: "sara@acme.example",
  action:    "Publish v2.4.0 to production",
  risk:      "high",
  expires_in_hours: 2
)
SaraPhone· 9 min later
Approvedwith a one-time code

"Go ahead, the 18:00 backup is done."

The agent picked up where it stopped and shipped.
For the people answering

Where the answer comes from.

By email

What the agent wants to do, and a button to review it. Nothing is decided from the inbox, so a link scanner can't approve.

On your phone

The private page shows the action and its risk. Approve or reject with a one-time code or your passkey. From the home screen app, a notification brings you there.

Questions

Good to know.

Why a code, if the approver already has the link?

Links get forwarded, previewed and scanned. A fresh code or a passkey, given at the moment of the decision, proves that the person deciding is the one the agent asked.

Can someone approve from their AI chat?

No, on purpose. An approval is always given on its own page, never by an AI on someone's behalf. Their AI can tell them what is waiting and link to the page.

What happens if nobody decides?

A request expires after 24 hours unless the agent picks another time, up to 30 days. The agent then sees that nobody approved it.

Is the decision kept?

Yes. The agent gets who decided, how they proved it and their note. If the agent asks for it, the decision also goes into the team's decision log, so other agents can find it later.

Free beta

Give your agent someone to ask.

Sign up with your email and create your workspace. We open Guizzo to new teams a few at a time: if every place is taken, you join the queue and we write when yours is ready.

Free during the beta. We send a code to sign in, no password. If every place is taken you join the queue, and we write when yours is ready. Privacy policy