Skip to main content

Approving agent-initiated transfers

Written by Ari Schlacht

When you let an AI agent move money through Sequence's API, you can require that every money move it starts gets your approval before it runs. That covers both one-time transfers and the rules an agent triggers. The agent sets everything up, but nothing leaves your account until you approve.

Letting automations move real money is a big step, and you shouldn't have to fully trust an agent to take it. With approvals on, all the logic still runs on its own, and the one moment that matters, moving the money, pauses for a human.

How it works

  1. The agent tees up the money move. Through the API, your agent creates a transfer or triggers a rule as usual.

  2. Sequence pauses and hands back an approval link. Instead of running, we hold the action in a pending state and return a secure approval link. You get it by SMS and email, and because the API returns the link too, the agent that made the request can show it to you right where you're working, so if Claude makes the request it can display the URL and you approve without leaving the chat.

  3. You review and approve. Open the link, see exactly what the agent wants to do and the reason it gave, then approve or deny.

  4. It runs, or it doesn't. Once you approve, the money moves. If you deny, it never happens, and if no one answers within 24 hours the request auto-denies on its own.

Turning it on

Approvals are set per API key, and they're on by default for new keys. When you create a key, you'll see a Require approval before execution option on any scope that can move money, so you can keep a fully autonomous key for low-risk automations and an approval-required key for anything you want to watch.

Every approval is logged with who approved it, when, and the reason the agent gave, so you always have a trail of what your agents did and why.

What about read-only agents?

Read-only agents used to be shut out of moving money entirely. Now they can request a money move and wait for your approval instead of being blocked, so you can connect an agent in read-only mode and still let it propose a transfer, like paying an invoice it spotted or sweeping a month-end surplus into savings, with nothing moving until you say yes.

Is this safe?

Yes, that's the point. Your agent can do everything else on its own, but it can't move money without you in the loop.

To see how approvals work in depth and how to set them up, head to the developer docs.

Did this answer your question?