Answer · The hard limits

What can RelayLaunch AI do without my approval?

Almost nothing that touches the outside world. RelayLaunch AI reads your schedule, finds revenue leaks, and drafts recovery messages, but it cannot message a customer, charge a card, or accept outside AI input without the owner acting first. Every outbound action is held until you approve it.

The limits, one by one

Each limit below is an enforced behavior of the product, not a policy promise. If any of these ever changes, this page changes with it.

Can RelayLaunch message my customers without me?
No. Every customer-facing message is prepared as a draft and held. You read it in your Morning Brief and tap approve or skip; nothing sends until that tap. New pilots also start with a send lock that blocks all outbound messages entirely until the owner turns it off, so a brand-new account cannot send anything at all.
Can it spend money or charge cards on its own?
No. RelayLaunch AI refuses card charges outright rather than asking permission for them. Your subscription is a flat monthly price billed through Stripe, and you manage or cancel it yourself in the Stripe Customer Portal. There are no per-action fees the AI could run up, and it has no path to initiate a payment.
Can other AI agents write into my account?
Not today. The public agent profile at /.well-known/relay-recovery.json declares Level 1, read-only autonomy. Outside agents can read the public scoreboard and the sample receipt. The propose-action endpoint is published as planned and not live; the site is static and a write attempt gets a 405 refusal. When it does activate, proposals land as held drafts for owner approval.
What happens when I am unreachable?
Outbound sends hold. If the owner is offline, external messages wait rather than going out on a guess, and anything that surfaces at an awkward hour defaults to next-day approval instead of a late-night send. The system is built to prefer a missed evening over a message you never saw.
What if the approval check itself breaks?
It fails closed. If the send path cannot positively confirm that an owner approved a message, the message does not go out. A broken check means silence, not an accidental send. That is a deliberate design choice: the failure mode of the approval system is always "nothing happened," never "something sent itself."
Are the numbers RelayLaunch shows real?
Yes, and the proof is that the public recovered-revenue counter shows $0 today, on purpose. It only moves when a real owner-approved recovery is measured, starting with our own books before any client result is claimed. Every unmeasured figure on the site is labeled Sample or (projected), and our own build checks fail if one is not.
Can I leave and take my data with me?
Yes. You can export your clients, recovery history, decisions, and reports from your console at any time, and the recovery record is yours to keep and show your accountant. Cancelling is self-serve in the Stripe Customer Portal in about a minute. No exit call, no data hostage, no per-seat unwinding.

Why we lead with what it cannot do

Most AI tools sell you on autonomy first and mention controls in the fine print. We think an owner should know the ceiling before the pitch: the AI prepares, you approve, and the record of what happened is yours. That order is the product. "AI prepares. You approve." is not a tagline bolted on afterward; it is the shape of every workflow above.

Want to see the workflow? The free AI-Readiness Score uses 15 self-assessed questions in about 90 seconds, with your own numbers, before you sign up for anything.

By , veteran founder of RelayLaunch. We dogfood every workflow on our own business first.