What actually happens, start to finish.

11:04pm. A burst pipe.

1. The call comes in on your number

   Not a new number you have to advertise. We sit behind the line you already give out — either always, or only when nobody picks up, depending on which tier you bought.

2. The agent answers and says what it is

   First line, every call, no exceptions: it's an AI assistant and the call is recorded. You can't switch that off and neither can we. It's the same script in every state.

3. It works out whether this is an emergency — using your rules

   Not a generic template. During onboarding you tell us what counts as a true emergency at your shop, in your words. Burst pipe at 11pm and no heat with an infant in the house go one way; a dripping tap goes another.

4. Your on-call tech gets a text within seconds

   Address, the problem in the customer's own words, and a callback number. Not a notification telling him to log into something — the actual details, on his phone, so he can decide and call back.

5. If it can wait, it gets booked instead

   Into your calendar or your CRM, in a slot you said was bookable. And if it turns out the caller needs a person right now and one is available, the agent transfers instead of trying to be clever.

6. You see all of it in the morning

   Recording, transcript, what was decided and why, in the dashboard. Nothing to take on faith.

What the twenty-minute call covers

We publish the agenda because a promise about your time is worth nothing if you can't see what's in it. This is the whole thing:

Then you're done. If we need something afterwards we'll chase it by text or email — we won't book you a second call, because that would be the same as breaking the promise.

Before it ever answers a real call

Every agent we build has to pass a scenario suite — a couple of dozen scripted calls run against it, with the transcripts read by a human, not scored automatically. Among them:

• A burst pipe at 2am.
• No heat, with an infant in the house.
• Someone who only wants to know how much a water heater costs.
• A price shopper who has already called four other shops.
• A wrong number, and a salesperson.
• An angry existing customer.
• Someone who only speaks Spanish.
• Someone who talks over the agent the entire time.
• Dead air.
• Someone who just wants a human, immediately.

An agent that fails the suite doesn't go live. Live in 14 days, or the setup fee is refunded. We keep working — a missed deadline is a much better outcome than your customer finding the bug at 2am.

What the agent will not do

That last one is the whole design. A confident wrong answer at 2am costs you a customer and a story that travels — in a trade where everybody knows everybody, that's the failure that ends a vendor. An honest handoff costs you nothing.

We also do not dispatch, do not take payments, do not do your general marketing, and do not do recruiting. The full out-of-scope list is on the pricing page.

What we store, and who we'll call

Call recordings, transcripts, and the messages we send and receive on your behalf. Plus a consent ledger: every time a customer of yours gives permission to be contacted, we record when, from what page, and the exact wording they were shown.

That last part is not bureaucracy. It is what makes follow-up lawful, and it is the thing you'd want to be able to produce if anyone ever asked. We will not dial a list you upload — not a purchased list, not an old customer export, not "they're all opt-ins." That is the single fastest way for a vendor like us to destroy a client like you, and there's no version of the conversation where we say yes.

Our full AI and calling disclosure →