How it works

What happens to a request

Customers ask your business for things, and more and more often their AI agents do the asking. Here is what happens to each request.

An AI agent asks a bike shop for a service on Saturday at 10:00. The request lands in the shop's inbox as a booking, marked requested. The shop's rule sees the slot is free and confirms it, the agent gets the answer, and the inbox signs a receipt. Then a quote request arrives. The shop's rule sends quotes to a person, so it waits, marked needs you.

Step one

What arrives

Every request becomes an item with a type and a state, before anyone has read it.

Say someone wants their bike serviced on Saturday. Their agent can book it with you directly. Claude and ChatGPT use MCP, and other software uses the REST API. Email works too, for messages and replies. A web form is on the way. Being built

Each request lands as one of four types, with its own set of states, so the inbox always knows where it stands. Refunds are a fifth type, handled inside the app.

A booking, an order, a quote request and a message arrive, and each settles into the inbox as its own type with its first state: requested, received, received and open.
A quote request open in the inbox, with the customer, the quote lines, the next step as a button and the conversation below
One item The request, its state and its history.
Message

open, answered, closed, spam

Quote request

received, needs info, quoted, accepted, declined, expired

Booking

requested, needs info, proposed, confirmed, completed, no show, cancelled by the customer, cancelled by the business, declined, expired

Order

received, needs info, accepted, awaiting payment, paid, fulfilling, fulfilled, completed, declined, cancelled

Step two

Your rules run first

Write down how your business works once. Simple rules then handle the routine without you.

Start with your hours, services and prices, and the order size above which you want to decide yourself. Then switch on a few rules. Six come ready-made for appointments, trades and shops.

Rules don't use AI, so the same request always gets the same answer. The reason goes in the item's history. You can try a rule on a request you already have before you switch it on.

A rule is switched on: when the slot is free and inside opening hours, confirm the booking. Two bookings pass and are confirmed. A third, late in the evening, stops and waits for a person.
The rules screen, each rule written as one plain sentence with a switch beside it
Rules in plain English Each rule is one sentence you can switch off or edit.
Appointments

When a new item arrives and it is a booking and it is not a test and (the total is under 50.00 or there is no total) and the slot is free and it is inside opening hours: confirm it. Stop there.

Shop

When a new item arrives and it is an order and the total is over 200.00: flag it for a person and set priority 2. Stop there.

Step three

Your own AI answers

Connect the AI you already use. It works like a member of staff, with the same limits.

Your own AI, connected over MCP, answers a booking for Saturday at 10:00: that time is taken, would 11:00 suit? The booking moves to proposed, within your rules.
The settings screen with the business name, domain, time zone, currency and languages
The settings your AI works inside Your AI follows the same settings you do.

It signs in as staff

Connect Claude, ChatGPT or any AI that speaks MCP. It signs in with your account and can answer customers or suggest another time.

It answers inside your rules

It can only take the steps a person could, and everything it does is logged.

You watch it first

In test mode, requests from agents are practice ones. Let it handle a day of them before it meets a real customer.

Why use your own AI

  • You already use it

    It may already know your business and the way you talk to customers. A new one would start from scratch.

  • Your customers stay yours

    Their details live in your own database, and the inbox only sends them where you tell it to.

Safety

Safer than a shared mailbox

These protections are built into the inbox, so they hold even when an agent misbehaves.

Two agents ask for the last place at 10:00 on Saturday at the same moment. One gets it and is confirmed. The other is told the slot is taken. When the first agent retries with the same key, it gets the same answer back.

Requests have a fixed shape

An agent can only send a booking, an order, a quote request or a message. If a detail is missing, the reply says which one.

No double bookings

If two agents want your last Saturday slot at once, one gets it. If a connection drops and an agent retries with the same key, it gets the first answer back.

The history is never rewritten

Every change is logged, with who made it and any reason given. Nothing in the log is edited later.

Rules can check who is asking

A rule can confirm callers with a verified account and pass anonymous ones to you.

Your data stays with you

It lives in one database on your Cloudflare account or your own server. If you join a network, it gets hourly counts and receipts, never names or messages.

Receipts are signed by you

When a booking is confirmed or an order is paid, your inbox signs a receipt. The customer's agent can counter-sign it.

All of this works today. Checking an agent's own signature is still to come.

Why now

Why prepare now

People already ask their AI to book and buy things. Those requests will reach you, ready or not.

Say someone asks their AI for a dentist with a free slot on Thursday.

Unprepared

The request sits in a mailbox behind forty others. Or the agent finds no way to ask, and books the dentist down the road.

Prepared

Your inbox shows the agent your free times and how to book. Your rules confirm the slot, and you only see what needs you.

Nothing locks you in. The software is open source and the data is yours to take with you.

Get started

Three steps, in this order

Most of it is writing down what you already know.

  1. Pick where it runs

    Deploy to your own Cloudflare account in one click, or run it on your own server. We will also host it for you once that opens.

    See the three ways

  2. Write down the business

    Add your hours, services, prices and limits. For a bike shop or a bakery, that takes an afternoon.

    Read the quickstart

  3. Connect your AI

    Point Claude or ChatGPT at your inbox and let it handle a day of practice requests in test mode first.

    How to connect

You can start small and add more as you go.