Why the next shopping app is a conversation

Think about the last time you needed something slightly awkward to buy. A replacement charger for an old laptop. A present for a nine-year-old who has everything. A specific size of filter for a coffee machine. Chances are you either spent twenty minutes across several tabs, or you messaged someone and asked.

Messaging is where most of us already ask for help. It is quick, it fits around the rest of the day, and it does not demand that you learn a new interface. That is the starting point for Ottr: a shopping assistant you message like a contact, rather than an app you have to open, learn and remember.

The problem with one more app

Shopping apps are good at one thing: helping you buy from the company that made them. Each has its own account, its own saved cards, its own notification settings and its own idea of what you should see first.

For the person doing the shopping, that creates a few familiar frustrations:

• Fragmentation. Comparing three retailers means three apps or three browser tabs, each with a different layout.
• Repetition. You explain your size, budget and preferences again and again, to systems that never talk to each other.
• Friction at the end. Logging in, finding the card, confirming the address and passing security checks all happen at the moment you least want to think about them.

An assistant that lives in your messages changes the shape of the task. You describe what you need once, in plain language, and the work of searching and comparing happens on the other side of the conversation.

Why messaging fits shopping

It is where intent starts

Most purchases begin as a thought, not a search query. "We need a new kettle, the old one is leaking" is a perfectly good brief. Messaging lets you capture that thought the moment you have it, in whatever words come naturally.

It is asynchronous

A conversation does not need your full attention. You can ask Ottr for options on the bus, glance at the shortlist at lunch and approve the purchase in the evening. Shopping fits around your day rather than the other way round.

It keeps a record

A chat thread is a natural history of what you asked for, what you were shown and what you decided. With Ottr, that history belongs to your account, not to a particular app, so you can start on the web and pick up the same conversation in Telegram later.

Nothing to install

You save Ottr as a contact and message it. No download, no new password to manage, and nothing extra on your home screen.

Where Ottr is available

We have designed Ottr so that the channel is simply a way of delivering messages. Your identity, memory, permissions and payments never depend on which app a message came from.

• Web chat is available now.
• Telegram is supported, and Ottr is reachable there where the bot is enabled.
• WhatsApp is built and waiting on Meta's approval, which covers the business, the number, message templates and its policy on AI assistants.
• Apple Messages for Business is planned. It requires registration through an approved provider, and we will not use unofficial workarounds to automate iMessage.

OTTR is in development, and purchasing runs in a sandbox against a test merchant for now. We say more about that below.

What has to be true for chat shopping to be safe

The convenience of messaging comes with a real risk. Chat apps are informal. Messages are short, typos are common, and "yes" can mean "yes, buy it" or "yes, I see it". Accounts can also be shared, borrowed or compromised. A shopping assistant that treats a chat reply as permission to spend money would be building on sand.

So we drew a firm line early on: a reply in a chat never buys anything.

Approval happens on a secure OTTR page

When you decide to go ahead, Ottr sends you a short-lived approval link. Opening it requires your signed-in OTTR session. The page shows the exact item, price, retailer, delivery address and card, and the approval is tied to those exact terms. If any of them change, including the live price, the approval no longer counts and nothing is charged. You can read the reasoning in Why a 'yes' in a chat should never be enough to buy something.

Your card details never enter the chat

Card numbers are never typed into a conversation and never shown to the AI. Ottr pays with a single-use virtual card created for that one purchase, as explained in Single-use virtual cards.

New contacts start as guests

Anyone can message Ottr and start chatting straight away. But owning an account, and therefore being able to approve a purchase, requires a separate linking step that connects a signed-in web session to control of that chat. Simply knowing someone's phone number is not enough.

Conversation is the interface, not the whole product

It would be easy to read "messaging-first" as "chat window with a model behind it". The conversation is only the front door. Behind it sits a purchase engine with a defined sequence of steps, a payments ledger, a durable workflow that survives restarts, and reconciliation that checks with the merchant before retrying anything, so an order is never placed twice.

That machinery is what turns a pleasant chat into something you can rely on. It is also what allows other AI agents to use the same platform, with your permission, as described in Agent-to-agent commerce.

What this looks like in practice

A typical exchange is short:

1. You message: "I need a USB-C charger for a 14-inch laptop, at least 65 watts, under £40."
2. Ottr asks one clarifying question if it needs to, then comes back with a small number of options, with the price, delivery estimate and the trade-offs between them.
3. You pick one. Ottr prepares the purchase and sends an approval link.
4. You review and approve on the secure page. Ottr completes the checkout and confirms the order.
5. Updates arrive in the same thread until the parcel lands.

You never leave the conversation except to approve, and that one step is deliberate.

The takeaway

The next shopping app may not look like an app at all. It looks like a contact in the messaging app you already use, one that understands plain language and handles the tedious parts. What makes that possible is not just a capable model, but a careful separation between talking about a purchase and authorising one.

If you would like to try it, you can get started with Ottr, and for tips on asking well, see How to brief an AI shopping assistant.