Memory

The assistant remembers a returning shopper's preferences so it doesn't re-ask: what it keeps, for how long, and where that memory lives.

The assistant remembers useful things about a shopper so it doesn't ask the same questions twice. Tell it "I'm a medium" or "I'm allergic to peanuts" once, and it applies that later in the conversation, and on a return visit, without being reminded.

What it remembers

Durable, personal facts the shopper volunteers: their name, sizes, favourite colours, budget, dietary needs, who they're shopping for. It keeps a small, curated set of the most useful things, rather than a transcript of everything said.

How long each fact lasts depends on the fact. Lasting details like a size or an allergy stick around; incidental ones ("looking for a gift this week") fade on their own after a while, so old context doesn't pile up and mislead later recommendations.

The shopper doesn't manage this: there's no "memory saved" popup and no settings screen. It happens quietly as they chat, and the payoff is a concierge that already knows them next time.

Where the memory lives

• Memory is per store. It's tied to a durable, anonymous identity, so the shopper never has to sign in for it to work.

• For a shopper who isn't signed in, memory is per browser/device: clearing browser data, or moving to a different browser or device, starts a fresh shopper with a clean slate.

• A shopper who is signed in to a Shopify store keeps their memory across their browsers and devices. Past conversations don't travel with them; it's the memory, not the chat history.

• Voice and text share the same memory: a preference stated out loud is remembered in text, and vice versa.

"This" and "that": knowing what's on screen

Separately from long-term memory, the assistant is aware of what the shopper is looking at right now: the product card open in the chat, or (in the embed) the page they're on. So "is this in stock in medium?" resolves to the thing in front of them. That live awareness is covered under In-chat tools.

Memory is on by default and needs no configuration. The most reliable way to see it working: state a preference, then start a new conversation and ask for a recommendation; the assistant applies it without re-asking.