AI Shopping Guides for 3D Web Stores and Configurators

A 3D store lets shoppers spin a sofa or repaint a car, but the moment they have a real question ("will this fit under my window?") they are back to scrolling a spec table. Neocortex puts a product guide inside the scene: an AI character who knows your catalogue, answers in plain language and changes what the shopper is looking at while it talks.

What it looks like

Fernhill Living is an original furniture brand with a browser room planner. A shopper builds a rough copy of her living room: one long wall, a window, a radiator she can't move. In the corner of the screen stands Odile, the shop's interior guide, who has opinions about rugs and doesn't hide them.

Shopper: I like the Halden sofa but my window wall is only 2.3 metres. Does it fit? Odile: The three-seater is 2.2 metres, so yes, with a little room to spare. Here it is against your wall. Shopper: Can I see it in green? Odile: Moss or bottle? I'd go moss with that oak floor. Showing it now.

Each reply comes with actions. The first returns SHOW_PRODUCT targeting the Halden three-seater, and the planner drops it against the window wall. The second returns CHANGE_COLOR targeting the moss swatch, and the fabric changes on screen. When the shopper says "fine, I'll take it", Odile returns ADD_TO_CART, and the store's own cart code shows its usual confirmation.

Odile knows the sofa's width because it sits in the catalogue documents she was given. She knows the shopper's wall because the planner sends the room as metadata: each wall, window and placed item with an ID, a name and its measurements as properties. Nobody wrote a branch for "green sofa, short wall".

How Neocortex makes it work

• AI characters. Odile is designed in the Neocortex visual node editor: a personality, a tone of voice, what she knows and the actions she may return. The merchandising team can chat with her in the browser dashboard before any store code exists.
• Knowledge base. Product sheets, dimension tables, fabric and care guides go in as PDF or TXT documents. Free includes 1 upload, Starter 5, Pro 20 and Team 100, so a bigger catalogue fits on the paid plans.
• Actions. SHOW_PRODUCT, CHANGE_COLOR and ADD_TO_CART are Action nodes on the character. A reply can carry several, each with a name and a target ID, and your page runs them.
• Metadata targets. Send what is on screen as metadata: the products in the room, the available swatches, the walls and their sizes. Action targets point at these IDs, so "show it in moss" lands on the right sofa and the right fabric.
• Events. Tell the guide what the shopper just did ("added the oak table", "switched to the bedroom view"), so her next answer starts from there.
• Session memory. Keep the returned sessionId. A shopper who comes back tomorrow can resume the shopper's session, and Odile picks up from the sofa they were planning.
• Voices and emotions. Replies can stream in the guide's voice with an emotion tag for her avatar. Shoppers who prefer talking can use the Unity WebGL microphone (over HTTPS) or your own mic capture sent to Neocortex speech-to-text.
• More than 20 languages. One guide can serve shoppers in their own language from the same catalogue.

Neocortex never touches the cart, prices or payment. ADD_TO_CART is a request your store code handles through its own cart and checkout, exactly as a button click would.

Ideas to build

The car configurator. A product specialist walks a buyer through trims, paint and wheels. "Show me the estate in dark blue with the bigger wheels" comes back as three actions on three targets, and "how much fits in the boot?" is answered from the spec sheet while an OPEN_BOOT action swings the tailgate open.

The fashion showroom. A stylist character stands beside a mannequin in a virtual boutique. Shoppers ask what goes with a jacket, and the stylist dresses the mannequin item by item from the current collection, explaining each choice. Sizes and fit notes come from the knowledge base, so she can say the cut runs small.

A game's in-world shop. In a browser game, the shopkeeper behind the counter is the store. Players ask which armour suits a stealth build, the shopkeeper sets three pieces on the counter with SHOW_PRODUCT, and your game handles the in-game currency. Our post on AI shopkeeper NPCs goes further into that idea.

The spare parts counter. A machinery supplier shows an exploded 3D view of a pump. A maintenance engineer describes the noise it makes, the guide suggests the likely worn part from the service manual, and a HIGHLIGHT_PART action lights it up in the model with its part number.

Building it

Unity WebGL stores use the Unity SDK. Stores built on any other web 3D stack call the Web API from their own code, so the guide sits on top of the configurator you already have.

1. Gather the catalogue. Export product sheets, dimensions and option lists as PDF or TXT, and keep them in sync with what the store sells.
2. Create the guide in the visual editor with a personality that fits the brand, the uploaded documents, a voice and Action nodes for each thing the page can do.
3. Describe the scene as metadata. Give every product, swatch and room element on screen an ID, a name and useful properties such as size or price band.
4. Map actions to page code. SHOW_PRODUCT loads a model, CHANGE_COLOR swaps a material, ADD_TO_CART calls your existing cart function.
5. Test in the dashboard with real shopper questions, especially the ones about fit, delivery and returns.
6. Send events as the shopper changes things, and keep the sessionId per shopper.

Prices and stock change faster than documents. Let the guide describe products from the knowledge base, and have your page show the live price and availability from your own system next to anything she suggests.

Store guides are one entry in our AI NPC, ambient AI and voice command use cases. For guiding people around a venue rather than a catalogue, read AI hosts for virtual events. The virtual assistants page and the interactables docs show how metadata and actions fit together. Start with one product line and three actions.

Checked October 2026.

Frequently Asked Questions