It's live. I built and shipped a real-time automotive WebGPU product configurator, and it already speaks agent.
Most "3D configurators" are slow, plugin-heavy, or quietly faking it with pre-rendered turntables. This one isn't. It renders photoreal product configurations in real time, directly in the browser, on a WebGPU pipeline. Pick a trim, swap the wheels, drop in an option package, throw on a custom wrap, repaint the whole thing, and watch the price update live.
A note on the stack, because people always ask. The 3D viewer itself runs on vite + three.js targeting WebGPU. Next.js is not doing the rendering; it powers the surrounding product and marketing pages. Firebase Firestore holds the data and state, and semantic vector embeddings give the configuration an intelligent 3D-data RAG layer, so an AI agent can reason over the model, the options, and the pricing instead of just spinning a camera around it.
It started as a car configurator, but the architecture does not care about cars. Trims, wheels, accessories, wraps, paint, pricing: the same engine drops into any industry and ships to desktop, mobile, XR, and VR from one codebase.
This is the part I am most excited about: a configurator an agent can actually sell with.
See it in the portfolio → Automotive WebGPU Product Configurator