Now available for Muse, Grok Bot, ChatGPT Dots, Hermes Agent and more.
01
Three experiences. One vehicle journey.
Sales
Web 4.0-enabled Vehicle Configurator
Show, don't tell. Claude or ChatGPT can now help you build, and bring home, your dream car. Your agent walks you through the build, prices it, explores options, and configures the vehicle right in front of your eyes.
18 WebMCP tools
Get help navigating dynamic pricing, financing, energy use and multi-year ownership cost
ASI = Always Save (w/) Incentives. An agent-readable incentives and offers engine is an agent query away.
Showroom controls like camera placement, feature spotlight and component focus via WebMCP
Know your vehicle like never before. A real-time digital twin 3D model synced to your vehicle, with a visual and agentic library of components: what they do and where they go.
25 WebMCP tools
Detailed component library and modeling allows a user and their agent to view, inspect and highlight key vehicle elements; the agent can ask what is on the sheet and measure real clearances
Comprehensive visual demonstration capabilities
Fully operable. Drive. Open doors, front trunk, liftgate and charge-port
An agent-native spatial-reasoning tool and framework for code-native 3D model creation by large language models. A ruler that measures the model directly, instead of measuring a picture of it. It is how the vehicle above got built.
16 instruments
Includes tolerance and component verifier, and an aperture gate that names fixed geometry inside any door, hood or liftgate opening
Maps sections, clearances, symmetry and silhouette deviation; camera-less loop
Every screen publishes its tools on document.modelContext. There is no scraping and no screenshot-guessing: the agent reads the same revisioned build the person is editing, and every change it makes lands in the same state, on the same screen, at the same moment.
The person always wins
Tools that change the build take an expectedRevision and reject on conflict. A multi-stage agent transaction can be interrupted by clicking any control, and the receipt says which stages landed. Presentation tools move the camera, never the configuration.
Every claim is citable
Incentive outcomes carry source records with titles, URLs and retrieval dates. Digital-twin measurements state their basis: an independent reconstruction fitted to published dimensions, not manufacturer CAD. The agent can say where a number came from.
03
Try it with an agent
ChatGPT desktop
Update to the latest version, then open any AutoLab URL in the app's built-in browser. Ask it to read the current build. The header chip on each page switches from Manual mode to the number of registered tools when the API is picked up.
Chrome 149+
Enable the flag, restart Chrome, load the page:
chrome://flags/#enable-webmcp-testing
Any other browser gets the same experiences by hand. The Configurator exposes the identical tool API on window.autolab and the Owner's Guide on window.r2, for devtools and automation.
Read the build, price, incentives and revision; list valid options; simulate; apply an interruptible multi-stage transaction; interrupt; undo; present showroom, camera, Blueprint and focus; capture buyer context; estimate payment, energy and ownership cost; compare up to three builds; read the twin; list 42 parts; reveal, frame and highlight; authored views; lights, motion, dissolve, openings, explode; measure; switch surfaces; run the twin's guided tour.
The embedded twin's own drawing tools, including set_vehicle_context, so the configured paint, wheels, interior, range and price arrive with the vehicle.
State, view, motion, camera, orbit, frame, list and get parts, measure, highlight, annotations, published specification, reset; start and stop the guided tour; what is visible on the sheet by component and coverage; the real surface-to-surface clearance between two components; look at any coordinate in the vehicle frame; choose the vehicle; start, drive and stop a run on the Tesla Roadster.
Read the measured fit report, the aperture gate report and the instrument list for the model that both experiences render.
This page
3
List the experiences, open one, read the overview and how to run it with an agent.
Registration
const api = document.modelContext ?? navigator.modelContext;
await api.registerTool({
name: "open_autolab_experience",
description: "Open the Configurator, the Owner's Guide or the Engine.",
inputSchema: { type: "object",
properties: { experience: { type: "string",
enum: ["configure", "garage", "engine"] } },
required: ["experience"], additionalProperties: false },
execute: async ({ experience }) => open(experience)
});
Contract
Closed top-level schemas. Human-readable titles. Explicit read-only, non-destructive and idempotency hints. Errors carry the recovery: no part "flux capacitor". Call list_parts for the 42 available ids. Structured JSON in metres. Build-changing tools are revision-gated; presentation tools are not, because they cannot lose anyone's work.
Until about 3 years ago, a car was the most complex piece of machinery you'd ever own.
What if AI could better help you choose what you're buying?
Or the car you already own?
Or help the company that sold it to you fix the thing, the right way, at the lowest cost, the first time you bring it in?
Your car keeps getting smarter.Your AI should get smarter about your car.
