Your first design

This is the five-minute version. By the end you will have a 3D bookshelf in the viewport, you will have tweaked a slider to widen it, and you will know which button opens the export menu.

1. Open the app

Go to prompt2cad.com/app. If you are signed in (see Sign up), you land on the home lobby: a Direct prompt card, a Guided setup entry, and a Recent projects grid. Type your prompt into the Direct prompt card and click Generate model to spin up a fresh project and start designing.

Start from an example instead

You can also browse the public model library and open a design such as a fitted wardrobe, a kitchen cabinet run, a workshop bench, or a mechanical bracket. Choose a model page, inspect its image or interactive preview, then click Customize this design. The editor opens a new project with that editable model loaded; it does not copy another customer's project or chat. Browsing and opening an example do not start an AI generation or spend credits. Parameter sliders are free; asking the agent to change the design uses credits, and CAD/3D-file downloads unlock after a paid top-up. These example models are starting geometry for your own measurements and downstream detailing, not production-verified plans or ready-to-cut instructions.

2. Write your first prompt

Type in the chat input at the bottom of the left panel:

A simple oak bookshelf, 180 cm tall, 80 cm wide, 30 cm deep. Five shelves.

Hit Enter (or ⌘/Ctrl + Enter). The agent starts streaming a response.

Be concrete about dimensions and material

The more grounded your prompt is, the less back-and-forth. "An oak bookshelf 180×80×30 cm with five shelves" needs no clarifying questions; "a nice bookshelf" needs three or four.

Prefer not to start from a blank prompt?

Click Start guided setup on the launcher instead. You'll pick a category (furniture, mechanical parts, enclosures, DIY, etc.), then answer a short structured form — dimensions, materials, interfaces, clearances — with recommended values where you're unsure. The agent turns your answers into a structured brief and plans from there. Same result, less typing.

3. Answer the clarifying questions

For anything ambiguous, the agent will show a small form inline in the chat — a few quick toggles, a number with a unit, or a "choose one" group. Pick the option that matches your intent and hit Continue. You can always type a custom answer in the Other… field.

The agent will then show you a plan: the overall dimensions, the rough internal layout, and the style direction. Skim it, hit Approve if it looks right, or paste a follow-up message asking for a change.

4. Watch the model build

Once you approve the plan, the agent writes the geometry code and the viewport renders the model live, primitive by primitive. The viewport estimates about 2 minutes to first geometry; complex designs and retries can take longer. Chat is charged from actual token usage, so the credit cost varies with the request and agent response.

When it finishes you will see a small popover near the Export button: "Your model is ready". That is the cue that the chat turn finished. Follow-up prompts are also charged from actual token usage and are typically cheaper than creating a model from scratch.

5. Tweak a slider

In the bottom-right of the viewport, a small Parameters panel lists every adjustable knob the agent declared — typically width, height, depth, shelf_count, and any style enums. Drag the width slider from 80 cm to 120 cm: the model re-renders on the next frame, with the same proportions stretched to fit. No new chat turn, no credit cost.

Sliders are free

Parameter sliders re-execute the geometry code locally in your browser. They never cost credits.

6. Refine with the agent

Sliders cover small changes. For everything else, type a follow-up:

Make the bottom shelf taller for vinyl records.

The agent edits the geometry code and the viewport updates with the change. The cost is based on the tokens used by that chat turn, not a fixed iteration fee. You can keep iterating — add doors, change the finish, add a back panel — until you are happy.

7. Export

Click Export above the viewport. You will see:

• CAD / Engineering → STEP, DXF
• 3D / Web → GLB, DAE, OBJ, STL
• Image → PNG

PNG is free at any time. The CAD and 3D formats are gated until you top up at least once — see Top-up. The lock dialog walks you through it.

What just happened

You went through the four-stage loop the whole app is built around:

1. Prompt — natural language describing the piece.
2. Plan — the agent's interpretation, surfaced for you to approve.
3. Build — the agent writes geometry code, the viewport renders it.
4. Refine — sliders for small tweaks, follow-up prompts for everything else.

Everything from here is variations on that loop. Next: Interface tour for the layout, or jump to Writing prompts for prompt craft.