Interface tour

The editor is intentionally one screen — chat on the left, viewport on the right, balance and projects tucked into the header.

Home lobby (/app)

When you open the app without a specific project in the URL, you land on the home lobby instead of being dropped into the last project you touched.

• Direct prompt (left card) — type a one-shot description and click Generate model (⌘/Ctrl + Enter also sends). Attach up to four reference images or up to four OBJ, STL, GLB, DXF, DWG, or SketchUp files with the buttons next to Generate model. Images are sent as visual context. CAD/3D files are associated with the new project but remain out of the viewport until the agent explicitly uses them; it can inspect them without placing them. Accounts with experimental agent overrides also see Agent options beside the attachment buttons, where the model, including experimental options such as DeepSeek V4.1 Flash, and thinking effort can be selected before the first message is sent. Thinking effort defaults to Medium. The same controls remain available under Chat settings inside the project; they affect subsequent agent calls and do not alter existing geometry by themselves.
• Guided setup (right card) — switches the lobby to the step-by-step design wizard inline (no project is created yet). Category and object-type cards use pregenerated reference images to make the choices easier to scan. These images are illustrative, not a preview of your configured model; the wizard shows no generated geometry until you click Generate model at the end. That action compiles the brief, spins up a fresh project, and starts the real CAD generation. Use Back to home to leave the wizard without creating anything.
• Recent projects (grid below) — the six most recently updated projects, newest first. Click any card to resume it. Select Browse all projects to open the project browser when you need another design. The grid shows name and last-edited time; rename, duplicate, and delete live in the project header once you open one.
• Open file — choose a .p2cad project file to import a local copy and open it. You will confirm before import because project files contain design code; only open files you trust. Open CAD file starts a new project from DWG, DXF, SketchUp, OBJ, STL, GLB, or self-contained glTF geometry. It does not read STEP, IGES, or native SolidWorks/Fusion files yet.
• Measurement system — use the m / ft control in the header to choose metric or imperial display units before starting a design. It changes displayed values, not the meter-based geometry saved in projects.
• Feedback — select Feedback in the header to send a rating or note about the product.

If your last session ended inside a project, that project will be in the recent grid — but the app no longer reopens it automatically. Use Direct prompt or Guided setup at the top to start something new.

Deep links still jump straight into a fresh project: ?prompt=..., ?example=..., and the post-checkout redirect land in the editor exactly as before.

Header (top)

• Logo / Wordmark — back to the landing page.
• Project name — click to rename. The agent suggests a name automatically once your first model exists; the placeholder until then is "Untitled Project".
• History (rotate-counter-clockwise icon) — opens the checkpoint panel. Every successful build is auto-checkpointed; you can name and pin checkpoints manually to keep them past the auto-retention limit of 20.
• Projects (folder icon) — opens the project manager. New project, switch project, delete project.
• Credits counter — shows your remaining balance. Click it to open the top-up flow.
• Avatar — choose the editor appearance: Light is the default, while Dark stays dark regardless of your device and System follows your device setting. This preference is saved in this browser; it does not change the public marketing pages, which stay light. You can also sign out here.

Chat panel (left)

The conversation with the agent. From bottom to top:

• Input box — type your prompt. Enter sends. Shift + Enter inserts a newline. ⌘/Ctrl + Enter also sends.
• Image attach (camera icon) — paste or upload up to four reference images per message (photos, sketches, or technical drawings). They count toward the token budget.
• Project files — CAD/3D attachments from the direct prompt remain visible above the composer as file badges. The badges mean the agent can inspect the files; they do not mean the source geometry is in the viewport or exports.
• Stop (square icon) — interrupts a streaming response. The agent stops where it is, no credits are refunded for tokens already generated.
• Messages — your prompts, the agent's text replies, plan diagrams, tool-call cards (one per geometry edit, screenshot, search, etc.), and the inline answer forms.

"Your message is queued"

If you type a follow-up while the agent is still answering, it is queued and sent the moment the current turn finishes. No need to wait.

Viewport (right)

The 3D scene that renders your model. Default camera is perspective; controls:

• Left-drag — orbit around the model point where the drag starts. Dragging empty space falls back to the current view centre.
• Middle-drag (scroll wheel click) — the same point-focused orbit, in CAD style.
• Right-drag — pan.
• Scroll — zoom toward the cursor.
• R — reset to the auto-fit view.
• Click a part — select it and open its details. On desktop, hold Shift while clicking to select several parts.

Top bar (overlay)

• View buttons — Front, Back, Left, Right, Top, Bottom, Perspective. Click any to snap.
• Display mode — Rendered (default) or Wireframe.
• Parts list (list icon) — opens the geometry-derived parts list: stable IDs, dimensions, quantities, materials, and CSV export (see Parts list & cut list).
• Print (printer icon) — generate a scaled, printable PDF template of the current view or an orthographic plane, with outlines, holes, and scale bars (see Print templates).
• Render (sparkle icon) — generates a photorealistic image of the current view. Costs 50 credits per render.
• Export — opens the format dropdown (see Exporting overview).

Bottom-left

• Dimensions — the model's overall bounding box, in your chosen unit (metric or imperial — toggle is in settings).
• Primitive count — handy for performance: very dense models (1,000+ primitives) are slower to re-render on slider drags.

Bottom-right

• Parameters panel — one row per adjustable knob declared by the geometry code. Drag a slider, the viewport re-executes the code with the new value. Enums show as button groups; booleans show as switches.
• Selection panel — click a model part to see its name, type, dimensions, and material when available. Select Add to chat before writing your next message so the agent receives the exact selected part as context; a Shift-click selection adds each selected part. The attached parts appear as removable chips above the chat input and are consumed by the next message. Select Show in code to open the generated geometry at that part's call. For names generated in loops, Prompt2CAD shows the containing section instead; it does not make the code editable from this dialog.
• Terms / Privacy links — only visible on larger screens.

Empty state

If you have not built anything yet, the viewport shows a dashed-border box and "Your model will appear here." While the agent is working on your first build, that placeholder is replaced by a live progress card: Understanding your request → Planning the design → Building your 3D model, with a running timer. The stages follow what the agent actually does (reading your brief, planning dimensions and materials, then writing and executing the CAD geometry). Large designs can appear progressively as complete assemblies are accepted, while the agent continues adding the remaining planned assemblies; these are usable portions of the requested design, not a reduced final result. If the agent pauses to ask you something or waits for plan approval, the card switches to an amber notice pointing you back to the chat. Once the build completes, the progress card is replaced by the model and the parameter panel.

Error banner

If geometry code fails to execute (rare, but it happens — usually a parameter combo the agent did not test), a red banner appears at the top of the viewport with the error message. Either drag the offending parameter back, or ask the agent to fix it. The agent can read its own errors via the same banner.

Modals

A few things appear as full-screen overlays rather than panels:

• Project Manager — list, create, delete, switch projects. Local projects sit alongside cloud-synced ones.
• Checkpoint Panel — list of auto- and manual checkpoints for the current project. Restore reverts the model state to that checkpoint.
• Parts List Sheet — slide-over panel with the geometry-derived parts list and CSV download.
• Print Template Dialog — preview and export a scaled PDF template (page size, scale, view, component).
• Render Modal — opens after a Render finishes, shows the photorealistic image full-bleed with a download button.
• Plan Review Panel — appears when the agent has a plan ready for you to approve before it starts building.
• Paywall Dialog — appears the first time you click a gated export format, the full parts list, or a print template. Sign in / top up from here.

Keyboard shortcuts

Action	Shortcut
Send message	Enter or ⌘/Ctrl + Enter
New line in message	Shift + Enter
Reset viewport camera	R
Cancel a paste of an image	Esc in the input

The rest of the app is mouse-driven by design — the goal is to keep the surface small enough that a new user does not need a cheat sheet.

Next: Chat-driven design — how the agent actually decides what to build.