Both hands are on the mouse. You do not know what the cabinet is called. The question you want to ask takes four seconds out loud and four minutes to find in a help page — so ask it out loud.
Every piece of kitchen design software has the same silence in it. You are dragging a cabinet, something turns red, and there is nobody to ask. So you stop, open a help page, and try to guess what the thing you are looking at is called — which is hard, because not knowing what it is called is the entire problem. Voice fixes that particular silence better than any menu ever has: you say “why won’t this go where I want it” and keep your hands where they were.
Yes, but only with an assistant wired into the planner itself — a general chatbot cannot see your plan. In the Flip designer the voice assistant can list items, highlight one, add a cabinet beside an existing one, slide it along its wall by an inch amount, change a wall cabinet’s height, remove, undo and save. Plumbing, gas and structure stay your decision.
Any general assistant will happily discuss kitchens with you. It knows the work triangle, it knows quartz is harder than marble, and it can talk about all of it beautifully. What it cannot do is see the thing on your screen. Ask it “why is this cabinet floating” and it has no idea what this is.
A voice attached to the planner is a different animal, and the difference is entirely about what it can read:
What they search at midnight “kitchen design software help” — at the exact moment they cannot describe the problem in words the help page uses.
The obvious way to build this is to screenshot the canvas and hand it to a vision model. We deliberately did not, and the reason is worth knowing if you are comparing tools.
Everything on that screen is already an exact number inside the planner: the cabinet is a B30, it is thirty inches wide, it sits on wall 2, it is eight inches off the floor. Showing a picture of that to an image model and asking it to guess those numbers back takes facts we already have and turns them into an estimate. So instead, every time she needs to know what is going on, the planner hands her a short written statement of the current state:
Which is why the answers sound like a colleague leaning over: “your B30 on wall two is eight inches off the floor — that is why the toe kick disappeared”, rather than “please check whether the cabinet is elevated.”
That last exchange is the part people are surprised by: the cabinet appears in the design, in a real size, from the real catalogue, as one undoable step. She is not drawing it — she is pressing the button you would have pressed.
A voice that can change your design is only pleasant if it is boring and predictable about it. Four rules, all of them on purpose:
And the honest gaps: she cannot judge whether your kitchen looks nice, cannot see colour or finish the way you can, and cannot take a measurement of your real room. Nothing about a microphone changes the fact that the tape measure is still your job. Chapter 01 is still the chapter that matters most.
Live voice models cost real money per minute, which creates an obvious temptation: leave the line open and let it run. We went the other way, and made the state visible so you never have to wonder.
Faded means she is not listening and nothing is being spent. Solid, with a ring means the microphone is live. Press the mouse on the canvas and she fades further out of your way — the seconds when you are actually placing a cabinet are the seconds she should not be in your field of view.
The connection closes. Your microphone stays on, but nothing leaves your computer — the page just watches the level locally, and reconnects when you have actually been speaking for a moment rather than at every cough. She keeps the thread of the conversation when she comes back. Twenty minutes of silence and she releases the microphone entirely.
Why it works that way: a call is billed in blocks, so an open line during a long silence is a block you paid for and did not use. Every account gets ten minutes of voice to try, and after that voice runs on GH Coin like the other AI extras. Typing to her is free, needs no account, and is always there as the fallback.
If any of that is unavailable — a browser that will not do live audio, a microphone you would rather not grant — you land in the typing panel with an explanation rather than a dead button.
You can do a great deal of it by talking, but not all of it, and the difference matters. Voice is excellent at explaining what you are looking at, naming things, diagnosing why something will not move, and recommending and placing the next cabinet. It is poor at precise placement and cannot judge how the room looks. The realistic workflow is hands on the mouse, questions out loud.
Our 3D kitchen planner is free to design in with no signup, and Emma, the assistant inside it, can be typed to for free and anonymously. Live voice needs a signed-in account and microphone permission; every account gets ten minutes of voice to try, after which voice runs on GH Coin like the other AI extras.
In our case it is handed a short written summary of the design straight from the planner — current view, walls and ceiling heights, cabinet count, the selected item with its code, size, wall and height off the floor, any overlap warnings, your recent edits and the last message shown to you. It is not a screenshot and no image model is involved, so the numbers she quotes are the real ones rather than an interpretation of a picture.
It can change it, but only in one narrow way: after you ask what to do next, the planner produces one concrete recommendation, she reads it to you, and she places it only if you say yes. It is a single undoable step. She will not invent cabinets, resize things on a whim, or act without an explicit yes.
The connection closes after five minutes of silence. The microphone stays open after that, but nothing is uploaded — the page only measures the audio level locally so it can reconnect when you genuinely start speaking again. After about twenty quiet minutes it releases the microphone entirely. The avatar shows which state you are in: faded means nothing is being sent.
English, Mandarin Chinese and French are the defaults and she will follow whichever you use. Korean, Japanese, Hindi, Spanish and Portuguese can be chosen for a call from the language menu.
For questions, comprehension and next steps, considerably. For putting a specific cabinet at a specific inch, no — a mouse is a precision instrument and speech is not. Treat voice as the thing that removes the interruptions, not as a replacement for the drawing itself.
No. The planner and the voice both run in the browser. Voice asks for microphone permission the first time, which you can decline — the typing panel needs neither a microphone nor an account.
Draw a wall, drop a cabinet, then click the avatar in the corner and ask why it went where it went.
Open the 3D planner →