KeystoneAI

How a sentence becomes a working floor plan.

Six stages, in order. Each one hands the next a stricter version of the house, until there is a plan that passes the checks.

The six stages

stories: 2area: 2,400 sq ftbedrooms: 3baths: 3private baths: 1garage: 1 carprimary suite: level 2front faces: southkitchen: rearlaundry: level 1extras: studybudget: mid

The survey is normalized into a usable brief

Your answers do not stay as loose text. They become structured constraints: story count, area target, garage type, primary-suite level, bathroom rules, frontage and lot context.

At a glance

Input
Guided home brief, or one plain sentence
Core engine
Deterministic layout and validation, not an image model
Output
Plan, elevations, DXF, cost range and interactive 3D
Boundary
Concept study, not permit documents

Where it stops

The output is a concept aid for discovery and a first meeting. It is not a permit-ready drawing set, not a stamped document, and not a substitute for architect, engineer or builder review.

Why a deterministic engine and not an image model?

An image model paints something that looks like a plan. The engine builds one: rooms with real areas, walls that meet, doors that connect, stairs that land. That is why the elevations, the cost range and the 3D preview can all come from the same geometry.

What if my brief has no valid layout?

The studio tells you plainly and lists what failed, instead of showing a plan that breaks the rules. Some very large or unusual briefs are outside what the engine supports today.

Try a real example

Plan around a fixed 14 × 12 ft bedroom, then inspect the DXF export of the same house. Both walkthroughs show actual Keystone outputs and explain their limits.