GenArch

Generative Architecture — AI-powered building generator for Graphisoft Archicad.

Describe a building in plain English, and GenArch draws the complete 3D model — walls, doors, windows, slabs, columns, beams, roofs, rooms, stairs, furniture, and more — directly into a running Archicad instance.

"2-story modern house with 4 bedrooms, open-plan living, and a flat roof"
                              ↓
                    AI Planner (Ollama Cloud / DeepSeek / Ollama)
                              ↓
                    Structured JSON building plan
                              ↓
                    C++ Add-On inside Archicad
                              ↓
                    Complete .pln model drawn automatically

Quick Start

1. Install the Add-On

git clone https://github.com/Eliahhango/GenArch.git
cd GenArch
.\install.ps1

This copies the pre-built Add-On into Archicad's Add-Ons folder. Restart Archicad.

2. Generate a Building

cd python
pip install -r requirements.txt
python main.py --backend ollama-cloud "1-bedroom cabin with gable roof"

Requires: Python 3.10+ and an API key (Ollama Cloud or DeepSeek).

Features

• 14 building element types — Walls, Doors (wall-linked), Windows (wall-linked), Slabs, Columns, Beams, Roofs, Zones, Stairs, Curtain Walls, Meshes, Railings, Furniture, Objects
• Multi-floor — Any number of stories with per-floor element arrays
• 3 AI backends — Ollama Cloud (recommended), DeepSeek API, Ollama Local
• PDF/image import — Sketch a floor plan and let AI turn it into a 3D model
• Architecturally aware — The AI follows real architectural conventions (wall thickness, room sizes, structural logic, passive solar principles)
• Multi-style — Modern, Victorian, Japanese, Brutalist, Mediterranean, Prairie

Setup

1. Python

cd python
pip install -r requirements.txt          # or pip install requests ollama pydantic
copy .env.example .env                   # then edit .env with your API keys

2. Add-On

Copy the compiled .apx into Archicad's Add-Ons folder:

%APPDATA%\GRAPHISOFT\ARCHICAD 29\Add-Ons\

Restart Archicad 29. To build from source, see Building the Add-On.

3. Generate a Building

python main.py "3-bedroom house with pool and 2-car garage"

CLI Reference

Command	Description
python main.py "description"	Generate and draw in Archicad
--dry-run	AI plan only, no Archicad
--output-json file.json	Save the generated plan to file
--input-json file.json	Replay a saved plan (no AI call)
--backend <name>	ollama-cloud, deepseek, or ollama
--model <name>	Override the AI model
--import-file <path>	Import a PDF/image for AI analysis
--verbose	Detailed logging

Examples

# Preview without Archicad
python main.py --dry-run "modern 2-story office with curtain walls"

# Save and reuse a plan
python main.py --output-json plan.json "L-shaped villa with courtyard"
python main.py --input-json plan.json

# Import a floor plan sketch
python main.py --import-file sketch.pdf

# Choose backend and model
python main.py --backend deepseek "small cottage"

Building the Add-On

Requires Visual Studio 2022, CMake 3.19+, and the AC29 DevKit.

cd addon
.\build_ac29.bat

Output: addon/Build/AC29_NMake/AchicadAutomation.apx

Architecture

python/main.py           → AI generates JSON building plan
python/archicad_builder.py → Sends JSON to Archicad via Add-On Command API
addon/Src/AddOnMain.cpp  → Registers 16 commands
addon/Src/Create*.cpp    → Each command creates an element type

Elements are created in dependency order: Walls first (GUIDs captured for door/window linking), then Slabs, Columns, Beams, Doors, Windows, Roofs, Zones, Dimensions, Stairs, Curtain Walls, Meshes, Railings, Furniture, Objects.

Project Structure

├── addon/               # C++ Add-On source (.apx)
│   ├── Src/             # 16 command pairs + helpers + entry point
│   ├── build_ac29.bat   # One-command build
│   └── CMakeLists.txt
├── python/              # Python CLI
│   ├── main.py          # Entry point
│   ├── backends/        # AI backends (DeepSeek, Ollama, Ollama Cloud)
│   ├── schemas.py       # Pydantic models
│   └── architect_soul.py # Master architect AI prompt
└── USER_GUIDE.md        # Full documentation