A.U.R.A

Architectural Understanding & Rendering Agent

Upload any floor plan image. Describe your use case in plain English. Walk through a fully configured 3D scene in seconds.

A.U.R.A turns a static 2D floor plan into an explorable, furnished 3D space. Gemini Vision reads the plan's rooms, walls, and doors; a three-stage agent pipeline places furniture and fixtures for the use case you describe (a family home, an office, a clinic); and the result opens in an in-browser 3D viewer you can walk through, edit, and share.

Live demo: https://aura-bay-nu.vercel.app/

Features

• Floor plan to 3D — upload a PNG/JPG/WebP plan and get a normalized 3D scene with walls, doors, and rooms.
• Use-case-aware furnishing — describe how the space will be used and the placement agent furnishes it accordingly.
• Walk and orbit modes — explore in first person or orbit the model, with a top-down minimap.
• Scene editor — select, move, add, and delete objects in the sidebar inspector, then save a shareable copy.
• Bring your own Gemini key — visitors can paste their own Gemini API key on the landing page; it is stored in the browser and sent with uploads instead of the server key.
• Resilient pipeline — Backboard → direct Gemini → rule-based fallbacks, plus SHA-256 caching of repeat uploads.

Tech Stack

Layer	Technology
Frontend Framework	Next.js 14 (App Router)
3D Rendering	Three.js + React Three Fiber + Drei
Animation	Framer Motion
AI Vision + Language	Gemini 2.5 Flash API
Agent Orchestration	Backboard (3 sequential agents)
Database	PostgreSQL (Vultr Managed)
File Storage	Vultr Object Storage (S3-compatible)
UI Primitives	Radix UI, Lucide React
Language	TypeScript throughout

Getting Started

# 1. Install dependencies
npm install

# 2. Copy the environment template
cp .env.example .env.local
# Fill in all required variables (see Environment section below)

# 3. Run the database migrations
psql "$DATABASE_URL" < migrations/002_floorplan_ai_schema.sql
psql "$DATABASE_URL" < migrations/003_add_image_hash.sql

# 4. (Optional) Create Backboard assistants
npm run setup-backboard

# 5. Validate the setup
npm run check

# 6. Start the dev server
npm run dev

App runs on http://localhost:3000.

Environment

The root .env.example is the only active environment template. (The internal package name is still floorplan-ai.)

Required variables:

• APP_BASE_URL — Public base URL (e.g. http://localhost:3000)
• GEMINI_API_KEY — Google Gemini API key
• DATABASE_URL — Vultr PostgreSQL connection string (sslmode=require)
• VULTR_STORAGE_ENDPOINT — Vultr Object Storage S3-compatible endpoint
• VULTR_STORAGE_BUCKET — Storage bucket name
• VULTR_STORAGE_ACCESS_KEY — S3 access key
• VULTR_STORAGE_SECRET_KEY — S3 secret key
• BACKBOARD_API_KEY — Backboard platform API key

Optional variables:

• BACKBOARD_AGENT1_ASSISTANT_ID — Pre-created Backboard assistant for Agent 1
• BACKBOARD_AGENT2_ASSISTANT_ID — Pre-created Backboard assistant for Agent 2
• BACKBOARD_AGENT3_ASSISTANT_ID — Pre-created Backboard assistant for Agent 3
• BACKBOARD_LLM_PROVIDER — LLM provider for Backboard (default: google)
• BACKBOARD_MODEL_NAME — Model name (default: gemini-2.5-flash)
• GEMINI_MODEL — Override Gemini model (default: gemini-2.5-flash)

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗