Abra is a text-to-XR engine. Describe a scene in plain language — Abra generates a complete, interactive WebXR experience you can explore on any device. Phone. Headset. Laptop. AR. VR. Web. One URL, no app install.
Built in February 2026 — one month before Google announced Vibe Coding XR.
- Describe — Pick a curated scene or type your own prompt
- Generate — AI creates a complete A-Frame WebXR scene with 3D models, lighting, animations, and interaction logic
- Explore — Open the scene in your browser. Walk through it in VR. Drop it on your desk in AR. Share the link with anyone.
Every generated scene is a self-contained HTML file. No build step. No dependencies. Just a URL.
| Phone | Headset | Desktop | |
|---|---|---|---|
| 3D Explore | ✅ Touch + D-pad | ✅ Head tracking | ✅ Mouse + WASD |
| VR Mode | — | ✅ Quest, Vision Pro | ✅ (with connected headset) |
| AR Mode | ✅ Android Chrome | ✅ Quest 3 | — |
| Share via URL | ✅ | ✅ | ✅ |
Tested on Apple Vision Pro, Samsung Galaxy Z Flip 5, iPhone, and desktop Chrome.
Abra ships with curated scene prompts that showcase different environments and interactions:
- 🏕️ Forest campsite with tent, campfire, and surrounding trees at dusk
- 🪐 Solar system with planets orbiting the sun you can walk through
- 🏴☠️ Pirate island with a ship docked at a wooden pier and treasure chests
- 🚀 Spaceships and asteroids where you can shoot lasers
- 🏰 Medieval castle with towers, walls, and a courtyard
- 🏘️ Suburban neighborhood street at golden hour
- ❄️ Frozen wilderness with snow-covered pines and a frozen lake
Each scene is generated from a single text prompt.
- Frontend: Next.js, TypeScript, Tailwind CSS, App Router
- AI: Claude by Anthropic (model-agnostic architecture — swap in any LLM)
- 3D Rendering: A-Frame (WebXR framework built on Three.js)
- 3D Assets: 75 CC0 models from Kenney.nl (nature, pirate, space, castle, suburban, industrial)
- Deployment: Vercel
- AR/VR: WebXR Device API — no native app, no SDK, just the browser
User prompt
→ Claude generates complete A-Frame HTML
→ Self-contained scene with 3D models, lighting, animations, interactions
→ Preview in-app
→ Share via URL
→ Open on any device: phone, headset, desktop
→ AR mode (Android) | VR mode (Quest, Vision Pro) | Web (everywhere)
The system prompt (SYSTEM_PROMPT.md) is the engine. It encodes spatial design principles, A-Frame component patterns, cross-platform compatibility rules, and a curated 3D model library. This is where domain expertise becomes product differentiation. See SYSTEM_PROMPT.md for the full spatial design system that powers scene generation.
- Cross-platform by default — A-Frame targets every WebXR-capable browser, not a single headset ecosystem
- Model-agnostic — Currently powered by Claude, but the architecture works with any LLM
- No $1,799 headset required — Works on the phone in your pocket
- Real 3D, not video — Generated scenes are actual WebXR with geometry, physics, lighting, and interaction — not AI-rendered frames
- One URL — Every scene is a shareable link that works on any device
Brandi Kinard — Design Engineer & AI Developer
Solo-built. From architecture to prompt engineering to cross-device QA.
- Clutch — AI agent on Ray-Ban Meta smart glasses for real-time step-by-step task guidance (video)
- Living Draft — Autonomous pipeline: smart glasses capture a space → AI publishes a website, no human at keyboard (video)
- Zero Touch Content Engine — One git push auto-generates and publishes a complete content kit
- Face Game Controller — OpenCV face detection as hands-free game input
MIT