← All work
Prompt to ProductAI tool2026

SoulClaw

A browser studio where anyone can preview, tune and export an AI character's personality — in seconds.

Role
Product designer & front-end
Year
2026
Type
Self-initiated · vibe-coded concept
Built with
Next.js · Three.js · AI-assisted coding
Live demo
EN ↗ · 中文 ↗
SoulClaw — cover
Overview

OpenClaw is an open-source framework for AI characters. Its personalities live in plain SOUL.md files — powerful, but only if you're comfortable with GitHub and a terminal.

SoulClaw turns those files into something you can see, try and take home — right in the browser, with no backend at all.

The challengeGreat AI personalities were buried in repos — impossible to preview without a local setup.

Three friction points kept coming up: templates hidden in GitHub, no way to test a personality before deploying it, and customisation that meant typing terminal commands. The goal: make a soul tangible, testable and exportable within three seconds.

Approach

How it came together

01

Perceive

An “Aura” visual language — deep-space black, micro-glass panels and a single electric-orange accent, with breathing soul orbs and claw-mark light sweeps built in custom shaders.

02

Create

A split-view studio: edit SOUL.md and IDENTITY.md side by side with a live preview, plus one-click tone modifiers — More sarcastic, Gentler, More efficient.

03

Deploy

Zero backend. The browser packages templates, avatars and configs into a ready-to-use ZIP, so nothing ever touches a server.

The work

Full project

Technical Highlights

  • Framework: Next.js 15 App Router with static export to Vercel; achieved zero server costs with full interactivity preserved

  • Graphics: Custom fragment shaders for organic soul orb animations with Perlin noise and parallax scrolling; implemented 48px rounded card design system with hover-lift micro-interactions (2px translate + 40px orange glow)

  • Performance: Optimized LCP < 1.8s through aggressive code-splitting, dynamic shader lazy-loading, and Next.js image optimization; mobile-first vertical scrolling with touch-optimized gesture handling

  • Internationalization: Complete bilingual deployment (EN/CN) with natural localization—not translation. Adapted UX patterns for Chinese users including vertical rhythm adjustments and culturally resonant character naming (e.g., "毒舌码农" Toxic Coder, "温柔陪伴者" Gentle Companion).

Key Features

  1. 3-Second Try-On: Real-time chat simulator with persona switching—test character responses against different user profiles before deployment

  2. Live Markdown Atelier: Monaco-powered editing with instant preview and syntax-aware highlighting

  3. Aura Shader Background: WebGL-powered immersive environment with mouse-follow interactions

  4. One-Click ZIP Export: Client-side workspace packaging with automatic dependency resolution

  5. Masonry Gallery: CSS Grid-based soul browser with horizontal scroll on desktop, vertical flow on mobile

Impact

  • Cost Efficiency: 100% static architecture = $0 server costs at scale

  • Accessibility: Reduced OpenClaw onboarding from 30 minutes (CLI setup) to 3 seconds (browser-based)

  • Engagement: "Try-before-you-fork" model increases template discovery and community contributions

Key decisions

Why it looks the way it does

  1. 01

    Try before you fork

    A real-time chat preview lets people meet a personality before committing to it.

  2. 02

    No server, on purpose

    Everything runs client-side: zero hosting cost, instant load, nothing to maintain.

  3. 03

    Localise, don't translate

    Chinese personas got their own names and tone rather than word-for-word copies of the English ones.

Outcome

What shipped

  • Onboarding cut from a ~30-minute CLI setup to about 3 seconds in the browser
  • 12 anime-style character portraits designed for the library
  • Fully bilingual — English and 中文
Reflection

What I learned

Building with AI as my engineer changed how I scope. I spent less time asking “can this be built?” and more asking “should it exist — and how should it feel?”