A production web-app platform — assembled by AI, maintained by humans.
Every template you build today becomes tomorrow's ecosystem.
Built by Neoulsoft Inc. (est. 2019, Seoul) — Open source initiative launched April 2026
10+ production apps running across KR / JP / ZH markets
NSKit is a production web-app platform designed for the era when AI writes most of the code. The artifacts it produces are not mockups, pitch decks, or prototypes — they are Self-Contained templates that run in production, maintained by humans, composed by AI.
This puts NSKit in a different layer from AI visual-artifact generators such as Claude Design, v0, Lovable, or Figma Make. Those tools are excellent at producing quick visuals; NSKit is where those ideas become services that actually run and earn. We treat them as complementary inputs, not competitors.
Core philosophy — "Bound by structure, free to combine": uniform structure is the precondition for infinite composability, like LEGO blocks.
| Traditional stacks (React/Vue/Next) | NSKit | |
|---|---|---|
| File structure | Small modules + imports | Self-contained (HTML + CSS + JS in one file) |
| Build pipeline | Webpack / Vite / esbuild required | None — F5 refresh = live |
| AI token cost per change | ~25,000 tokens | ~800 tokens |
| Parallel AI sessions | Merge conflicts likely | Zero conflicts |
| Final output | Deployable bundle | Running production service |
| Template reuse | Copy-paste, abstract later | Template Mixing — compound over time |
NSKit is opinionated about one thing: the framework constrains structure so that AI can safely compose templates. Everything else — visual style, domain logic, UX — is free.
NSKit accepts design input from two complementary paths:
[ AI Design Tools (Claude Design, v0, Lovable…) ] ──┐
↳ Fast ideation, landing pages, one-off screens │
├──→ NSKit Template Converter
[ Pencil (.pen) structured design ] ─────────────────┘ ↓
↳ Production components, design systems Self-Contained templates
bidirectional sync, stable node IDs (Part / Page / Panel / Dialog / Scene)
↓
Template Mixing ecosystem
Whichever path feeds the converter, the output conforms to the same discipline: PART.define() structure, Self-Contained closure, Template Mixing compatibility. Source diversity, destination uniformity.
| Project | Description |
|---|---|
growing-pool-cache npm |
Self-growing cache pool for AI content — 97% cost reduction |
| ai-native-design | Design principles for AI-optimized code structure |
| command-handler | AI-friendly backend API dispatch for Spring Boot |
| csw | Claude Subscription Worker — $200/mo subscription → production API |
A living AI companion that runs entirely on local hardware — personality, emotion, memory, and 3D presence.
| Project | Description |
|---|---|
| nvatar | Project hub — architecture, stack, and documentation |
| nvatar-demo | Live demo — 3D room, voice chat, web search |
| nvatar-sdk | BehaviorPattern SDK — build custom avatar behaviors |
| nvatar-code-assist | Code Assist — Claude Code integration via MCP channel |
| avatar-chat | Prompt engineering patterns for local LLM character AI |
| chat-like-human-memory | 9D emotion + personality evolution + 3-tier memory |
| customize-local-llm | Local/cloud hybrid routing — personality local, facts cloud |
| vrm-studio | 3D VRM avatar chat room with Three.js + WebSocket |
| portable-ai-companion | Cross-app franchise architecture — personality portability (CC BY-NC-SA) |
- 2026 — Framework stabilization, template refinement from live B2B projects (GosuSchool, NewMyoung, Prism, Chicver, Haru, BigFoot), Template Studio internal build-out
- 2027 Q1 — Template Studio public launch, Creator program, compounding ecosystem
- Beyond — Ecosystem that grows as creators ship: every new template widens the combinatorial space
This is a long game. NSKit is not trying to be the fastest way to generate a mockup. It is trying to be the most durable way to build and maintain web services in the AI era.
NSKit is developed and maintained by Neoulsoft Inc.
7 years of technical innovation · 150+ projects · 10+ industry domains