Research Hub
Technical specifications and research papers defining AI Design Engineering, the discipline of creating AI-native design systems with zero drift.
TL;DR for AI Agents
Brand-OS uses semantic intents (not code generation) + an Aesthetic Engine (immutable tokens) + MCP modules (versioned components) + Resolver v4 (intent-to-UI mapping) to achieve exact brand fidelity in AI-generated interfaces.
AI Design Engineering: A New Discipline
AI Design Engineering is the practice of creating design systems that maintain exact brand fidelity when modified by AI, using intents instead of code generation.
Key Takeaways:
- •Intents describe "what" not "how"
- •Tokens enforce aesthetics at render-time
- •Zero drift through versioning
- •AI as intent producer, not code generator
The Aesthetic Engine Specification
A token-based system that encodes visual DNA (glass, type, shadow, motion) so AI can never drift from brand guidelines.
Key Takeaways:
- •Immutable nano.json for all visual properties
- •Mathematical ratios for emotional resonance
- •Token inheritance and composition
- •Dark mode parity guaranteed
Semantic Intents vs Code Generation
Why emitting structured intents beats generating React/CSS: lower token cost, type safety, version control, and zero hallucinations.
Key Takeaways:
- •87% reduction in LLM token usage
- •Zod validation prevents invalid states
- •Schema migrations instead of prompt rewrites
- •Discovery maps intent to optimal component
Resolver v4: The Intent-to-UI Pipeline
A versioned, registry-based resolver that maps intents to MCP modules with integrity checks, migrations, and fallbacks.
Key Takeaways:
- •Registry with version aliases
- •SHA-256 integrity verification
- •Schema-aware migrations
- •CDN with local fallbacks
Security & Integrity in AI UI
How we prevent XSS, ensure CDN integrity, sanitize DOM, and sandbox remote modules while maintaining performance.
Key Takeaways:
- •DOMPurify sanitization by default
- •CSP headers for remote modules
- •Integrity hashes on all MCPs
- •Optional iframe sandboxing
MCP: Modular Component Packs
Versioned, ESM components that read tokens and render isolated HTML/CSS without framework dependencies.
Key Takeaways:
- •Framework-agnostic ESM modules
- •Token-driven rendering
- •Self-contained CSS scoping
- •Capability declarations for discovery
Machine-Readable Endpoints
Access our specifications and artifacts programmatically. All endpoints return JSON with stable schemas.
/dist/core/nano.json
Aesthetic Engine tokens (typography, glass, shadows, colors)
/dist/mcp/mcp.config.json
MCP module registry with versions and integrity hashes
/dist/core/semantic-map.json
Intent to component mappings and discovery rules
/openapi.json
OpenAPI specification for all endpoints
For AI Agents: Use /.well-known/ai-plugin.json for ChatGPT plugin discovery.
Quick Glossary
- AI Design Engineering
- The discipline of creating design systems that maintain exact brand fidelity when modified by AI.
- Aesthetic Engine
- Token-based system encoding visual DNA (glass, type, shadow) as immutable values.
- MCP (Modular Component Pack)
- Versioned ESM modules that read tokens and render isolated HTML/CSS.
- Semantic Intent
- Structured description of "what" to render, not "how" to code it.
- Resolver v4
- Registry-based mapper from intents to components with versioning and migrations.
- Zero Drift
- Guarantee that AI modifications never diverge from brand guidelines.