metaMe Runtime

The one-line description
metaMe Runtime is a generative interface engine: instead of a fixed UI, Claude generates a Surface Plan (overlay vs drawer vs embed vs micro “Liquid UI”) from intent + device + constraints, then metaMe renders the experience and verifies it with a Design Parity loop.
This makes the UI adaptive, fractal, and design-led — not “chat with tools.”

The 30-second pitch
“Most AI apps are a chat window. metaMe is different: you talk to metaMe, and Claude doesn’t just answer — it decides what the screen should become. A single request can spawn a tiny capsule badge, expand into an embedded card, then open a drawer for context, then promote into a full overlay for deep work — all based on device, content type, and design constraints. The UI is generated as an explicit plan, audited in real time, and stays stable and beautiful.”

Why it maps to “Generative Interfaces” at the top tier
The core innovation loop (this is what judges want)
User intent → Claude reasons → generates Surface Plan → UI reconfigures → Parity verifies → (auto-fix) → UI updates
That’s a novel interaction loop enabled directly by Claude’s reasoning.

How it hits top tier criterion
1) Working Prototype & Execution
deterministic Surface Selector + constraints + parity checks = stability
fast state (Redis or simple in-memory) + caching of plans

2) Interface Novelty & Playfulness
our novelty isn’t “pretty UI.” It’s fractal interfaces with “doors” that unfold.

What makes it feel playful
Capsules as living UI badges that “breathe” into drawers/overlays

A visible “UI metamorphosis” animation (even simple)
A small “why did the UI change?” affordance that reveals the plan

Has a Surface Plan HUD toggle (tiny):
shows the current surface ladder + the chosen density

“Claude picked Drawer because: mobile + act intent + module profile”
That’s surprising and judge-friendly.

3) Theme Alignment: Generative Interfaces
Claude is not generating content only.

Claude generates UI structure: surface, density, module composition, progressive disclosure.

The output is a plan, not prose.
“Claude generates the interface plan, not just the answer.”

4) Leveraging Claude’s Capabilities
Long context: keeps cartridge/codex/capsule state coherent

Multi-step reasoning: resolves conflicts between module profiles, device constraints, and user intent

Instruction following: obeys DIS/Constraint Manifest/Parity expectations

Optional multimodal: user drops an image/diagram → Claude turns it into a module + surface plan
our “UI Plan + Parity + Auto-fix” loop is the magic:
a basic LLM can chat

but reliably planning UI across devices and constraints is where Claude’s reasoning shines