Turn written text into rich, story-driven 3D concept visualizations using local LLMs or offline extraction engines.
- 3D concept visualization: Immersive PlayCanvas-powered scenes with orbit, fly, and follow camera modes
- Chat-driven input: Type concepts in a floating CodeMirror editor — extraction pipeline maps text to 3D layers
- Observation modes: Pluggable renderers (graph, morality, etc.) that map extracted concepts to different 3D spatial metaphors
- Layer-based composition: Build scenes from composable layers with per-entity animations and CRDT-ready fractional ordering
- Prefab system: Named entity templates for consistent visual language across observation modes
- Guided onboarding: 4-step first-run wizard — welcome, NLP model pre-download, LLM configuration, starter scene selection
- Scene management: Create, clone, rename, and switch between concept scenes — persisted in IndexedDB via Dexie.js
- Settings reset: Danger-zone option to wipe all app data with double-confirmation glass modals and auto-reload countdown
- Hex dial controls: Gamepad-inspired hexagonal dials for scene actions, camera modes, and theme switching
- Vim-like input mode: Command mode for shortcuts, input mode for text fields — status bar shows active mode
- Adaptive theming: System/light/dark theme cycling with glass-morphism UI
- Voice input: Web Speech API for hands-free text entry
- Four extraction engines: LLM (OpenAI-compatible), NLP (compromise.js), Keywords (RAKE), Semantic (TF.js + USE)
SvelteKit 2 (Svelte 5) | TypeScript | Tailwind CSS 4 | PlayCanvas | Dexie.js | CodeMirror 6 | Space Grotesk
npm install
npm run devOpen http://localhost:5173 for the 3D visualizer.
| Key | Action | Key | Action |
|---|---|---|---|
| W/A/S/D | Pan camera | ? | Toggle keyboard help |
| Z/X | Zoom in/out | F | Toggle fullscreen |
| C | Look at origin | H | Toggle controls |
| Shift (hold) | Toggle orbit/fly | I | Enter input mode |
| Space | Cycle follow target | Esc | Close / exit input mode |
| O/;/./,/M/K | Hex dial faces | L | Switch hex dial bay |
| 1-9 | Select fan-out option |
| Engine | Requires | Speed | Quality |
|---|---|---|---|
| LLM | Running OpenAI-compatible server | 2-10s | Best |
| NLP | Nothing (offline) | <100ms | Good for structured English |
| Keywords | Nothing (offline) | <10ms | Basic co-occurrence |
| Semantic | ~30MB model download (cached) | 2-5s | Good semantic understanding |
Configure the LLM endpoint and model in Settings.
- DSL Reference — Entity spec, animation primitives, layers, prefabs, observation modes
npm test # Run vitest
npm run build # Production build
npm run check # Type checkMIT