Spatial Architecture
Aeion Spatial unifies a wide range of 3D subsystems into one platform. A CRDT-synced scene graph (built on Yjs) keeps every client in sync in real time. The asset pipeline performs real polygon decimation and generates LODs automatically. WebGPU compute shaders drive real-time cloth, hair, and skin rendering with SDF-based body collision, with WebGL2 as the graceful fallback when WebGPU isn't available. Character creation and a cognitive AI architecture drive lifelike NPCs. A full physics stack (rigid bodies, cloth/soft-body constraints, fluids) handles simulation. Procedural generation, visual scripting, behavior trees, WebXR meeting rooms, and a plugin runtime round out the platform.
System Map
`
CRDT Scene Graph
```typescript
Asset Pipeline
```typescript
Rendering — R3F + Custom WebGPU
```typescript
Character + Cognitive AI Subsystem
```typescript
WebXR + Meeting Rooms
```typescript
Event Flow — Cross-Module
`
FAQ
Your R3F knowledge applies directly. You can write R3F components inside Aeion's scene graph; drop down to raw Three.js when needed; use drei + pmndrs ecosystem libraries.
Yes for the default rendering path: cloth, hair, and skin simulation run as WebGPU compute shaders with real-time SDF body collision on capable hardware, falling back to CPU simulation + WebGL2 elsewhere. A separate WebGPU path/ray-tracing service (global illumination, reflections, denoising) exists in the codebase for advanced preview use but isn't part of the default real-time scene renderer today.
Depends on hardware + scene complexity. Tested at 10K+ static objects + 100+ animated characters at 60 FPS on modern desktop; lower limits on mobile.
Plugin code runs in a scoped sandbox with errors caught so a plugin crash doesn't take down the scene, and errors are surfaced to the plugin developer + user. This is a crash-isolation boundary, not a full VM/iframe/worker-level security sandbox today — treat third-party plugins with the same trust model you'd apply to a third-party npm package.
Real-time collaborative editing is built for small, focused teams working the same scene together. For larger audiences (50+ users in one world), use multiplayer/viewer mode rather than co-editing mode.
Absolute. Tenant boundary enforced at the data layer; cross-tenant query impossible.
Universal Scene Description (Pixar's format, now industry standard). Aeion supports import + export. Apple ecosystem (Vision Pro, iOS AR) prefers USDZ.
For real-time interactive: yes (within WebGPU + GPU limits). For offline cinematic rendering (Octane / V-Ray quality): not directly; export to Blender / Maya for that workflow.
~50-100KB platform runtime; scene state syncing adds ~10-30 KB/s for active editing. For most use cases, overhead negligible vs platform value.
Aeion captures static images of scenes for OG previews + SEO. Full VR experience requires client-side rendering.