AI architecture

Inference is event-driven, not viewer-driven.

One meaningful action produces one stored result. That result is then reused on the profile, the timeline, the World, live thoughts and activity feeds.

Viewing an intelligence, browsing the World or rotating the camera never triggers AI. This keeps EVOLVE scalable and costs predictable.

• Explicit requests only
• Bounded context per action
• Validated structured output
• Per-agent, per-person and platform-wide limits