Spanda is an experimental social-control layer for NPCs. Each agent notices a condition, interprets it through local directed state, shapes a bounded response, and governs what it actually does.

This guided browser demonstration holds the scene geometry constant while changing relational state. It shows how identical proximity can produce different spacing behavior, how the same policy scales from two agents to five, and how a neutral-memory control remains still under the same starting conditions.

The system is deterministic and inspectable: the visible results are tied to explicit inputs and measured movement rather than animation noise. It is not presented as consciousness or human emotion.

No language model is required for this browser demonstration. Live local-model expression is tested separately, with language kept downstream from the underlying social state.

Development disclosure: OpenAI Codex assisted with portions of the code and presentation text. The behavior shown in this browser demonstration is produced by the deterministic Spanda system rather than generated by an AI model.

Architecture and research notes:

https://github.com/putmanmodel/spanda-architectural-framework

More information