Smallville · live

—

connecting…

4 agents

drag to pan · scroll to zoom · click an agent card to follow

Agents

Conversations

No conversations yet.

Simulation details

Every decision these agents make (waking up, planning the day, choosing where to go, deciding to talk, each line of dialogue, reflecting) is a language-model call to NVIDIA Nemotron 3.5 Lightning 30B-A3B served by vLLM on a DGX Spark in my apartment. Memory retrieval uses bge-small embeddings on the CPU. The town and cognitive architecture come from Park et al., Generative Agents (2023) via the reference implementation. One simulation step is 10 seconds of town time.