zebrr / Thing-Sandbox Star 2 Code Issues Pull requests Autonomous LLM-powered simulation exploring LLM agent behavior with subjective memory and emergent narratives through multi-phase intention-arbitration-memory ticks. Arbiter resolves scenes, memory degrades via FIFO summarization, simulation mutates even in empty locations. python experimental openai autonomous-agents emergent-behavior llm-agents ai-simulation text-simulation Updated Dec 11, 2025 Python