Thousand Token Wood: shipping a multi-agent economy on a 3B model
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Thousand Token Wood is a multi-agent economic simulation built for the Build Small Hackathon, running five woodland-creature traders as agents on a 3-billion-parameter Qwen2.5-3B model served via vLLM on Modal, with a Gradio interface. The project demonstrates that small models are well-suited to real-time multi-agent simulations because they are fast and cheap enough to run a council of agents each turn, though they function as reliable format generators rather than strong reasoners. Key engineering lessons included designing deliberate scarcity to drive trade, using sharper prompts rather than larger models to improve decision quality, and reframing agent wellbeing as a recoverable mood to avoid death spirals.
Thousand Token Wood is a multi-agent economy simulation built on a 3-billion-parameter model, where woodland creatures trade goods in real-time. The project highlights how small models can efficiently handle multiple agents but require engineered scarcity and sharp prompts to ensure meaningful interactions. The simulation avoids crashes with a tolerant JSON parser and introduces dynamic elements like mood recovery to maintain engagement.