Building AI civilizations in real-world environments
How AI beings remember, plan, and live together in the real world.
Long-horizon tasks
Persistence
Cognition
System 2
Free will
Emergence
Multi-agent worlds
Self-replication
Models
Internal LifeSim AI models that run the beings of LifeSim.
Memory · Oct 2026
LifeSim-memory-0.1
Ranks the memories of an AI being by how much each one helps the decision it makes now, and gives it the ten that matter. Trained on LifeSim worlds, with labels from three judge models.
1 ms latency on CPU
103 decision trees over 43 features
Social · Oct 2026
LifeSim-social-0.1
Our internal LLM-based System 1 model, used in agent-to-agent and human-to-agent interactions. It decides how an AI being reacts when someone approaches it: decline, a brief chat, carry on, suggest later, or accept.
20 ms latency on CPU
92 to 95% accuracy
Publications
What we learned from building LifeSim, and the earlier research of the founder.
