Multiplayer interaction data
Engine-grade records of how multiple agents act, change a shared world and experience the consequences.
A scalable multi-agent world model
Built in collaboration with RhOS.ai
Khora is a playable multi-agent world model that allows an arbitrary number of players to interact within the same generated world in real time. It is the first scalable multi-agent world model with linear-time complexity and maintains long-horizon consistency through a shared world state.
Khora scales toward interactive worlds through generalized environment support and richer gameplay interactions. It is the first public proof of Ophilus's mission: building the intelligence behind the next generation of interactive worlds.
We're a frontier lab building the data foundation for multi-agent interaction. Shared-world models can support games, robotics, simulation and embodied AI. Each requires agents to reason about actions, space and one another inside the same evolving environment.
Engine-grade records of how multiple agents act, change a shared world and experience the consequences.
Controllable, synchronized worlds for multi-agent training, simulation and embodied AI.
Signals for measuring coherence, causality and generalization across interactive worlds.
A multiplayer world model isn't just entertainment. It's infrastructure for any domain that needs shared, interactive simulation.

Generate worlds that evolve with player actions while preserving one coherent shared state.

Diverse simulated environments for training embodied AI agents without relying entirely on physical infrastructure.

Films and stories you don't just watch. You step into shared narrative experiences in generated worlds.

Drop into a generated world alongside others: genuine co-presence inside one shared reality, not avatars in a lobby.
Population-Scalable Multi-Agent World Modeling presents the architecture, evaluation and results behind Khora.

How Khora uses one shared world state to support real-time multiplayer interaction, multi-map generalization and approximately linear scaling.

Why shared AI worlds need synchronized interaction data, controllable environments and engine-grounded evaluation.

Try the shared-world demo or talk with us about research, games, and simulation partnerships.