Khora: Population-Scalable Multi-Agent World Modeling

World models have advanced visual prediction and interactive generation, but multi-agent environments expose a fundamental scalability limit: most methods are bound to a fixed number of agents.

At Ophilus, together with RhOS, we built Khora to address this constraint. Khora decouples shared world-state evolution from visual rendering and uses a population-agnostic interface to generate each agent's observation.

Our architecture supports arbitrary agent populations at inference time without retraining while preserving cross-view consistency and scaling approximately linearly with the number of queried views in real-time interaction.