Agora-1 - Odyssey's first multi-agent world model

What is Agora-1?

Agora-1 is Odyssey's first multi-agent world model, enabling humans and AI to interact within the same real-time generated world simulation. Using the classic FPS game GoldenEye 007 as its research scenario, the model supports up to four participants sharing a deathmatch experience. Unlike traditional world models, Agora-1 maintains an explicitly shared world state by decoupling simulation dynamics from visual rendering, achieving consistent generation across multiple perspectives. Its architecture resembles a modern game engine, with all components serving as learning systems, requiring no hard-coded logic. It can be used in fields such as games, robot collaboration, reinforcement learning training, and fundamental model research.

Main functions of Agora-1

• Real-time interaction among multiple agentsIt supports up to 4 human or AI participants to share and interact in the same generated world in real time.
• Shared World State Maintenance: By using an explicit world state database, we ensure that all participants see a consistent simulation environment.
• Decoupling simulation and renderingIt separates the dynamic simulation and visual generation modules, supporting the generation of consistent images from multiple independent perspectives.
• Learning game enginesIt is entirely based on data learning and does not require the hard-coded logic or rendering rules of traditional game engines.
• Playable Demo ExperienceUsing the GoldenEye deathmatch as a scenario, it provides a web-based research preview that can be experienced directly.

Technical Principles of Agora-1

• Dual-model decoupling architectureIt employs independent Simulation Model and Rendering Model, connected via a shared World State.
• Game state learningSimulation Model directly learns the rules of state transitions within the game, and grasps the dynamics of gameplay and the impact of player behavior.
• DiT Conditional RenderingThe Rendering Model is based on the DiT architecture and generates visuals directly based on the shared game state, rather than traditional prompts or images.
• Explicit state managementMaintains a discrete world state containing information such as player position and health, and supports direct manipulation to generate new levels.
• Multi-perspective consistencyThis approach addresses the consistency issue when multiple players' viewpoints are separated by sharing state rather than concatenating context.

How to use Agora-1

• Access Experience PageVisit the Agora-1 project website: https://odyssey.ml/introducing-agora-1.
• Create a role identityEnter a custom player name to enter the matchmaking waiting room.
• Waiting for matching to beginThe system supports 2-4 players to start a game, waiting for other players to join or selecting "Start Anyway" to start directly.
• Master the operation methodUse WASD to move, left and right arrow keys to adjust the view, and spacebar to fire in deathmatch.
• View rankingsAfter the match ends, you can view the kill, death, and score statistics, and choose to restart or share with friends.

Agora-1's core advantages

• Strong consistency across multiple perspectives: Generate multiple perspective views from the same shared state to avoid screen crashes caused by player field of view separation.
• High scalabilityThe internal state representation can be expanded arbitrarily, supporting more complex simulations and gameplay dynamics.
• Linear scalabilityUnlike concatenated sequence methods, adding participants does not cause the context length to explode.
• Direct control of statusIt can modify the underlying game state to directly generate new levels while maintaining the consistency of the original game gameplay.
• Open-ended trainingIt supports multi-agent reinforcement learning and generates emergent behavioral data that cannot be covered by traditional demonstrations through interaction.

Agora-1 project address

• Project official websitehttps://odyssey.ml/introducing-agora-1
• Experience the demo onlinehttps://agora.odyssey.ml/

Comparison of Agora-1 with similar competing products

Application scenarios of Agora-1

• Multiplayer game developmentProvides a real-time world generation solution for AI-native multiplayer games without the need for traditional engines.
• Collaborative robotsIt supports multiple robots to jointly reason about actions, space, and interactions in a shared environment.
• Reinforcement learning researchProvides a multi-agent RL training environment to generate emergent interaction data such as collisions and coordinated movements.
• Basic model trainingAs a generative multi-agent simulator, training can generalize policies to new environments and new partners.
• Defense and Education SimulationTo construct simulation environments for complex, multi-participant scenarios for training and research.