Olympus introduces a novel approach that enables Multimodal Large Language Models to coordinate and execute a wide range of computer vision tasks through intelligent task routing.

• Uses a controller MLLM to delegate 20+ specialized vision tasks across images, videos, and 3D objects
• Enables complex workflows through chained actions without training heavy generative models
• Creates a modular architecture that easily integrates with existing MLLMs
• Supports instruction-based routing for intuitive task management

For security applications, Olympus offers powerful capabilities in surveillance monitoring, object detection, and anomaly identification across various visual data types—all through a unified interface that simplifies complex visual analysis tasks.

Olympus: A Universal Task Router for Computer Vision Tasks