Matrices is a platform of training environments for multimodal LLM agents learning to complete real computer tasks. It targets the path toward “self-driving computers,” where AI can work reliably and safely across interfaces, apps, and the web.
Realistic computer-use environments
Matrices simulates practical desktop scenarios so agents can practice step-by-step interaction and improve with feedback.
- Working with windows and app UIs
- Browser-based tasks and web navigation
- File and folder operations
- Interface actions like clicking, typing, switching, and choosing options
Built for agent research and development
Matrices is designed for researchers, engineers, and teams building and evaluating LLM agents. It supports systematic testing of how an agent performs when it must both understand instructions and take concrete actions inside a UI.
Toward more autonomous assistants
By combining realistic scenarios with a controlled environment, Matrices helps accelerate development of AI assistants that can carry out longer sequences of computer actions with less ongoing human involvement.
