Exo Architecture Enables AI Agents to Autonomously Self-Edit Their Operational Harnesses

The Exo framework allows AI models to iteratively modify their own underlying architecture while executing work. This system incorporates safety protocols adapted from operating systems research to prevent instability. The project is framed as a form of online continual learning.