Noema Overfit is a system designed to overcome memory limitations when running large AI models on consumer devices. It employs model paging, where non-expert weights remain in memory while expert weights are loaded from storage as needed. This allows for the execution of models that would typically exceed device memory capacity. The primary benefit is the ability to run powerful AI models on hardware like smartphones. The trade-off for this increased accessibility is a longer time to first token and slower overall generation speeds compared to models that fit entirely within memory.