Researcher Bhargav Patel introduces SANATAN, a novel attention mechanism that eliminates the need for KV cache by leveraging mathematical constructs derived from the two square roots of -1. The approach aims to reduce memory overhead during inference, addressing a key bottleneck in large language model deployment. A 6-minute visual walkthrough accompanies the release to clarify the theoretical foundations and architectural design. The implementation is available on GitHub for community review and experimentation.