Not long ago I created an app for fast and private inference. Was easy to create the version 0.1 but so hard to continue developing the next. Now I can proudly call it upgraded because: 1 Good coding setup with Sessions as mother and tabs as children, with persistence memory in local storage. 2 Supports image to text or vision uploads (great for, create code to build a website like this image). MMproj support. 3 Preview mode, see what your code does right away. 4 Privacy friendly, your project stays private until you decide to share it. 5 GUI friendly, choose LLM, load and voila. 6 Terminal for better understanding. 7 GPU Nvidia and some more, and CPU support. No data collection and free. I set a 1 time password for fun that is Welcome to Hugston Any suggestion on how to optimize the memory for 100k code lines?