I hear you on the RAM issue, Lalit. This is definitely on our roadmap. Let me explain what's happening:
Currently, we run Whisper locally and the system tailors the local model size to your computer's specs. If you have more RAM, it loads a bigger model. If not, it loads a smaller one to minimize memory usage.
The solution we're planning is dynamic, on-demand AI model loading - instead of keeping the model in memory permanently, we'll load/unload it between uses. This will unlock our ability to do even mor on-device AI, which is very memory-intensive when running but shouldn't eat resources when idle.
While this solution is on the roadmap, I can't promise a near-term fix. We've got some foundational issues I need to tackle first - mainly understanding the codebase thoroughly so Claude and I can make improvements without breaking functionality. It's a complex problem I'm working through. I don't have it working yet, but I think I'm close solving the problem with AI.
Sorry you're experiencing these issues. We have a solution planned, but I'm not able to implement it just yet. I understand how frustrating it must be to see the app sitting there eating memory when you are not using it - especially when it's causing stability issues. This was my great fear with the current system. I honestly expected way more complaints about this. I've been grateful / surprised that more people don't complain.
Before we have dynamic model loading, we will have smaller, faster and more accurate local AI models, which will eat less RAM.
Much work to do. Thank you for your patience. 🙏
