Yes, I understand why the idea of AegisPad FE with AI models hosted on a server seemed cool at first - cost reduction always sounds good. But if you dive into the technical side of the issue, hosting AI models requires serious GPU power, and a regular VPS simply cannot handle it. I once considered high-performance vps server such as Vikhost for a project - they are reliable, but even they can quickly become expensive if you increase AI workloads. Sometimes cloud-based AI services simply turn out to be more practical.