Hi everyone, I’ve been trying to find a way to isolate and sandbox an AI connected application on Linux, such as Zed Editor, so that any AI interaction can be kept separated from the host OS. Currently Zed and other apps have not deliberately implemented security and privacy protection in their foundational designs.
The problem with running Zed Editor or any other AI interactive application in a Virtual Machine is that there is still no reliable, performant GPU acceleration for virtual machines. It is quite shocking that DRM Native Context, which looks very promising, is essentially ignored in virtualisation development. Phoronix published an article earlier this year but since then nada, nothing, zip. Nothing is established and no concrete pathway to implementing DRM Native Context exists. It is mind blowing because this would enable virtualisation to explode on Linux Desktops and Workstations at a time when it is sorely needed as the world descends into a Digital ID based global technocracy.
Proxmox is ignoring GPU acceleration because it is desktop centric and not a useful technology for the datacenter which is their primary market. Same with all the other hypervisors including Incus which states that it is not a priority.
KVM announced some kind of support for DRM Native Context but nothing concrete has materialised. It is vaporware at this point.
I want to run Agentic applications on my workstation but refuse to allow it access to the host. Is there by chance something I have overlooked? Something solid, real and well implemented? Cheers.