@OwariDa
Post
Post 1 of 2
Running a full desktop environment in a microVM, and accessing it remotely over LAN (Omarchy, ping @DHH and @vaxryy 😊) QVM now supports ultra low latency desktop streaming, for purposes such as VDI / Virtual Desktop Infrastructure, RBI / Remote Browser Isolation and agent sandboxes Supported host-OS:es and hypervisors include both Linux/KVM, macOS/HVF and Windows/WHPX VirGL-acceleration is available for guests, and both X11 and Wayland based desktop environments are supported Hardware-accelerated H264 encoding is performed on the host side, avoiding the need to expose direct GPU access to the guest 
Video thumbnail from X post Watch video
Post 2 of 2
When connecting to local VMs it's possible to use a native viewer that skips the H264 video encoding step, but a use-case I personally really like is to host VMs on dedicated boxes on my LAN and then connect to them from a laptop, my phone w/ AR glasses or my VR headset to get a safe and isolated workspace that still feels like I'm using a local native desktop:) Some bonuses: \- Supports docker-less Dockerfile-based VM building, and pulling images from dockerhub and other OCI compliant registries \- Integrates a custom scriptable TLS-intercepting MITM proxy that keeps the source browser TLS fingerprint \- Integrated LLM router/proxy on the host side, allowing for API tokens etc to be kept out of the VM \- Supports egress filtering, which out of the box denies the VM from accessing any LAN resources \- Supports proxy/VPN chaining outbound traffic \- Root-filesystem via virtiofs, extremely space efficient. Can build a base image and instantly create multiple VMs from it, where each VM only stores changes (which also makes filesystem changes very easy to track!) \- Root-filesystem is FUSE-mountable from the host Particularly useful for agents, that can run both within and outside of the VM context: \- Custom browser automation within the VM, avoiding the normal bot detection traps and allowing for executing arbitrary scripts directly in Chrome extension context (supports any chrome/chromium-based browser, Manifest version 3 compliant) \- Mouse and keyboard input, screenshot and video capturing accessible via APIs It's perfect for both interactive use as well as agent sandboxes, and what I like the most, a combination of the two. Being able to let an agent jump in and automate some task within a VM I'm using interactively, as well as to be able to jump in interactively into a VM that an agent is controlling, is super convenient
Explanation
What it says QVM can run a full graphical Linux desktop inside a microVM and stream it over LAN with very low latency. It supports Linux/KVM, macOS/HVF, and Windows/WHPX hosts; X11/Wayland guests; VirGL graphics; and host-side hardware H.264 encoding, so the guest does not need direct GPU access.
It is aimed at both human-operated virtual desktops and agent sandboxes: OCI/Dockerfile image building, TLS-intercepting proxying, host-side LLM/API-token routing, default LAN egress blocking, VPN/proxy chaining, copy-on-write-ish virtiofs roots, browser automation, and API-controlled keyboard/mouse/screenshots/video.
Context The interesting model is: run disposable, strongly isolated desktops on dedicated LAN machines, then interact with them from a laptop/phone/AR/VR device. An agent can take over the same desktop you are using, or you can jump into one the agent is controlling.
Why it matters This collapses VDI + browser isolation + agent sandboxing into one system. The particularly useful security property is keeping credentials and GPU access on the host while giving the VM—and therefore an untrusted agent—a realistic GUI/browser environment with tightly controlled networking.
Images The screenshot shows the actual remote desktop UI: terminals plus YouTube running simultaneously, with connection/codec/bitrate controls and H.264 streaming, suggesting this is a working interactive demo rather than just an architecture proposal.