Start with what's running under you

Everything else builds on one mechanism. Apple Silicon Virtual Machines, Explained covers what the Virtualization framework is, and how it works goes device by device. If you're weighing virtualization against the alternatives, VM vs. container vs. sandbox on macOS and Virtualization.framework vs. QEMU cover the isolation and emulation trade-offs. ARM64 vs. x86-64 and why ARM VMs feel native cover the architecture underneath.

Get Linux running, and know its gotchas

Since Velo Workspaces 1.2, the fastest start is an official cloud image: Ubuntu, Debian, Fedora or Rocky Linux, ready to sign in to about a minute after it downloads. What's new in 1.2 walks through it. Each distribution has its own post with what's different about it: Ubuntu, Debian, Fedora, Rocky Linux, Arch, Kali and NixOS, plus which one fits your work, Server or Desktop, and how VM networking works. For something other than Linux, see macOS guests.

When arm64 isn't enough

x86 software doesn't vanish on Apple silicon. Rosetta 2 in Linux VMs translates individual binaries, can you install x86 Linux at all answers the whole-OS question, and running x86 Docker images covers containers. If something won't run and you don't know why, the triage guide sorts it out.

Running AI agents and local models, safely

This is the deepest part of the blog, best read in roughly this order: how AI coding agents use your computer, what they can access, the Git-specific risk, and what goes wrong unsandboxed, with dated incidents rather than hypotheticals. Then the complete guide to running agents in an isolated VM ties it together, backed by the AI Bridge architecture and what leaves your Mac and what doesn't. On the model side: tools and formats, how much RAM you need, what quantization does, and local vs. cloud models for agents. To pull it together: a safe local AI agent environment checklist.

A disposable-by-default workflow

The habit this blog argues for: what disposability means, how it differs from a snapshot, and how base images, snapshots and clones relate. Applied: one VM per Git branch, clean-room bug reproduction, testing untrusted software, the QA and DevOps workflow, matching your CI environment, and making it a daily coding habit. Managing the images underneath: versioning base images and tuning the guest before you save one.

What it costs, measured

Why ARM VMs perform the way they do, then the deep dives on CPU, disk and network, sizing guides for memory and how many VMs you can run, and what decides boot time. For AI workloads: the inference benchmark and what sandboxing costs across two inference engines.

How it compares

Against other Mac VM apps: Velo Workspaces vs. Parallels vs. UTM, and moving a VM over. Against containers: Velo Workspaces vs. Docker Desktop vs. OrbStack, Apple Containers vs. Docker Desktop vs. OrbStack and Apple Containers vs. Linux VMs. Against the cloud: Velo Workspaces vs. GitHub Codespaces vs. Ona and the trade-offs underneath. Building agents for research? See the AI researchers guide.

Where to start

  • New to VMs on Apple silicon: the fundamentals, then create an Ubuntu workspace from the cloud image.
  • Running an AI coding agent and want it safe: the incidents first, then the complete guide.
  • QA, DevOps or CI work: the disposable VM workflow is the quickest path to the payoff.
  • Deciding whether you need this at all over Docker or a cloud IDE: the comparisons above, read against your own constraints.

This page grows with the blog, so it's worth coming back to. Or skip the reading, download Velo Workspaces, and try it yourself.