Velo Workspaces
Guides, benchmarks, and architecture deep dives.
Everything we've written about running AI agents, Linux VMs, and disposable workspaces on Apple Silicon — how it works, what it costs, and how to set it up.
What Actually Goes Wrong When AI Coding Agents Run Unsandboxed — and Which Isolation Level Stops It
Four real 2026 incidents — a Cursor allowlist bypass, a zero-click Copilot data leak, and a nation-state espionage campaign run through Claude Code — and which level of sandboxing would have actually contained each one.
ComparisonVelo Workspaces vs. Parallels Desktop vs. UTM: Disposable Linux VMs on Apple Silicon
An honest, sourced comparison — pricing, what each one actually does with a Linux guest's GPU, and which one fits which workflow.
Complete guideZero-Tax Virtualization: Running AI Agents Safely in Velo Workspaces
A complete guide to running AI coding agents inside an isolated Linux VM on Apple Silicon, with the model — MLX or Ollama, your choice — served at near-native speed on the host GPU via AI Bridge.
BenchmarkWhat Sandboxing an AI Coding Agent in a VM Actually Costs on Apple Silicon
I benchmarked the same sandboxing architecture against two different inference engines and got two different answers about what it costs under load. Both are real. Here's why they disagree, and why I'm showing you both instead of picking the one that sounds better.
ArchitectureAn Architecture to Run AI Agents Safely and Efficiently in a Linux VM on Apple Silicon
How to let an AI agent run generated code safely in a VM without giving up GPU acceleration — the architecture behind AI Bridge, based on vsock, and the benchmark numbers behind it.
BenchmarkI Sandboxed My AI Coding Agent in a VM — It Cost Me Near-Zero GPU Speed
Three benchmarks on Apple Silicon. Single-request speed, concurrent load, and a real agentic coding loop, to measure what a Linux VM actually costs you when the model itself stays on the host GPU.
Setup guideRun an AI Agent Inside a Linux VM with Local LLMs on macOS at Near-Native Performance
A practical setup guide: MLX on the host, a disposable Linux VM for the agent OpenCode, and a vsock bridge that connects them at near-native speed — a fully local deployment.