Shuru is a lightweight sandbox that spins up Linux VMs on macOS using Apple's Virtualization.framework. Boots in about a second on Apple Silicon, and everything is ephemeral by default. There's a checkpoint system for when you do want to persist state, and sandboxes run without network access unless you explicitly allow it. Single Rust binary, no dependencies. Built it for sandboxing AI agent code execution, but it works well for anything where you need a disposable Linux environment.
The agent stack is splitting into specialized layers and sandboxing is clearly becoming its own thing. Shuru, E2B, Modal, Firecracker wrappers.
Earlier this month I wrote about how these layers have very different defensibility profiles and why going monolithic is the wrong call: https://philippdubach.com/posts/dont-go-monolithic-the-agent...
EDIT: Spelling
I've noticed claude forks parallel agents on an assigned task. How would they communicate in isolated sandboxes like these? Would it be cleaner and more effective for a harness to orchestrate swarms of agents in a single clean linux environment like OrbStack?
Use OrbStack. It’s faster than Virtualization.framework because it has its own hypervisor.
[dead]
What does local first mean in this context? Does it just mean local? Like, the software runs locally?