Hey Product Hunt I built miniVE for the moment I realized I was letting coding agents run `rm`, `curl | bash`, and `pip install` against my real home directory, my real SSH keys, my real everything. The cloud sandboxes that fix this (E2B, Modal) are great, but they're metered, account-gated, and your code leaves your machine. miniVE is the local, free version: a disposable Linux box on your own hardware you can point an agent at, or use to try any repo without wrecking your setup. - Hand your agent a box it can't escape: it thrashes `/workspace`, not your host - Pick a runtime: Python 3.10-3.12, Node 18-22, or blank Ubuntu - Drop in files or clone a repo by URL: git runs in the box, not on your machine - Real multi-tab terminals; installs persist until you say otherwise - Map ports and preview web apps in-app - Delete = truly gone. Container and volume removed; your machine stays clean. Honest about the threat model: this is container isolation, not a microVM. It stops "my agent nuked my home dir" and "that repo leaked my .env," not a determined kernel escape. For your own agents on your own code, that's the right trade. No cloud, no account, no telemetry; everything it creates is auditable with plain docker commands. MIT licensed. Requires Docker Desktop. Windows, macOS, and Linux. Would love your feedback, especially what's missing before this becomes your default box for agents and untrusted repos.

miniVE

A local throwaway sandbox for AI agents & any repo

A local throwaway sandbox for AI agents & any repo