Stoat is a Go binary that enables AI agents to easily spin up and manage QEMU virtual machines for sandboxed code execution, featuring snapshots, access control, and integration with tools like Claude.
Heya folks! I've been using CC for quite some time now, and a lot of my recent projects needed sandboxed code execution. Now docker/VMs are pretty good for this but I couldn't quite find any good agent native tools that let my agents work with them. Running them through Claude meant either approving every bash command or letting it run whatever it felt like on my machine and neither is quite right. VirtualBox and VMware assume a human clicking through a GUI, and there is nothing an agent can call. docker is fine until the VM needs systemd, kernel modules, or a loopback device, and a container still shares my kernel. So... I wrote stoat. A simple asf tool, one go binary that starts qemu directly and exposes specific tools for agents to interact with VMs on their own and safely. Configs route through a plain .toml file per project: $ stoat create dev --image debian-13 $ stoat up dev $ stoat mcp install claude-code After that Claude has tools for exec, files, packages, ports, services, logs, and snapshots all inside the same VM. If it breaks something, I restore a snapshot. Every VM has an agent_access level from none to exec, so I can hand over one guest and keep another read-only. Linux with KVM only for now tho, I'll be working on GCP/AWS connectors and OSX/Windows binaries :) PS: god bless claude cause my ass could NOT be asked to spend so much time developing an internal tooling for side-project. So... how do y'all handle VMs with agents? 👀
smol cloud is now live in beta, offering virtual computers with dynamic CPU/MEM allocation up to 4cpu/8gb, integration with smol CLI for running agent instances, and $5 monthly credit for active usage.
Krowoc is an open-source agent workspace that launches micro-VMs to give each AI agent an isolated machine with shell, filesystem, and headless Chrome. It supports multiple model providers via Fireworks or bring-your-own-key, and lets long-running tasks persist independently of your laptop.