Been thinking about how to support sandbox machines / vms / docker containers for Halo.
How would you prefer to use and connect to a sandbox env in an AI Agent?
// ... or a container on this machine: no account, no network.
tool Local : Remote("docker", image: "python:3.12-slim", network: false)
// A container that can reach PyPI and nothing else: pip works, nothing else gets out.
tool Deps : Remote("docker", image: "python:3.12-slim", hosts: ["pypi.org", "*.pythonhosted.org"])
// A server you already have, over your own SSH keys and ~/.ssh/config.
tool Box : Remote("deploy@build.example.com", connect_timeout: 10)commit c43597e