This is the whole loop, end to end. It takes about ten minutes and it leaves you with a machine that keeps running when you stop looking at it.
1. Get the command
tnt is a single static binary. There is no runtime to install and no language toolchain to set up.
# macOS (Homebrew)
brew install tnthub/tnt
# macOS and Linux (install script)
curl -fsSL https://tnthub.com/install.sh | sh
# any platform — download the binary directly
# https://tnthub.com/downloads/tnt_<version>_<os>_<arch>
Check that it works:
tnt version
2. Create a sandbox
tnt box create --name research
You get a machine with an SSH server already running on it. The default image carries the usual build tooling; anything else you need, you install yourself.
See what you have:
tnt box list
tnt box describe research
3. Get in
tnt box ssh-proxy ssh research
tnt box ssh-proxy ssh starts a local proxy that carries the connection to the sandbox, and hands you an ordinary SSH session. Your editor, your terminal multiplexer, scp, rsync and your ~/.ssh/config all work the way they do against any other machine.
The sandbox never listens on a public port. The connection is always initiated from your side, so there is no inbound firewall rule to get wrong.
4. Put your work in it
Over SSH, with the tools you already use:
git clone git@github.com:you/your-service.git
cd your-service
npm ci
The point of doing this here rather than locally is that the process keeps running after you disconnect. A test suite that takes forty minutes finishes. An agent with a long to-do list keeps going.
5. Run one command and come back
You do not need a session for this:
tnt box exec research -- npm test
exec runs one command and returns its output. It is the path to use in a script or a CI job — no proxy, no session, no interactive shell.
6. Take your results out
This is the step people skip, and it is the one that loses work.
# out to your machine
tnt box exec research -- tar czf - ./reports | tar xzf -
# or straight from the sandbox
tnt box exec research -- git push origin my-branch
7. Destroy it
tnt box destroy research
A destroyed sandbox is gone. There is no recycle bin, and nothing is inherited by the next person or the next tenant. If nobody touches a sandbox for the idle window, it is torn down automatically and its contents go with it.
When something goes wrong
Every sandbox keeps a log, and you can read it without a shell:
tnt box logs research --follow
If you would rather look at the machine itself, tnt box ssh-proxy ssh research still works — the sandbox is not hidden from you while you are diagnosing it.
What to do next
- Running an agent in the sandbox — the case the sandbox was built for.
- The sandbox documentation — lifecycle, snapshots, and where it runs.
- The CLI reference — every command and every flag.