myapp.boxd.sh, subdomain proxies like api.myapp.boxd.sh, and your org’s wildcard domain) and the same TLS certificates. What changes is who can reach them. Those names now resolve and route only inside your tailnet.
This is enforced at boxd’s edge, well beyond hiding a DNS record. Someone who knows the public IP and your machine’s name gets the same 404 a nonexistent machine gets. HTTPS, SSH, and raw port forwards all move onto the tailnet together.
Management stays where it was. The CLI, console, SDKs, and boxd connect go through the boxd API, so they keep working from anywhere. It’s traffic to your machines that needs the tailnet.
Set it up
Tailscale runs per organization and the boxd team enables it together with you, so start by reaching out.1
Reach out with an auth key
Send your org name and an auth key to contact@boxd.sh, or post them in the Slack channel you share with the boxd team. Generate the key in the Tailscale admin console with Reusable on, Ephemeral off, and Tags empty. Reusable matters because boxd enrolls one device per edge server. The boxd team enables Tailscale for the org and stores the key encrypted, so edge servers added later can enroll with it too.
2
Approve the devices
boxd’s edge servers appear in your tailnet as devices named
boxd-proxy-*. If you use ACLs, allow your users to reach them on all ports. Then disable key expiry on each device, because Tailscale expires untagged devices after 180 days and your org’s access would stop with nothing pointing at the cause.3
Connect
Join the tailnet on your laptop and use your machines exactly as before:Off the tailnet, the names stop resolving.
Good to know
- Everyone needs to be on the tailnet. DNS for your machines answers with the tailnet address only, so org members outside the tailnet lose access to the hostnames. That is the point, but tell your organization before turning it on.
- Tailscale or NetBird, not both. An organization is served over one private network at a time. To move to NetBird, the boxd team turns Tailscale off first.
- Regions. The tailnet covers the regions the boxd team sets it up in, by default every region your org had at that moment, so a tailnet set up before New York existed covers Frankfurt only. Machines in New York keep their names, such as
myapp.nyc.boxd.sh. In a region the tailnet doesn’t cover, the org’s machines stay on the public internet, and creating one there is refused, a fork or a machine from a snapshot included. To add a region, ask the boxd team and send an auth key made as in step 1, since setting up the regions again takes one. Then approve the region’s newboxd-proxy-*devices and disable their key expiry, as in step 2. - Custom domains: org-wide only. Your org’s wildcard domain follows the tailnet, because boxd serves its DNS. A per-machine domain can’t follow, because its apex A record lives in your own DNS zone pointing at boxd’s public IP.
boxd machine domain addrefuses the combination and names the conflict. - Auth keys expire after 90 days at most. That only matters when boxd adds edge capacity, since an expired key means a new edge server can’t enroll in your tailnet. If a machine ever stops being reachable over the tailnet, mint a fresh key and send it over. Enabling again replaces the stored key.
- Bot protection is off on the tailnet. Only your tailnet can reach the machines, so requests over it never get the bot protection check. Webhooks, API clients, and
curlfrom inside the tailnet wake a sleeping machine as before. - Turning it off. Ask the boxd team to disable it. Your machines return to the public path, and boxd’s devices log out of your tailnet.
If machine names resolve on one network and fail on another, check the local resolver. Some corporate and router DNS setups strip answers in Tailscale’s
100.64.0.0/10 range as DNS-rebinding protection. Point your device at a resolver that keeps them, or use Tailscale’s own DNS.