Self-hosted fleet hub for Linux boxes and Macs

Every Linux box,
one place.

A tiny Rust agent on each box dials out to your hub — like a GitLab runner. You get the box list, their web ports and SSH through it. No inbound ports, no VPN, no port forwarding.

curl -fsSL https://<your-hub>/install.sh | sudo bash -s -- --token <registration-token>

That's the whole onboarding. Your hub's address and a token — the web app shows the exact command under Add box.

Tether web app listing boxes with online status and ports

The whole setup in one picture

Boxes dial out to the hub. You log in to the hub's web UI from any device, see every box, open its web ports with a click and drop into a shell on it right in the browser — or ssh in from your own terminal.

How Tether connects Linux boxes, the hub and your devicesSeveral Linux boxes behind NAT run the tether agent and open one outbound connection each to the tether hub on Kubernetes. A Mac, an iPhone and a terminal connect to the hub over HTTPS and reach the boxes' web ports, a terminal in the browser and SSH through it, so shell commands run on the boxes without any inbound port.Linux boxes, anywherebehind NAT, LTE or home routers · no inbound portYour hubYou, from any devicebox1 · office serverDebian · sshd :22 · Grafana :3000tether agent · systemdpi-shop · Raspberry PiPi OS · aarch64 · behind LTE routertether agent · systemdvm-7 · cloud VMUbuntu · x86_64 · firewall closedtether agent · systemdoutbound WSS onlytether-hubon Kubernetes (Helm) or plain Docker▸ web UI: log in, see every box▸ click a box → its web ports▸ terminal on the box, in the browser▸ ssh from your laptop via the hub1 container · 25 MB RAM · SQLiteMac · hub web UIclick box1 → Grafana :3000 via hubor Open terminal → shell, no sshiPhone · same web UIshop Pi offline? tap it, open itsadmin page or a terminal on the go$ tether ssh box1box1:~$ sudo apt upgrade -yweb UI · HTTPS loginShells run on the box: your terminal ⇄ hub ⇄ box1, or the terminal in the browserssh stays end-to-end encrypted with your own key · the browser terminal runs as the unprivileged tether user and is audited

How it works

Run the hub

One container on Kubernetes (Helm chart included) or anywhere Docker runs. SQLite inside, nothing else to operate.

Add a box

Create a registration token in the web app and paste the one-liner on the box. The agent registers itself and installs a service that starts at boot — systemd on Linux, launchd on a Mac.

Connect

Click a port to open its web UI through the hub, or tether ssh box1 from your laptop — plain OpenSSH, tunnelled through the hub.

Built for boxes you don't control the network of

Outbound only

Agents open one WebSocket to the hub and keep it alive. Works behind NAT, LTE, hotel Wi-Fi and corporate firewalls.

SSH stays SSH

The hub forwards ciphertext as a ProxyCommand. Your keys, your known_hosts, your ssh config — nothing new to learn.

HTTP ports, one click

Grafana, a Jupyter notebook, a device admin page: every listening port shows up and opens through the hub, behind your login.

One static binary

5 MB, musl, no dependencies, x86_64 and aarch64. 3 MB of RAM at runtime, under a hardened systemd unit. Footprint →

Macs, too

The same one-liner tethers a Mac (Apple silicon or Intel, macOS 11+), or brew install it. Web terminal, ports and tether ssh work like on any box. Mac install →

Tokens like a runner

Registration tokens onboard boxes; each box then holds its own secret. Revoke a token without touching registered boxes.

Two-factor sign-in

Optional TOTP second factor per user — Microsoft Authenticator, Google Authenticator, 1Password or any TOTP app. The web UI and the CLI both ask for the code.

Audit log

Every login and failed attempt, password and 2FA change, token, user and box change, and every SSH or HTTP session through the hub — with who, when and from which IP. Kept 180 days.

Fleet facts at a glance

Agent version, OS release, kernel, CPU count, RAM and disk usage of every box, refreshed every 15 seconds for metrics and every 5 minutes for system facts.

Self-hosted freeware

Free of charge for personal and commercial use. The hub runs on your cluster, so box names, metrics and tunnels never leave your infrastructure.

Small footprint, on the box and on the cluster

Measured on the current release, idle with one box connected. No database server, no Redis, no message broker — one process on the box, one container on the cluster.

On every Linux box

  • 5 MB static binary, zero dependencies, nothing to apt-get
  • 3 MB resident memory, one process, idle CPU ≈ 0 %
  • 1 outbound connection, a 15 s heartbeat of a few hundred bytes
  • Runs as an unprivileged user; fine for a Raspberry Pi or a 512 MB VPS

On Kubernetes

  • 1 container, 1 replica, 1 Gi volume — the whole Helm chart
  • 25 MB resident memory; chart requests 50 m CPU / 64 Mi
  • 40 MB image pull (101 MB on disk), web app and agents included
  • SQLite inside the container: backup = copy one file

Or just Docker

  • Same image runs on a single VM with docker run
  • Starts in under a second, ready to onboard boxes immediately
  • Hundreds of boxes per hub: each idle connection costs a few kB

Live view of every box

Hostname, IPs, OS, kernel, load, memory, uptime and listening ports — refreshed with every heartbeat.

Box detail page with metrics, HTTP ports and SSH instructions

Ready in ten minutes

Helm chart, Docker image and static agents are on the download page. The docs walk you through the first box.

Get Tether