Your servers.From your terminal.

A lightweight server monitoring and control tool for Linux machines.

Coming soon. ETA 2027See the TUI
alrith - spark-01LIVE 00:00:00
CPU - 8 CORE
034%
152%
228%
361%
444%
519%
673%
738%
MEM / NET
MEM 31.2/54G58%
SWAP 1.9/16G12%
DOWN84.2 Mb/s
UP12.6 Mb/s
HOSTspark-01
GPU - GB10
UTIL64%
MEM 60/128G47%
TEMP61C
PWR118W
JOBtrain.py
PIDPROCCPU%MEM%
1842python train.py61.218.4
921alrith-agent1.80.6
1204docker dind8.44.1
402sshd0.40.2
1systemd0.20.3
1567nv-mgmt2.11.2
q quit/ filterarrows move

Kernel signals, CPU, memory, processes, and GPU in one view

One terminal for a single box or four Sparks

Rows of rack hardware in a data center

Rack hardware, the kind Alrith watches

Stupidly simple setup.

No cluster to bootstrap. No dashboard to host. Just the CLI and a machine you can reach.

Install
Install the CLI.

One static binary for Linux on your workstation.

Setup
Run setup.

Three prompts. No account and no cloud to join.

$alrith setup
Connect
Connect to the conductor.

Point alrith connect at the conductor and the TUI opens live. The conductor URI works over Tailscale the same as LAN.

$alrith://user:pass@ip:port/

Four Sparks. One terminal.

A box of DGX Sparks meshes on your LAN or Tailscale network. One is the conductor. Connect once and watch kernel signals, GPU, and processes on all of them.

$alrith://user:pass@ip:port/
MESH - LAN4 HOSTS - 1 VIEW
spark-01CONDUCTOR
MESHPEERS LINKED
  • spark-02PEERGPU 61% - KSIG 3
  • spark-03PEERGPU 44% - KSIG 1
  • spark-04PEERGPU 72% - KSIG 5
YOUR TERMINAL -> SPARK-01alrith://user:pass@ip:port/

The conductor serves the aggregated view. Local machines only. Works on Tailscale IPs the same as LAN.

Three rules. No exceptions.

Simple.

One CLI drives the whole workflow. The agent is a single Go binary with no sidecars to run.

Open.

MIT licensed. That will not change. Self-hosted from day one. Your config and metrics stay on your network.

Local.

Built for Linux servers, GPU boxes, and DGX Sparks. Compatible with Tailscale from day one. If you can reach it on your LAN or Tailscale network, Alrith can watch it.

One small agent on each machine.

One Go binary on each machine. It reads CPU, memory, disk, network, GPU, and kernel signals, then talks directly to your TUI. Nothing leaves your network, and you can inspect and control processes without leaving the terminal.

  • +Single Go binary, no sidecars to run
  • +Collects CPU, memory, disk, network, GPU, and kernel signals
  • +Metrics stay on the machine, nothing phones home
  • +Watch live in the TUI, stop and inspect processes
  • +MIT licensed, self-hosted, yours to keep. That will not change
In development

Alrith is coming soon.ETA 2027.

MIT licensed and self-hosted from day one. We will never change the license.