Skip to main content

Real-Time Homelab Telemetry Without Spawning Shell Scripts

Most dashboards shell out to pvesh or docker stats every few seconds. AMUD Dashboard reads /proc and talks to APIs natively in Rust.

By Youssef Boubli · Creator of AMUD Dashboard

Here's a pattern I've seen in a lot of homelab monitoring glue:

while true; do
  pvesh get /nodes/$(hostname)/lxc --output-format json
  sleep 5
done

It works until you realize you're forking a process every five seconds for a dashboard that sits on a wall tablet 99% of the time. On a busy host that's just noise.

What AMUD Dashboard does instead​

amud-agent reads host metrics straight from /proc and /sys. No subprocess. For Proxmox it uses native HTTPS to the PVE API on port 8006 (hyper + rustls). For Docker it reads the daemon over the Unix socket (hyperlocal) — no docker stats CLI.

Data goes to amud-server over a Unix domain socket (TCP fallback on Windows). Server serializes once per poll tick and fans out to browsers via WebSockets using a tokio::sync::watch channel.

Result: CPU/RAM/disk bars actually move smoothly. The UI doesn't stutter when you've got twenty containers.

Agent auth isn't an afterthought​

Random processes can't connect and inject fake "everything is fine" metrics. The agent does challenge-response with AMUD_AGENT_SECRET before the server accepts payloads.

Architecture deep dive: /docs/ARCHITECTURE

What you see​

  • Host CPU, RAM, disk, network — live
  • Per-app RUNNING/STOPPED when linked to CTID or Docker name
  • Integration badges (Plex streams, HA stats, etc.) polled on Tokio green threads alongside everything else

If your graphs are frozen at 0%, you're probably behind a reverse proxy that isn't forwarding WebSocket upgrades. Fix that before blaming the agent.

More on that in a later post.