Beszel: monitor everything, weigh nothing
Most homelabs monitor their Docker containers and forget the machines themselves — until a disk fills or a mini-PC cooks quietly in a closet. Beszel is the lightweight answer the ecosystem converged on: a small hub plus a tiny agent per machine, giving you history, container stats, temperatures and alerts across every host, at a resource cost you will not notice. It is the monitoring gap-filler our Docker chapter was missing.
Published Nov 18, 2026
Hub and agent: the shape of it
Beszel splits into two pieces. The hub is one container with a web UI — the dashboard, history database and alert engine. The agent is a featherweight service (also a container, or a binary) that runs on each machine you want watched and reports in. Compare the alternatives honestly: Uptime Kuma (which we run and like) answers 'is it up?' — Beszel answers 'how is it doing, and how has it been trending for a month?' The big observability stacks (Prometheus + Grafana + exporters) answer everything and cost you a weekend and ongoing rent in RAM and attention. Beszel is the deliberate middle: purpose-built for homelabs, deployable in minutes, honest about its scope.
Deploy the hub
services:
beszel:
image: henrygd/beszel:latest
container_name: beszel
ports:
- "8090:8090"
volumes:
- ./beszel/data:/beszel_data
restart: unless-stopped
Open the hub, create your admin account, and you have an empty dashboard waiting for agents. The hub's data folder holds the history — back it up with the rest of your configs and Beszel remembers your fleet's past across container recreations.
Add agents everywhere
- In the hub: Add System, name the machine, and Beszel generates the agent configuration (including the key the agent uses to authenticate).
- On each target machine, run the agent — as a container with the Docker socket mounted if you want per-container stats (you do; container-level CPU/memory is half the value), or as a bare binary for non-Docker hosts.
- Within a minute the machine appears with live CPU, memory, disk, network and — where the hardware exposes it — temperature and per-container breakdowns.
- Repeat for the NAS, the mini-PC, the Proxmox host, the VPS. The agent's footprint is small enough that 'should I monitor this?' stops being a resource question.
Alerts: the point of the exercise
- Disk usage — the alert that saves homelabs most often. Set 85% per machine; media disks fill silently and everything downstream fails weirdly.
- CPU/memory sustained — catches the runaway container and the transcoding pile-up before the family notices buffering.
- Temperature — closet machines and passively-cooled mini-PCs announce their suffering here first.
- Status (down) — overlaps Uptime Kuma; keep host-down alerts in one system to avoid double-paging yourself at 3 AM.
Delivery channels cover the usual suspects (email, webhooks into your notification stack — pair with our notifications guide if you want the house itself to announce problems). Set thresholds with hysteresis in mind: alert on sustained states, not spikes, or you will train yourself to ignore the system within a month.
Where Beszel fits our stack
Our Docker chapter's monitoring story is now three layers, each doing its own job: Dozzle for logs, Scrutiny for drive health prediction, Uptime Kuma for up/down, and Beszel for host-and-container resource history across every machine. That is not redundancy — it is the difference between 'the container restarted' (Dozzle), 'the disk is dying' (Scrutiny), 'the service is down' (Kuma) and 'the whole box has been memory-starved since Tuesday' (Beszel). Deploy the hub once; add agents as machines join the fleet.
Official documentation and references
- Beszel documentation — what Beszel is and how hub and agent fit together
- Beszel documentation — hub installation
- Beszel documentation — agent installation
- Beszel on GitHub — source code and releases
Frequently asked questions
Beszel vs Uptime Kuma — do I need both?
They answer different questions. Kuma pings services and tracks uptime; Beszel tracks resource health and history on the machines themselves, including Docker containers. Most homelabs end up with both, and this guide's stack does too.
Does the agent need the Docker socket? Is that safe?
Only for per-container stats — host metrics work without it. The socket grants broad access to that machine's Docker, so treat agent hosts accordingly: trusted LAN, authenticated hub connection, and no exposure beyond it.
How far back does history go?
As long as your hub's data volume lives — history is stored in the hub's database. Back up /beszel_data with your other container configs and your trends survive migrations.

