henrygd/beszel

Beszel is worth a look if you run a few boxes or a homelab and want historical graphs and alerts without standing up a heavyweight monitoring stack. The main check before committing is whether the hub-plus-agent model fits your network, since every monitored machine needs to reach the hub.

The problem

You have a handful of servers or a homelab and want to know, later, why something got slow: which machine was hot, which container ate memory, whether a disk is dying. Checking live metrics one box at a time makes that archaeology painful, and full monitoring stacks take time to configure. Beszel changes the workflow by putting a web dashboard in front of a small agent on each host, so you configure once and then read history and receive alerts from a single place. The output is a shared view of host and container trends rather than a pile of one-off terminal sessions.

What's different

The mechanism is a central hub — a web application built on PocketBase — plus an agent on each monitored system that sends metrics to the hub. That split keeps the heavy UI and storage in one place and the per-host footprint small, which matters when you monitor low-powered devices. History is retained at the hub, so past spikes and container behaviour remain inspectable instead of disappearing with the terminal.

The scope is broader than CPU graphs: the README lists Docker/Podman container metrics, disk I/O, temperatures, fan speeds, GPU usage and power draw for Nvidia, AMD and Intel, battery, S.M.A.R.T. disk health, and ZFS pool and dataset statistics. Alerts can be configured for most metrics across many notification services. The trade-off is administrative: agents on each host must be installed and kept able to reach the hub, so this is not a zero-touch, agentless design. Multi-user support with per-user systems and admin sharing, plus OAuth/OIDC and optional password disabling, adds setup choices that a single-user homelab can ignore but a shared deployment must decide on.

Who should try it

Good fit: a homelab or small self-hosted fleet where you control the network and can run the agent on each machine, especially if Docker stats, disk health or GPU metrics matter. Poor fit: environments where you cannot install an agent per host, or where you need the monitoring system to discover and watch machines without touching them. The documented path assumes Docker-style deployment of hub and agent; check the quick start guide before assuming your OS mix is covered.

Requirements and limitations

Every monitored system needs an agent that can reach the hub, so network layout and credentials are recurring upkeep, not one-time setup. Host-level metrics depend on what the agent can read on that OS; some sensor data, such as fan speed, is documented as Linux-only, and ZFS/S.M.A.R.T. detail depends on the host exposing it. Backups are supported to disk or S3-compatible storage, but restoring and verifying them is your responsibility. The README notes the API is commented out, so don't plan integrations around it yet.

HotRepo's take

HotRepo recorded the repository at 26,508 stars on 2026-10-24 (+28) and 26,509 on 2026-10-25 (+23), with two daily-board appearances. That is a thin snapshot: persistence of interest cannot be judged from it. The project's own claim of being lighter than leading solutions is unverified here; compare it against your existing stack before migrating. Verify agent-to-hub connectivity and per-OS sensor support on your least standard machine first.