This adapter mirrors a Beszel Hub into ioBroker. Beszel is a lightweight server monitor: small agents run on the machines you want to watch and report to a central Hub; the adapter reads that Hub over its REST API and writes one device per monitored system.

Everything is read-only. The adapter never writes to the Hub and creates no writable datapoints.

Before you start

You need a running Beszel Hub with at least one agent connected, and a login for that Hub. The adapter authenticates as a normal Beszel user — the same e-mail address and password you use for the Beszel web interface; Beszel does not accept a username there. An admin account is not required, but multi-factor authentication must be off for that user: the adapter cannot answer the one-time code.

The user only sees the systems it is assigned to: the ones it added itself, the ones a Hub admin added it to (PocketBase admin panel at /_/, collection systems, field users), or every system when the Hub runs with SHARE_ALL_SYSTEMS=true. A user that is assigned to nothing logs in fine and sees an empty list — the connection test says so.

Setting it up

Install and create an instance. In ioBroker, install beszel and open the instance settings.
Enter the Hub URL under Beszel Hub URL — the same address you open the Beszel web interface with, for example http://192.168.1.100:8090. An IPv6 address goes in brackets: http://[fd00::1]:8090. Both http and https work; an https Hub needs a certificate the ioBroker host trusts. A Hub behind a reverse proxy keeps its path (https://example.org/beszel). Spaces and a trailing slash are removed; a URL with ?, # or a user name and password in it is refused.
Enter e-mail and password of your Beszel login.
Press Test Connection. It performs a real login against the Hub and reports how many systems your user can see — or the actual error if something is wrong: a refused login, an unreachable host, a typo in the URL.
Choose your metrics on the Metrics tab (see Datapoints and metric switches). The defaults cover uptime, CPU, load average, memory, disk, disk throughput, network and temperature. Everything else is off until you switch it on.
Save. The instance starts, reads the Hub once, and creates the object tree.

Poll interval and timeout

Poll Interval accepts 10 to 300 seconds and defaults to 60. Beszel's agents record one measurement per minute, so a value below 60 seconds produces extra requests without newer data. A value entered outside that range — for example by a script writing the config directly — is clamped rather than accepted.

Request timeout (5 to 120 seconds, default 15) is how long a single request may take. Raise it for a slow link or a Hub with many containers.

What the adapter creates

beszel.0.
├── info.connection      is the Hub reachable
├── info.systemsTotal    systems registered on the Hub
├── info.systemsOnline   of those, how many report "up"
├── info.systemsAllUp    true while all of them do
└── systems.<name>.      one device per monitored system

The device name is the system name from the Hub, lower-cased with anything that is not a letter or digit replaced by _. Two systems whose names reduce to the same id get a short hash suffix so they cannot overwrite each other, and the adapter warns once when that happens. A name without any Latin letter or digit (Cyrillic, Chinese, …) becomes sys_ plus a short hash of the Hub's system id, so it stays the same across restarts.

Renaming a system on the Hub moves it to a new device id: the adapter logs System renamed on the Hub: systems.a → systems.b, and the old tree goes — including history and other settings you made on its datapoints. Removing a system is logged the same way.

How the adapter behaves when something is missing

A system goes down or is paused. Its info.online turns false and info.status shows what the Hub says. The measured values stay at their last reading rather than jumping to zero — the adapter reports what it knows, and it knows nothing new.
The Hub becomes unreachable. info.connection turns false, every system goes to info.online: false and info.status: unknown, and the fleet counters drop to zero. The same happens when you stop the instance, so nothing keeps claiming to be online while nobody reads.
The Hub answers with an empty list. Nothing is deleted. PocketBase answers a login it no longer accepts (a changed password, a deleted user, a restored Hub database) with an empty list instead of an error, so the adapter first logs in again and asks once more. If the list stays empty, the tree stays as it is and the log says once that the user sees no systems.
The login is refused. The log says why — wrong e-mail or password, multi-factor authentication, or password login switched off on the Hub. After three failures the adapter retries at growing intervals, at most every 15 minutes, instead of sending the password every poll.
A sensor, fan, GPU, filesystem, container or any other group member disappears. Its datapoints are removed once it has been missing in two consecutive polls — a single hiccup does not clear anything.
A list is longer than the adapter reads. The adapter reads at most 50 pages per list (1000 records each for the lists of systems, containers, units and devices). On a Hub large enough to exceed that, the cut-off list is reported once and leaves the tree as it is, instead of deleting the systems at its end.

Updating

An update reapplies names and descriptions to the datapoints you already have, so corrected wording and new translations reach existing installations, not just fresh ones. The consequence is that a datapoint you renamed yourself in the admin gets the adapter's name back on the next start.

Changelog

0.19.0 (2026-09-25) — stable

New: network monitors (Beszel 0.20.0) as an opt-in metric — response time, hourly average/fastest/slowest and loss for every ping, TCP, HTTP and DNS monitor set up on the Hub
New: containers show whether an image update is available (Beszel 0.20.0)
New: btrfs filesystems appear next to the ZFS pools (Beszel 0.20.0), with their own name, the pool type and a flag for raw physical sizes
New: the connection test tells how many systems your user can see, and says so when it is none
Fixed: after a password change, a deleted user or a restored Hub database the adapter kept every system green without new values for up to a day — it now logs in again right away
Fixed: a refused login says why — wrong e-mail or password, multi-factor authentication, or password login switched off on the Hub — and the adapter stops retrying every poll
Fixed: a paused or never-connected system no longer shows uptime 0 or empty system details; its last values stay
Fixed: on current Hubs, swap, ZFS cache, GPU memory and GPU package power appeared on hosts that do not have them — they are removed
Fixed: drives without a temperature or capacity reading showed 0; they now get no such datapoint
Fixed: a storage pool that was removed came back with the next detail refresh and stayed until the next restart
Fixed: a system whose name has no Latin letters or digits (e.g. Cyrillic or Chinese) got no object tree; it now gets a stable fallback id
Fixed: two containers or group members whose names turn into the same id could swap their datapoints after a restart, and a container's id suffix changed with every re-create
Fixed: a member of a group (sensor, container, unit, …) that was missing from a single poll was deleted at once; it now has to be missing twice
Fixed: spaces and a trailing slash around the Hub URL are removed; a URL with ?, # or a user name and password in it is rejected with a clear message, also in the connection test
Fixed: on a very large Hub, the systems at the end of a long list lost their datapoints — a cut-off list now leaves the tree as it is and is reported once
Fixed: a request that trickled in slowly could run far past the configured timeout
Changed: a Hub URL that does not lead to the Beszel API (e.g. a missing reverse-proxy path) is named as such in the log and in the connection test
Changed: a renamed or removed system on the Hub is reported in the log
Changed: the login field is called E-mail — Beszel does not accept a username
Changed: the SMART verdict also knows WARNING and UNKNOWN, the pool health UNKNOWN and the vdev state MISSING
Changed: help texts, descriptions and translations corrected; drive model, serial number, firmware and host name carry more specific roles

0.18.0 (2026-09-15) — stable

New: every system carries a pictogram of its operating system in the object tree — the same icons the Beszel web UI uses, readable in the light and the dark theme
Fixed: network upload/download were always empty against a Beszel Hub 0.19.0 or newer; they carry values again, and older Hubs keep working
Fixed: disk read/write, network upload/download and swap used show 0 while idle instead of an empty value
Fixed: containers and systemd units of a system that is down or paused were deleted after a few minutes — they now keep their last values like every other datapoint
Fixed: the last SMART device, ZFS pool detail or systemd unit of a system was never removed once it disappeared on the Hub
Fixed: hardware and OS details are refreshed when a system reconnects — a new kernel shows after the reboot, not after the next adapter restart
Fixed: a system that was still pending gets its hardware and OS details on its first contact
Fixed: a Hub that is slow at adapter start no longer blanks the hardware/OS datapoints of all systems for one poll
Fixed: renaming a system on the Hub in a way that keeps its object id (e.g. only the case) now reaches the object tree
Fixed: a system added later with the same name as an existing one no longer takes over the existing system's object tree; the newcomer gets the suffix
Fixed: a container, dataset or unit whose name equals a group name (e.g. gpu, network, containers) kept being renamed while its system was down
Fixed: stopping the adapter in the middle of a poll no longer leaves late value changes behind
Fixed: after the Hub briefly reported an empty system list, the offline markers written on errors and on shutdown reached no system
Changed: temperature, battery, swap and ZFS ARC datapoints exist only on hosts that report that hardware; existing empty ones are removed
Changed: uptime, load average and agent version appear only once a system has connected; existing empty ones are removed
Changed: the ZFS error counters and the SMART power-cycle counter no longer show an empty unit in the object tree
Changed: the four "Peak values" options are gone — a Hub never delivers peak values in the minute records the adapter reads, so they never produced a datapoint
Changed: the messages of the connection test follow the system language, and the test runs with the configured request timeout
Changed: SMART and dataset text columns the Hub does not carry read as empty (null) instead of an empty string
Changed: the warning about a plain-http Hub URL is gone — http on the local network is how Beszel is normally deployed
Changed: info.uptime_text is gone — it was info.uptime a second time as text; existing installations lose it on the first start

0.17.1 (2026-09-07)

Improved: sixteen datapoints now carry an explanation in the object tree — online state, OS name, load average, container and service CPU, ZFS scrub errors and drive power cycles
Fixed: the datapoint carrying the distribution name was labelled "OS Version" — it now reads "OS Name" in all eleven languages, matching what it actually shows

0.17.0 (2026-09-06)

New: SMART data per drive as an opt-in metric — the drive's own overall verdict plus temperature, capacity, power-on hours and power cycles
New: ZFS pool details as an opt-in metric — scrub status, per-vdev error counters and the datasets of each pool
New: systemd service details as an opt-in metric — state, sub-state, CPU and memory for every unit the agent reports
Improved: the two slow detail sources are read every 15 minutes instead of every poll, so switching them on costs your Hub almost nothing

0.16.0 (2026-09-06)

Fixed: switching a metric group off now really empties it — a system that was offline at the time kept the empty channel and got it back after every restart
Fixed: a stumble while starting no longer leaves the adapter alive but silent — it keeps going and updates your values as usual
Changed: the status words of a system, of a ZFS pool and of a container are shown in your ioBroker language instead of English
Changed: a container's health is now a proper status datapoint with its list of possible values, like the system status next to it
Improved: starting up puts far less load on the ioBroker database, which shows most with many systems or many metrics switched off
Changed: user documentation now covers the ZFS pools, the root disk name and the read/write totals

License

MIT License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Developed with assistance from Claude.ai

Beszel Monitor