It learns your network, then tells you when it changes.

ON-DEVICE AI

Three things it does, and none of them leave your computer.

Every vendor now writes "AI" on the box. Here is exactly what ours does, so you can judge it. There is no language model and no cloud inference — this is statistics that run on your own machine, on your own data, and keep running when the internet is down.

1. It learns what each interface normally carries

A static threshold is wrong twice. Set it at 80% and the backup link that always runs at 90% pages you nightly; set it high enough to silence that and you miss the access port that quietly went from 2 Mbps to 40. DeviceWatch instead builds a baseline per interface, per direction, at three levels at once: an overall moving average, a weekday-versus-weekend profile, and a per-weekday profile — each in 30-minute slots. Monday 09:00 is compared against what Monday 09:00 has looked like, not against a number someone typed in a year ago.

In a 21-day simulation over 30,240 samples with four injected faults, it found all four and raised zero false alarms. That number matters more than the detection rate: an anomaly detector that cries wolf gets muted within a week, and a muted detector detects nothing.

2. It learns which alarms you ignore

This is the part most monitoring tools never do. DeviceWatch watches what you do with each alarm. If you dismiss the same one five times, or acknowledge it within two minutes without acting, it learns that this alarm is noise on this device — and stops emailing it. It keeps showing it on screen; it just stops waking you.

It also learns the opposite. If a particular alarm keeps being followed by that device going down within the hour, DeviceWatch marks it a precursor and tells you so the next time it fires — the port errors that always come before this switch drops.

Two rules keep this honest. It never silences a device outage, no matter how often you dismiss one. And only your actual decisions are stored — the statistics are rebuilt from the event log every time, so nothing drifts into a black box you cannot audit or reset.

3. It forecasts when you run out of room

Robust trend regression over the stored history projects when a link, a disk or a memory pool reaches capacity. Robust, specifically, so that one bad afternoon does not become a fictional crisis three months out.

COVERAGE

Any brand that speaks SNMP.

Reachability, interfaces, traffic, errors and link state come from IF-MIB, which is identical on every manufacturer. CPU, memory, temperature and power have no such standard, so DeviceWatch carries vendor profiles — and we mark plainly which ones we have confirmed against real hardware and which come from vendor documentation alone.

Confirmed against actual devices: Cisco IOS / NX-OS, Cisco Small Business, Fortinet FortiGate, HP / Aruba ProCurve, Dell iDRAC / OpenManage, Silver Peak / Aruba EdgeConnect, and the standard MIBs that cover servers, Linux, Windows and appliances such as Palo Alto. Juniper, MikroTik and Huawei profiles ship from documentation and are labelled as unconfirmed inside the app.

A profile that turns out to be wrong costs you nothing — the device answers "no such object" and that one reading is simply absent. And when a device genuinely exposes no health data at all, as a cloud-managed access point does, DeviceWatch reports unknown rather than green. A device that has told us nothing has not told us it is well.

CAPACITY

How many devices can I add?

The licence covers up to 250 devices, and that is also the number we are willing to stand behind. It is not a marketing figure, but it is a projection: the sustained test ran at 40 devices, and the 250 figure is extrapolated from it. The binding constraint is disk, not polling. If your network is near that number, email us before you buy and we will size it with you.

Installation and administration guide → — first run, settings, firewall rules, and connecting it to MeshWatch Central.

Minimum requirements. macOS 12 Monterey or newer (Apple silicon or Intel), Windows 10/11 or Windows Server 2016 or newer with Desktop Experience (64-bit), or 64-bit x86 Linux with a desktop environment. 4 GB RAM and 2 CPU cores for up to 50 devices; 8 GB RAM and 4 cores for up to 250. Disk grows with history — budget 225 KB per monitored interface per year, so 250 devices at 24 ports is about 1.6 GB annually. UDP 161 outbound to your devices; UDP 162 inbound only if you want to receive traps. The full measured breakdown ships with the app as SYSTEM-REQUIREMENTS.md.

WHAT YOU GET

Seven screens, no upsells between them.

Dashboard — devices monitored, what is not responding, open incidents, ports up, saturated links, and the busiest ports right now. Interfaces — every port with traffic, errors, discards, utilisation and history graphs. Health — CPU, memory, disk, temperature, fans, power supplies and modules, read from wherever that particular manufacturer puts them, using the device's own warning and critical thresholds when it publishes them. Map — a topology built from LLDP and CDP neighbour data, colour-coded so a hardware fault does not look like an unreachable device. Alarms — grouped by cause, so one core switch going down is one entry and not thirteen, with per-type control over what leaves by email and quiet hours that outages deliberately ignore. Reports — PDF export and scheduled delivery. Settings — polling intervals, SMTP, SNMP credentials in the operating system keystore, and full configuration backup.

Discovery takes a single IP, a range, or a CIDR block, tries your community strings in order, and tells you what it found — manufacturer, model, port count, and whether the device supports 64-bit counters, which you want to know before you rely on its traffic graphs on a gigabit link.

The interface is English by default, with Japanese, Korean and Simplified Chinese available.