Agentless by default, extensible by script
Two questions get asked in the first ten minutes of any monitoring trial. Does it need an agent on every server? And what happens when the thing you need to watch is not in the catalog?
Key points
- Windows over WMI, Linux over SSH, network gear over SNMP, hardware over IPMI and Redfish — nothing installed on the hosts.
- Any numeric WMI property, or the output of a command, script, SQL query, file check or pushed reading, becomes a sensor with a graph and a threshold.
- A compiled MIB set names the common vendors' SNMP counters; upload a device's own MIBs for the rest.
- Custom sensors do not add to the license count.
- Syslog, SNMP traps and Windows events come in; email, webhooks and scripts go out; integrations get scoped API tokens.
On this page: Agentless Beyond the catalog Events in Data out Smart groups
The first question decides whether the rollout gets past change control. The second decides whether the tool still fits the network in three years, when the network has acquired a vendor appliance, a line-of-business database and a batch job that only one person understands. SysPeep answers the first with no, and the second with an escape hatch that takes whatever a shell, a query or a script can measure and treats it exactly like a built-in sensor.
Agentless where it counts
What is reached, and how
- Windows servers over WMI: services, Hyper-V memory pressure, DNS server counters, terminal sessions, battery charge, and temperatures and fan speeds where LibreHardwareMonitor is on the host
- Linux hosts over SSH
- Network gear, printers, firewalls over SNMP v2c and v3, with v1 and v2c traps received
- Server hardware over IPMI and Redfish through the management port
- Hypervisors and containers from vCenter, Proxmox and Kubernetes
- Anything exporting Prometheus or OpenMetrics, scraped rather than duplicated
Why it matters
Nothing to install or patch on the server itself, and nothing to get past change control host by host. A failing fan is seen over the management port before the operating system notices — and while the operating system is down.
Remote probes exist for branches, DMZs and OT segments, but a probe is one component per segment that reports back over mutual TLS, not an agent on every host. The number of things you have to deploy, patch and explain stays small.
When the catalog is not enough
The catalog holds 280 sensors across 38 collection methods, in five categories and forty sub-categories you can search by what a sensor measures, and every sensor can be tested against the device before it is added. Every site still has the odd system that none of them quite fit. For those, SysPeep makes a WMI property, or the output of a command, script or program, a sensor: it is graphed, thresholded and alerted on like any other.
| The odd thing on your network | How it becomes a sensor |
|---|---|
| A Windows counter nobody pre-packaged | The WMI explorer: browse the host's classes property by property, with sample values and units where the host reports them, and turn any numeric property into a sensor with no script to write |
| A switch whose interesting counters live only in the vendor's MIB | Named out of the box for Cisco, Juniper, Arista, Fortinet, Palo Alto, HPE, Dell, APC, MikroTik, Synology and the standard IETF and IEEE modules; for anything else, upload the device's own MIB files and the names appear, with missing dependencies and duplicates reported |
| A backup job that proves it ran by writing to a drop-folder | A file and folder check on the drop-folder, with a threshold on what you expect to find there |
| A nightly batch job on a box that cannot be polled | A passive push sensor: the job POSTs its own reading in when it finishes, and the reading gets a graph and a threshold like any polled value |
| A vendor appliance whose only interface is a web status page | An HTTP sensor that checks status, latency and the text or JSON the page should contain |
| A line-of-business application with a SQL backend | A database query sensor: any SQL result becomes the value — order backlog, queue depth, replication lag |
| A Linux box, an embedded device or anything reachable only by SSH | An SSH command sensor: whatever the command prints is the reading |
| A log that grows when something is wrong | A file check on its size, with a threshold on the growth |
| "Is the whole checkout path healthy?" | A calculated or composite sensor built from the web journey, the database query and the queue depth, with its own threshold and its own alert |
| A metric a developer already exposes for Prometheus | Scraped as OpenMetrics; no second exporter |
Custom sensors cost nothing extra. Licensing counts devices, not sensors, so the escape hatch is always open and nobody has to budget for using it.
Nothing is hidden by discovery, either: every sensor is offered on every device, and the ones discovery did not detect are marked with what was recorded and when, so a protocol the sweep missed is still one click away. Before every test and discovery scan a card names the credential that will be used, with its non-secret details, and the result says which credential was tested — no more guessing whether a failed test used the device's account or the global default.
Events in, alerts out
The single service also receives events: syslog lines, SNMP traps and informs, Windows event log subscriptions, and its own flow notices when a router stops exporting (the syslog server and trap receiver page covers this side in full). Rules turn them into alerts by kind, application and severity, and a built-in tester shows which rule wins for a given event and which rules are shadowed by earlier ones. The thousand events that do not matter stay in the events timeline, where they are still there when you need to know what happened right before something broke.
Getting data out again
- Notifications go out by email, webhook or script, so chat, ticketing and paging tools receive the incident in whatever shape they expect. There is no per-integration licensing.
- API tokens with read or write scope give scripts and integrations their own credentials, shown once at creation and revocable individually.
- Reports export as CSV for whatever spreadsheet or system wants the raw numbers, or, in the paid editions, PDF for the people who want the chart.
- Batched device-down notifications and digests of up to 24 hours keep the integrations from being flooded during a storm, and a test-send from the console proves each channel works before the night it is needed.
Organization that maintains itself
Smart groups have membership rules — subnet ranges, device types, discovering probe, manual pins — with subgroups and first-match precedence, editable in place. A new switch discovered in the branch subnet joins the branch group and shows up in the right place without anyone filing it. Decommissioned devices retire with their history intact; test boxes that were never monitored are deleted cleanly.
The practical result is a tool that grows with the estate instead of needing to be re-organised every time the estate changes, and a catalog you can leave behind without leaving the tool.
Bring the script, point it at the box
Install the free edition, add the one system nothing else could monitor, and watch its output turn into a graph with a threshold. Up to 5 devices stay free at full depth; the 30-day trial unlocks probes, scheduled reports and PDF export.
Download Free TrialNext: Evidence for the auditor