Skip to content
Server monitoring

Know when a server runs out of room

Install the open-source Spectate agent with one command and see CPU, memory, disks and network every 15 seconds, with alerts when a metric stays over its threshold or a host goes quiet.

One-line install
One command installs, registers and starts the agent.
Tiny footprint
About 13 MB of memory and 1 KB of traffic a minute.
Nothing lost offline
Unsent batches wait on disk and are replayed in order.
Install

One command, then it runs on its own

Copy the install command from the dashboard and run it on the server. The agent runs as its own unprivileged user and gets its configuration from Spectate, so there's no config file to keep in sync.

  • Linux on amd64 and arm64, macOS best effort
  • One install token for a whole fleet
  • Signature-checked updates
Add a host

Run one command on each server. The agent authorizes itself with a short-lived token and starts streaming metrics.

01
Choose platform
UbuntuDebianCentOS / RHELRocky LinuxAlmaLinuxFedoramacOS
02
Run the installer
curl -fsSL https://get.spectate.net/agent.sh | sudo sh -s -- --token sk_host_••••••
03
Wait for the first report
ams-app-01 reportedjust now
CloseContinue
What it collects

CPU, memory, disks and network every 15 seconds

CPU, memory and swap, load, disk usage per mount point and traffic per interface, sampled every 15 seconds and sent once a minute.

  • 15-second samples kept for 30 days
  • Hourly min, max and average for 13 months
  • Disk space and inodes per mount point
  1. Kestrel
  2. Hosts
  3. ams-app-01
ams-app-01Reporting
System uptime
41d
Since 13 Sep
CPU
8cores
AMD EPYC 9454P
Memory
61%
9.8 of 16 GB

Telemetry

CPU and memory, 15s samples

CPUMemoryalert above 85% for 10m
Alerts

Alerts on thresholds and on silence

Alert when disk usage stays above 90% or memory runs low, for as long as you choose so short spikes don't wake anyone. A host that stops reporting raises a no-data alert.

  • Must hold for up to 60 minutes before alerting
  • No-data alert after 5 minutes by default
  • Routed like every other alert
In detail

Everything in server monitoring

On-disk spool

Up to 50 MB or 24 hours of batches kept during an outage.

No duplicates

Replayed batches are recognised, so samples never double.

Clock correction

Samples are shifted to match when a server's clock drifts.

Open source

Written in Go, published on GitHub, checksummed releases.

Self-monitoring

The agent reports its own spool size, failures and memory.

Same price as a monitor

A host counts as one monitor on your plan.

Included on every plan. See pricing

FAQ

Server monitoring: questions

How do I install the Spectate server agent?

Create a host in the dashboard and copy the install command. It is one line, curl piped to sh with your install token. The script downloads and verifies the binary, creates a system user, registers the host and starts a systemd service. Linux on amd64 and arm64 is supported; macOS works on a best-effort basis.

Which server metrics does Spectate collect?

CPU usage split into user, system, I/O wait, steal, idle and more; memory and swap; load averages for 1, 5 and 15 minutes; uptime; disk space and inodes per mount point; and bytes, packets, errors and drops per network interface. Samples are taken every 15 seconds.

How much overhead does the agent add?

Very little. The agent is a static binary of about 10 MB that uses about 13 MB of memory. It sends one gzip-compressed batch per minute, about 1 KB for a typical server with two disks and two network interfaces. It runs as an unprivileged user under a hardened systemd unit.

What happens when the server loses its connection?

The agent keeps collecting. Batches that can't be sent go to an on-disk spool of up to 50 MB or 24 hours and are replayed in order once Spectate is reachable again. The server removes duplicates. If no data arrives for 5 minutes, or the time you set, Spectate raises a no-data alert.

Can I alert on high disk or CPU usage?

Yes. Add a metric condition to the host, for example disk usage above 90 or CPU usage above 80, choose a severity and how long the value must stay over the threshold, from immediately up to 1 hour. The alert follows the host's escalation path and includes a chart of the metric.

Is the Spectate agent open source?

Yes. The agent is written in Go and its source is published on GitHub under the Spectate organisation, so you can read exactly what runs on your servers. Releases are checksummed, and updates installed with spectate-agent update are signature-checked.

Know about it first.

Set up your first monitor in under a minute. 14 days free, no credit card.