Server monitoring is here: CPU, memory and disk from a one-command agent
Spot Downtime now monitors what's happening inside your Linux servers. Install the agent with one command and get alerted on full disks, pinned CPU and servers that stop reporting.
· 5 min read · By the Spot Downtime team
Until now, Spot Downtime watched your servers from the outside: pinging the host, opening its ports, loading its websites. That tells you the moment something has broken. It can't tell you that a disk is 93% full, or that CPU has been pinned for ten minutes and the site is about to start timing out. Server monitoring can.
Starting today, every workspace can add a Server monitor. A small agent on the server reports CPU, memory, disk and network every minute, and you're alerted when something stays over a limit, or when the server stops reporting at all. One server is included on the Free plan.
| Problem | Uptime checks | Server agent |
|---|---|---|
| Server off or unreachable | Yes | Yes |
| Website returns errors | Yes | — |
| A port stops accepting connections | Yes | — |
| CPU pinned at 100% | — | Yes |
| Memory running out, swapping | — | Yes |
| Disk filling up | — | Yes |
| Network traffic spikes | — | Yes |
How it works
The agent is a single small program, about 7 MB, that runs as a service on your server. Once a minute it reads the system's own counters, the same ones top, free and df read, and sends them to Spot Downtime in one HTTPS request.
spotdowntime-agent
reads /proc every minute:
CPU · memory · disks · network
stores every sample
no report for 3 min → down
over a limit → resource alert
Slack · Teams · Discord
email · PagerDuty · webhooks
mobile push
- Nothing connects to your server. The agent only sends. There's no port to open and no firewall rule to add.
- It can't be told to do anything. The agent never runs commands or code it receives. It reads numbers and sends them.
- It's locked down. It runs as its own unprivileged user, in a systemd service with a read-only view of the system and a 64 MB memory cap.
- It doesn't lose data. If the network drops, it keeps up to 3 hours of samples and sends them when it's back.
Set it up in two minutes
- In Spot Downtime, click Add monitor and choose Server.
- Give it a name, like
web-1, and choose when to be alerted. The defaults are CPU or memory averaging over 90% for 5 minutes, and any disk over 90%. - Open the server's page and copy the install command. It includes the server's private token.
- Run it on the server:
curl -fsSL https://spotdowntime.com/install-agent.sh | sudo sh -s -- --token YOUR-TOKENThe script downloads the agent for your CPU (x86_64 or arm64), checks its checksum, creates the service and sends a test report. If the token is wrong or the server can't reach us, it says so and stops. Within a minute, the server's page fills in:
Alerts that wait for a trend
A healthy server hits 100% CPU all the time: during a deploy, a backup or a burst of traffic. Alerting on every spike would bury the alerts that matter. So CPU and memory alerts use the average over a window you choose, from 1 to 30 minutes, and only fire when the whole window is over your limit.
Disks are different. A full disk breaks things straight away (databases stop writing, logs stop, deploys fail), so a disk alerts the moment it crosses its limit, and the alert names it: Disk /var/log is 93% full, above the 90% limit. Every resource alert sends a second message when it's back to normal, with how long it lasted.
| Alert | When it fires | Goes to |
|---|---|---|
| CPU high | Average over the window is above the limit | Chat, email, PagerDuty, webhooks, push |
| Memory high | Average over the window is above the limit | Chat, email, PagerDuty, webhooks, push |
| Disk high | Any disk is above the limit | Chat, email, PagerDuty, webhooks, push |
| Server down | No report for 3 minutes (2 to 10, your choice) | Every channel, SMS and calls included |
A server that stops reporting is treated like any monitor going down: it opens an incident, follows your escalation policy, and shows on your status page. Resource alerts are warnings, not outages, so like slowness alerts they don't use text or call credits.
What it costs
| Plan | Servers | History |
|---|---|---|
| Free | 1 | 30 days |
| Starter | 5 | 30 days |
| Pro | 25 | 30 days |
| Enterprise | 250 | 30 days |
Each server also counts as one of your plan's monitors.
What's next
The agent runs on Linux today. Windows and macOS agents are next, followed by longer history and a forecast of when each disk will be full. Until then, watch Windows servers from outside with ping, port and website checks.
Remove it any time
--uninstall. It stops the service and removes the agent, its user and its settings. Delete the server in Spot Downtime to remove its history too.Get new guides by email
Practical posts on monitoring, outages, error fixes and on-call, when they're published. No spam.
Keep reading
- Monitoring · Servers · UptimeKubernetes liveness and readiness probes vs external uptime monitoringProbes keep pods healthy inside the cluster. They can't tell you users can't reach the site. How liveness, readiness and startup probes work, and what to add outside.November 15, 2026 · 7 min read
- Servers · TroubleshootingServer agent not reporting? How to get it back in five minutesYour server shows as down with "the agent hasn't reported", or never showed up at all. How to check the agent's service and logs, fix a refused token, a blocked connection, a failed install or a machine that went to sleep.November 5, 2026 · 6 min read