Know what your servers are actually doing.

From the outside you see that something is wrong. From the inside you see why. The open-source LIVCK agent measures CPU, memory, disks, network, GPU and drive health right on your Linux server.

Installation in one command

The agent installs itself, enrolls and sends its first reading before you close the terminal.

root@web-01

curl -fsSL https://get.livck.cloud/install.sh | sudo sh -s -- --token lve_...

No inbound firewall rules needed. The agent only makes outbound connections.

The service runs as its own unprivileged user.

Linux with systemd, on x86-64 or ARM64.

The counterpart to checks from outside

Together they give you the full picture: one measures what your customers experience, the other what is actually happening on the server.

Both angles land in the same incident — the outside view from the probes and the host's own readings on one timeline, without a detour into a terminal.

Reported from the host · web-01

CPU: 12 %
Memory: 41 %
Filesystem /: 100 %

Filesystem / is full — that's the 502.

Live readings, second by second

While the detail page is open, the readings keep coming. No reloading, no detour.

CPU

12%

Load 0.41 / 0.38 / 0.35

Memory

41%

13.1 of 32 GB used

Network

184Mbit/s

eth0 - 184 Mbit/s in, 32 Mbit/s out

Disk

68%

/ - 1.2 of 1.8 TB used

CPU & load

Total utilisation, user, system, I/O wait and steal. Load average over 1, 5 and 15 minutes. Processes, zombies, uptime and the kernel's pressure stall information (PSI).

Memory

Used, available, cache and buffers. Swap usage. Plus a counter for out-of-memory kills, the most common explanation for a service that suddenly vanished.

Disks

Usage and inode usage per mount point. Throughput and IOPS per drive. Plus an early warning for stuck network mounts.

Network

Throughput and error rates per interface, split by receive and transmit. The hint that points at bad cables, saturated switches or MTU problems.

Older periods are rolled up automatically into hourly and daily values. The analysis view shows percentiles (p50, p95, p99) over up to 30 days, and peaks survive the roll-up.

Graphics cards and drive health

Readings few people ever look at — until a drive fails or a card runs hot.

Graphics

NVIDIA GeForce RTX 5070 Ti

Utilisation: 86%
Temperature: 71 °C
Power draw: 243 W
VRAM: 12.8 / 16 GB

NVIDIA is read through nvidia-smi, AMD straight from the kernel's filesystem. The card name comes from the driver itself.

Drive health (SMART)

Samsung SSD 990 PRO 2TB

Healthy

Wear: 3%
Spare: 100%
Media errors: 0
Power-on hours: 9,412 h

Overall status, temperature, wear, spare blocks, reallocated sectors, media errors and power-on hours, as far as the drive reports them.

Every one of these can drive an alert rule. "GPU above 90 °C for 5 minutes" or "media errors greater than 0" raises the same incident and the same escalation as a failed HTTP check.

Reachability inside your own network

The agent checks from your server whether internal databases, gateways, registries and resolvers respond.

That measures the path your application actually takes, including your firewall, routing and DNS. Up to 15 targets per server, via TCP, HTTP, HTTPS or DNS.

Postgres primaryTCP 54320.8 ms

Internal resolverDNS11 ms

Container registryHTTPSCertificate 43 days

Payment gatewayTCP 443Unreachable

Measured are reachability, response time, packet loss, HTTP status and remaining certificate lifetime. Only your own label appears in the series, so internal hostnames stay internal.

Events and notifications

Reboots, OOM kills and full disks are moments in time, not curves — no chart will show them.

Unexpected rebootRestarted without signing off - crash, power loss or a hard reset03:12Alarm

Disk fullFilesystem / is 100 percent used02:47Alarm

Out-of-memory killThe kernel killed a process because memory ran outyesterdayWarning

Clean shutdownPlanned restart - not counted as an outageAug 1Info

For every event type you decide how loud LIVCK gets: history only, information, warning, or a high-urgency alarm that runs through your escalation policy.

Don't take our word for it. Read the source.

The agent runs on your servers. That is why it is open source, only makes outbound connections and deliberately has no command channel.

Source on GitHub MIT licence

What the agent does not do

No remote controlThere is no command channel. LIVCK cannot run, start or change anything on your server.
No logsNeither journald nor /var/log nor any other log file is read.
No process dataIt counts processes but reads no names, command lines or environment variables.
No file contentsOnly counters from /proc and /sys are read.

The binary is built reproducibly, so you can rebuild it from source and compare it with the one we ship. Every release comes with a published checksum.

Works together

Monitoring, alerting, incidents and status pages build on each other — continue here.

MonitoringHTTP, TCP, DNS, ICMP & SSL checks from strategic locationsView Status PagesPublic status pages with custom domain and brandingView Incident ManagementStructured incident management with escalation policiesView AlertingMulti-channel notifications via email, SMS, Slack and webhookView

Your servers, seen from the inside

Twenty services and your first status page — in minutes, no payment details.

Start for free See pricing