Agent based monitoring installs a lightweight process on each host, which samples local counters and pushes them to a collection backend. It contrasts with agentless approaches such as SNMP polling, SSH scraping, or reading the cloud provider's own metrics API.
The trade is access against operational overhead. Only an agent can read /proc, enumerate mount points, and see per device block counters, and only an agent keeps recording when the network path to the monitoring vendor is the thing that broke, provided it buffers locally. The costs are real too: an agent is software you install, upgrade, and hold credentials for, it consumes memory on every host, and its own liveness becomes something you have to watch, because a silent agent looks exactly like a healthy quiet server. Judge one on three things: resident memory footprint, whether it queues to disk during an ingest outage, and how painful credential rotation is.
Hyperping's server agent installs with a single curl command, holds around 50 MB of resident memory with no JVM, and queues to /var/lib/hyperping/queue so metrics survive an ingest outage or a reboot. It runs on Linux with systemd and macOS with launchd, on amd64 and arm64, and re-running the installer rotates credentials while keeping the same server identity and history. Windows is not supported yet.
