Per-second metrics kept on each machine
Netdata is a monitoring agent you install on every machine, whether it runs Linux, FreeBSD, macOS or Windows. The agent works out what is running, samples its metrics every second, keeps them on local disk and serves a dashboard on port 19999. Other agents can stream their data to a "parent" agent that holds the history for all of them. Netdata Cloud, the vendor's SaaS, queries agents and parents directly and has no central metric store. The README's FAQ sets Netdata against Prometheus plus Grafana: no manual setup and no dashboards to build.
Your data therefore stays on your agents and parents, whereas SigNoz writes logs, metrics and traces to ClickHouse and OpenObserve stores them as Parquet files on object storage. Netdata started out with host metrics: it has accepted OpenTelemetry logs since v2.11.0 (August 2026) and traces since v2.12.0, as an early-access feature.
The comparison of open source observability platforms looks at what each tool stores, on which database, and at what volume it starts to pay off.
Finds running services and flags anomalies on the node
- One-second resolution out of the box: the agent keeps that grain for 14 days (1 GiB at most), then rolls up to per-minute data for 3 months and per-hour data for 2 years. The dashboard picks the tier that matches the zoom level, and the README puts storage at about 0.5 bytes per sample.
- Services found on their own: Docker, containerd, LXC and Kubernetes containers, VMs seen from the host, systemd services and processes. The README lists 800+ integrations, and you slice the charts without writing a query.
- Anomaly detection on the node: the agent trains several unsupervised models per metric and ships hundreds of preset alerts, which can notify by email, Slack, Telegram, PagerDuty, Discord or Teams.
- Hardware down to the sensor: on Linux, GPUs, PCI AER and EDAC memory errors, IPMI, S.M.A.R.T. and NVMe. The Redfish collector added in v2.12.0 reads the BMC (iDRAC, iLO) and keeps reporting with the host powered off, as long as the BMC has power.
Closed-source UI, and logs behind a Cloud login
Only the agent is GPL v3 or later. The web UI ships in the packages and updates over a CDN, but it is closed source: its NCUL1 license forbids modifying it or using it with anything other than Netdata agents and Netdata Cloud. Netdata Cloud is closed source too.
Logs need a Netdata Cloud login. Netdata's access-control docs list the systemd journal, Windows events, the process list and network connections as blocked for anyone who isn't signed in. A free account is enough, and traces need the same sign-in. Metrics, alerts and anomaly detection work without an account.
Past five nodes, you pay. Without a subscription, dashboard access stops at five nodes; on Netdata Cloud the rest show as locked, though they keep collecting. Configuring collectors and alerts from the UI, RBAC and SSO are paid features as well. The pricing pages list Business at $4.50 per node per month billed annually, Homelab at $90 a year for non-commercial use, and Enterprise On-Premises from 200 nodes.
Port 19999 is open by default. The security docs say anyone who can reach it can read the dashboard and metrics. The built-in token protection runs through Netdata Cloud; without it, you are left with the agent's IP access lists or a reverse proxy that handles authentication. On Windows, a standalone agent on the free tier keeps its local dashboard locked until it is connected to Netdata Cloud.
The installer defaults to nightly builds. kickstart.sh installs the nightly channel and turns on daily automatic updates; --stable-channel picks stable releases instead and --no-updates skips the updates. Anonymous telemetry is on by default; --disable-telemetry turns it off.
About 150 MiB of RAM per agent. That is the README's figure for a default production setup, alongside roughly 5% CPU, and it drops to about 100 MiB with ML and alerts off and ephemeral storage.
Five minor releases in under a year. v2.8.0 came out on November 19, 2025 and v2.12.0 on September 30, 2026, with patch releases in between. The v2.12.0 notes open with a warning about random crashes on startup and recommend v2.11.1 until v2.12.1 ships.