31 self-hosted Datadog alternatives

Last reviewed Aug 26, 2026 · 474 words

Datadog's infrastructure monitoring listed at roughly $15 per host per month at last check, before logs, APM, and custom metrics turn the invoice into a line item your CFO asks about. For a homelab or a small fleet, the self-hosted answer costs a few hundred megabytes of RAM instead. The catch is that "replacing Datadog" means replacing between one and four products, and the right pick depends on which ones you actually use.

What you gain and what you give up

Beyond the bill, you gain data gravity: metrics and logs stay on your network, retention is your disk-space decision rather than a pricing tier, and there's no per-custom-metric metering shaping what you dare to measure. You gain composability too — every tool here speaks open standards, mostly Prometheus exposition and OpenTelemetry, so nothing locks you in.

You give up the single pane of glass. Datadog's real product is correlation — click from a metric spike to the traces and logs behind it — and rebuilding that from parts is genuine work. You give up their hundreds of maintained integrations, anomaly detection you didn't configure, and the luxury of your monitoring not being your problem at 3 a.m. Monitoring the monitor is now on you.

Which replacement fits which user

For a single server or small homelab, install Netdata (80,301 stars) and be done in five minutes: per-second metrics, auto-discovery, zero configuration, 256 MB of RAM. For infrastructure teams, Prometheus plus Grafana remains the standard kit — pull-based metrics, PromQL alerting, dashboards over everything — and the Grafana vs Prometheus comparison explains why they're partners, not rivals. If what you'd miss from Datadog is APM — distributed traces, service maps, unified logs — the closest one-product answer is SigNoz, OpenTelemetry-native but hungry: 4 GB of RAM and a ClickHouse dependency. Long retention across many hosts is where VictoriaMetrics earns its place as a drop-in Prometheus storage backend that compresses years of metrics into modest disk.

Migration reality

Your dashboards and monitors are the sunk cost: Datadog's JSON exports don't translate to Grafana, so plan to rebuild the 10 dashboards people look at and ignore the other 200. Historical metrics effectively can't leave. The clean cutover path is instrumenting with OpenTelemetry now, dual-shipping to both backends for a billing cycle, then turning off the agent. Prometheus keeps 15 days of data by default — raise the retention flag before anyone asks for last quarter's graph.

The default pick

Prometheus and Grafana, with Netdata alongside for the real-time per-host view. It's the boring stack with a decade of answers on the internet, and it scales from one mini PC to the point where you'd hire someone anyway.