No single tool here replaces Datadog, and our page implies one does

Our page for Datadog lists twenty-two open source tools. It orders them by whether we featured a project and then by the health score we compute for it, and it paints the first three with gold, silver and bronze. A reader arriving from a search for how to leave Datadog gets a podium.

Grafana takes the gold. Grafana is excellent and the score behind that placement is not wrong, but Grafana stores nothing at all. It reads from Prometheus, Loki, Elasticsearch, InfluxDB and Postgres and draws what it finds, so a team that adopts our top ranked answer to Datadog and stops there has installed a window with no house behind it.

Netdata takes silver and is a per-node agent that shows you one machine in real time. Prometheus takes bronze and is a metrics store with no interface worth showing a manager. Three different products, three medals, one column.

This post is the thing the page should have said. What Datadog actually sells, where the twenty-two really sit, and which rows we would not hand anyone without a warning.

The podium ranks project health, not fit

The ordering is one line of code. Our alternatives pages sort by isFeatured and then by score, a number we derive from repository signals, so the list is ordered by how healthy a project looks to us and never by which part of your bill it cancels.

That produces a ranking that is internally consistent and practically useless. A team paying Datadog mostly for log ingestion and a team paying mostly for APM traces are shopping for different products, and both get the same podium in the same order.

The honest version of this page is not a ranking at all. It is a map.

Datadog is several products sold as one line item

The reason no single row replaces it is that the thing being replaced is not one thing. Datadog bills separately for infrastructure metrics, log ingestion and retention, application performance traces, error tracking, synthetic and uptime checks, and real user monitoring. Teams rarely buy all of it and almost never buy it evenly.

This matters before you shop, because the largest line on your invoice decides which tool you are actually looking for, and the twenty-two do not cover those lines equally.

Where the twenty-two actually split

Sorted by what they do rather than by score, the list falls into groups. Every project below is in our catalog with a commit inside the last two months.

• Metrics stores: Prometheus, VictoriaMetrics and GreptimeDB. These hold time series and query them. They are not dashboards and they do not ingest your logs.
• The visualization layer: Grafana. It reads from Prometheus, Loki, Elasticsearch, InfluxDB and Postgres, and on its own it stores nothing at all.
• OpenTelemetry platforms: SigNoz, OpenObserve, HyperDX, Coroot, Uptrace, Maple and Gigapipe. These are the closest structural match, since they take logs, metrics and traces in one system the way Datadog does.
• Error tracking and session replay: Sentry and Highlight. This is the part of Datadog many teams are actually paying for and it is a different product from metrics.
• Classic infrastructure monitoring: Zabbix, Checkmk, Icinga, Nagios and Netdata. Agent and check based, strong on hosts and networks, weak on distributed traces.
• Log search: Quickwit. Built for observability search rather than general dashboards.
• Uptime and status: OneUptime.

Prometheus and Grafana appear on almost every list of Datadog alternatives and are routinely presented as one answer. They are two halves of one answer, and our page ranks them first and third as though a reader should choose between them. Netdata sits between them and is neither half, since it watches hosts rather than storing a fleet's time series.

SigNoz is the row whose shape most resembles the product you are leaving, because it takes all three signals in one place. That makes it the easiest comparison to reason about and the hardest to stage, since adopting it is one decision rather than four.

Three rows we would not hand over without a warning

The list is honest about project health and silent about everything else, so here is the everything else.

• Sentry has no license recorded in our database. It is the fourth most starred row on this page at 44,738 stars, and the field that would tell you what you are agreeing to is blank. We wrote about why that field is blank for 89 of our tools in our previous post.
• Logfire is scoped to LLM and agent systems, not to general application performance. It sits on this page because it is observability, and it will not replace an APM bill.
• Temps claims to replace six products at once, including Vercel, Sentry, PostHog and Pingdom. It is also the one tool in the entire catalog that breaks our own cap of five alternatives, which is a sign we have not curated it carefully rather than a sign it does six jobs well.

What to do with the list instead

Read your invoice first and the page second. The useful sequence is short.

• Find your largest billed signal before you open any directory. Metrics, logs, traces and error tracking lead to four different groups above, and the podium ordering will not tell you which one you are in.
• Decide whether you want one system or an assembled one. An OpenTelemetry platform is one deployment and one failure domain, while Prometheus plus Grafana plus a log store is three of each and more control.
• Check the enterprise edition before you commit. Retention, downsampling and single sign on are the features most often reserved for a paid tier, which is the same bill in a smaller envelope.
• Ask what you are willing to operate at three in the morning. Several of these run on ClickHouse, which is excellent and is also a database your team now owns.

What this page should become

We are not going to pretend this post fixes the page. The ordering is still isFeatured then score, and the next reader who lands on Datadog from a search will still see a podium.

What should change is the shape. An alternatives page for a product billed in parts should group its rows by the part they replace, and it should say plainly when a row covers one signal rather than all of them. That is a change to how we model the relationship between a product and a tool, not a copy edit, and it is the more useful thing to build next.

Until then, the list is a good inventory and a bad recommendation. Twenty-two healthy projects sit on that page, every one of them worth running, and not one of them is Datadog.