Retention you decide
Ninety days by default, set to whatever you need in hours, days, weeks or years. Because ingest isn't metered, keeping a year of metrics is a storage question rather than a billing one.
Managed Prometheus
One line of remote_write, and you're done.
A Prometheus-compatible endpoint that takes your metrics, keeps them as long as you say, and answers PromQL, so your dashboards, rules and alerts carry over unchanged.
Getting data in
Point your existing setup at it.
There's no agent of ours to deploy and no format to convert to. If it speaks Prometheus remote write, and nearly everything does, it can write here. Local Prometheus servers can keep scraping exactly as they do now and forward the result.
- Prometheus remote_write from an existing server
- Prometheus Agent scrape and forward, no local storage
- OpenTelemetry Collector via the Prometheus remote write exporter
- Grafana Alloy or any agent speaking remote write
remote_write:
- url: https://your-cluster.clusternest.com/api/v1/write
basic_auth:
username: admin
password_file: /etc/prometheus/clusternest.pwWhat you get
Long retention, without the invoice surprise.
Two credentials, two jobs
A read-write admin user for your shippers and a read-only user for everything that only queries. Hand the read-only one to Grafana and dashboards can never write to your metrics.
PromQL, unchanged
The query API is Prometheus-compatible, so your existing rules, dashboards and alerts keep working. Nothing to port, no dialect to learn.
Custom hostname and allowlist
Serve the endpoint on your own domain, and restrict it to the networks your workloads actually come from.
Scale out, not up
Two storage nodes to start, up to ten on Extreme, added without taking the endpoint down.
Rotate on demand
Rotate either credential from the console, the API or Terraform, whenever you need to. No ticket, no coordination with us.
Reading it back
Your dashboards don't move.
Add the endpoint as a Prometheus data source in Grafana with the read-only credential and everything you've already built keeps working. Recording and alerting rules are PromQL, so they transfer as they are.
Scale
| Tier | Per node | Storage nodes |
|---|---|---|
| Basic | 1 vCPU · 2 GiB · 10 GB | 2 |
| Standard | 4 vCPU · 32 GiB · 500 GB | 2 |
| Extreme | 8 vCPU · 64 GiB · 2 TB | 2–10 |
Priced per storage node, per month, exactly as OpenSearch is. Pricing
Versions
The last 10 releases, always available.
The storage engine is VictoriaMetrics, which is why these version numbers don't match Prometheus's own. The API you talk to is Prometheus-compatible either way. New clusters default to 1.148.0.
- 1.140.0
- 1.141.0
- 1.142.0
- 1.143.0
- 1.144.0
- 1.145.0
- 1.146.0
- 1.147.0
- 1.148.0
- 1.149.0
Start with one cluster.
Ten days on a Basic cluster, no card. Point a test Prometheus at it and see what the queries feel like.