Prometheus logo
Observability

Prometheus

Open-source metrics collection and alerting. Prometheus is an open-source monitoring system and time-series database, a graduated Cloud Native Computing Foundation project. It scrapes metrics from services and exporters, stores them, and lets you query them with PromQL. Alertmanager handles routing and grouping of alerts.

When it is the right choice

  • You run Kubernetes or many services and need metrics without per-host licence costs
  • You want alerts defined as code and reviewed like any other change
  • You want metrics from tools that already expose Prometheus format, such as vLLM and Kubernetes
  • Plain Prometheus is not built for long-term storage across many clusters; add a system such as Thanos or Mimir for that

How we use it

  • Deploy Prometheus and Alertmanager with Helm and manage rules in Git
  • Write alerts on symptoms users feel, each linked to a runbook
  • Collect GPU, model serving and token metrics alongside infrastructure metrics
  • Review noisy alerts regularly and remove or fix them
Works well with

Often used together

GrafanaKubernetesOpenTelemetryvLLM
Services

Where it fits in our work

AI observabilityKnow what is wrong, and why, before your customers tell youLearn more »After launchInfrastructure that keeps getting cheaper, faster and safer after launchLearn more »
From the blog

Related articles

AI · October 6, 2026Observability for LLM ApplicationsWhat to measure in an LLM application: tokens, latency, cost per request, OpenTelemetry traces and the quality signals that show when answers get worse.Devops · December 4, 2024DevOps TrendsDiscover the top DevOps trends for 2024, including GitOps, AIOps, DevSecOps, and multi-cloud strategies. Stay ahead with actionable insights for your teamDevops · November 27, 2024Devops MetricsDiscover 5 essential DevOps metrics and KPIs like Deployment Frequency and MTTR. Learn how to track KPIs for efficiency, reliability, and user trust
Questions

Prometheus: frequently asked

How long can Prometheus keep data?

It can keep weeks or months on local disk depending on volume. For longer retention or many clusters, we add remote storage such as Thanos or Grafana Mimir.

How do we avoid alert fatigue?

Alert on user-facing symptoms, not every metric. Each alert should need action and have a runbook; the rest belong on dashboards.

Running Prometheus, or deciding whether to? Tell us what you have; the first consultation is free.

Let's talk »