● Beta — KubeEra is in active development. Profiles are live; AI-perception measurement is rolling out.
KubeEra
C

Cortex

IncubatingObservability
Open source · CNCF · CNCF Incubating · Apache-2.0

Cortex is a horizontally scalable, multi-tenant, long-term storage system for Prometheus metrics.

updated 2026-07-01

What it is

Cortex is a horizontally scalable, multi-tenant, long-term storage system for Prometheus metrics. It sits behind a Prometheus-remote-write endpoint, ingests samples at scale, and stores them in object storage (S3, GCS, Azure Blob) instead of local disk. It exposes a PromQL-compatible query API so existing dashboards and alerting rules work unchanged. Core selling points: multi-tenancy out of the box, independent scaling of ingestion vs. query paths, and durability that doesn't depend on a single Prometheus instance's disk.

Who builds it and why

Cortex originated at Weaveworks to solve the problem of running Prometheus at scale for many customers on shared infrastructure. It became a CNCF project and reached incubating maturity. The contributor base (345+) spans multiple companies, notably AWS — Amazon Managed Service for Prometheus is built on Cortex, which is a major reason the project has continued to get serious engineering investment even after Grafana Labs forked much of the same lineage into Mimir. The "why" hasn't changed: teams running Prometheus at fleet scale need object-storage-backed retention and tenant isolation that vanilla Prometheus doesn't provide.

Production readiness signal

CNCF incubating status plus 345 contributors and an active commit history (last commit June 2026) indicate the project isn't stagnant, but it's worth noting the metrics ecosystem has consolidated around Grafana Mimir and Thanos in mindshare over the past few years — Cortex's public visibility (5,819 GitHub stars) is comparatively modest next to those alternatives. That doesn't mean it's unsound: it's running in production at AWS-scale (AMP) and has a long track record of operating multi-tenant metrics at petabyte scale. v1.21.1 suggests a mature, incrementally versioned release cadence rather than a project still finding its architecture. Apache-2.0 licensing is standard and unencumbered.

Who should use this

  • Teams already running Prometheus who need durable, object-storage-backed long-term retention beyond local disk limits.
  • Platform teams building a multi-tenant metrics-as-a-service offering internally (e.g., one Cortex cluster serving many product teams with hard tenant isolation).
  • Organizations that want to build on the same codebase AWS uses for Amazon Managed Service for Prometheus, or that are already running AMP and want self-hosted parity/portability.
  • Teams with existing Cortex deployments who have no urgent reason to migrate — operational familiarity and stability matter more than chasing the newest fork.

Who should NOT use this

  • Small teams with a handful of Prometheus instances and modest retention needs — this is operational overhead you don't need; Prometheus's built-in storage or a managed SaaS product is simpler.
  • Anyone starting fresh today without an existing AMP or Cortex investment — Grafana Mimir forked from Cortex specifically to simplify operations and improve scalability, and it has more first-party tooling, documentation, and community mindshare right now. Starting a greenfield deployment on Cortex over Mimir requires a specific justification (usually AWS AMP compatibility).
  • Teams wanting logs or traces in the same system — Cortex is metrics-only; you'll still need Loki/Tempo or equivalents alongside it.
  • Anyone without dedicated SRE/platform capacity — running the ingester/distributor/querier/compactor topology correctly, including ring consistency and blocks storage compaction, is nontrivial operational work.

Alternatives

  • Grafana Mimir — direct fork of Cortex, now the more actively promoted option with tighter Grafana ecosystem integration and generally simpler operations.
  • Thanos — similar goal (long-term, scalable Prometheus storage) but a different architecture built around sidecars and a global query layer; strong alternative if you want less multi-tenancy overhead.
  • VictoriaMetrics (cluster edition) — Prometheus-compatible, generally lower resource footprint, popular where operational simplicity and cost are the priority over CNCF pedigree.

Pricing

Fully open source, Apache-2.0. No licensing cost. All spend is your own infrastructure (object storage, compute for ingesters/queriers) and engineering time to operate it — there is no commercial entity selling a Cortex-specific managed offering the way Grafana Labs sells managed Mimir/Grafana Cloud. If you want a managed equivalent, Amazon Managed Service for Prometheus (Cortex-based) is the closest thing, priced by AWS separately.

Frequently asked

What is Cortex?+
Cortex is a horizontally scalable, multi-tenant, long-term storage system for Prometheus metrics. It sits behind a Prometheus-remote-write endpoint, ingests samples at scale, and stores them in object storage (S3, GCS, Azure Blob) instead of local disk.
Who builds Cortex?+
Cortex originated at Weaveworks to solve the problem of running Prometheus at scale for many customers on shared infrastructure. It became a CNCF project and reached incubating maturity.
Is Cortex production ready?+
CNCF incubating status plus 345 contributors and an active commit history (last commit June 2026) indicate the project isn't stagnant, but it's worth noting the metrics ecosystem has consolidated around Grafana Mimir and Thanos in mindshare over the past few years — Cortex's public visibility (5,819 GitHub stars) is co
Who should use Cortex?+
Teams already running Prometheus who need durable, object-storage-backed long-term retention beyond local disk limits. Platform teams building a multi-tenant metrics-as-a-service offering internally (e.g., one Cortex cluster serving many product teams with hard tenant isolation).
Who should not use Cortex?+
Small teams with a handful of Prometheus instances and modest retention needs — this is operational overhead you don't need; Prometheus's built-in storage or a managed SaaS product is simpler.
What are the alternatives to Cortex?+
Grafana Mimir — direct fork of Cortex, now the more actively promoted option with tighter Grafana ecosystem integration and generally simpler operations.
How much does Cortex cost?+
Fully open source, Apache-2.0. No licensing cost. All spend is your own infrastructure (object storage, compute for ingesters/queriers) and engineering time to operate it — there is no commercial entity selling a Cortex-specific managed offering the way Grafana Labs sells managed Mimir/Grafana Cloud.