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

K8sGPT

SandboxObservability
Open source · CNCF · CNCF Sandbox · Apache-2.0

K8sGPT scans a Kubernetes cluster for problems — crashlooping pods, failed deployments, misconfigured services, pending PVCs, RBAC issues — and runs the raw error output through an LLM (OpenAI, Azure OpenAI, local models via Ollama, Amazon

updated 2026-07-01

What it is

K8sGPT scans a Kubernetes cluster for problems — crashlooping pods, failed deployments, misconfigured services, pending PVCs, RBAC issues — and runs the raw error output through an LLM (OpenAI, Azure OpenAI, local models via Ollama, Amazon Bedrock, etc.) to produce a plain-English explanation and suggested fix. It ships as a CLI, a Kubernetes operator, and an in-cluster analyzer that can run continuously and push findings to Slack, Discord, or other integrations. Analyzers are pluggable — there's a defined set covering core resources (Pod, Service, Ingress, PVC, HPA, etc.) plus community-contributed ones, and you can write custom analyzers for your own CRDs.

It does not replace kubectl describe or existing observability stacks. It sits on top of them, aggregating the same diagnostic signals you'd normally pull manually and formatting them into something a junior engineer (or a senior one at 3am) can act on faster.

Who builds it and why

K8sGPT is a CNCF Sandbox project. It originated as an open-source effort to reduce the time engineers spend translating cryptic Kubernetes error states into root causes — the kind of triage work that's mechanical but requires tribal knowledge. 122 contributors and active maintenance (commits as recent as mid-2026) suggest a real community, not a single-vendor side project, though the founding organization/date is not publicly documented. The project is stewarded under CNCF governance, which means roadmap and maintainership are at least nominally open rather than locked to one company's commercial interest.

Production readiness signal

  • CNCF stage: Sandbox — the earliest CNCF maturity level. This signals active development and community interest, not proven production hardening at scale.
  • Release cadence: v0.4.35 indicates a project still in pre-1.0 iteration; expect breaking changes between minor versions.
  • Community health: 7,935 GitHub stars and 122 contributors is a healthy signal for a sandbox project — active, not abandoned.
  • License: Apache-2.0, no vendor lock-in risk on the code itself.
  • Gap: No published case studies, adoption data, or enterprise SLAs are publicly available. Treat it as a diagnostic aid, not a system of record.

Who should use this

  • Platform teams who want a fast, low-effort way to surface "what's actually wrong" summaries for common cluster failures without building custom tooling.
  • SRE/on-call engineers who want AI-assisted triage to shorten the loop between "alert fires" and "root cause hypothesis," especially for less-experienced responders.
  • Teams already comfortable sending cluster diagnostic data to an external LLM API (or running a local model) who want to reduce MTTR on routine, well-known failure patterns.
  • Platform engineers building internal developer platforms who want to surface human-readable error explanations to app teams instead of raw K8s error strings.

Who should NOT use this

  • Regulated environments where sending pod logs, resource specs, or error text to a third-party LLM API is a compliance non-starter — unless you're running a fully local model and have validated the data path.
  • Teams expecting root-cause analysis for complex, multi-service, or novel failure modes — K8sGPT is built for known, common misconfigurations, not distributed-systems-level incident investigation.
  • Organizations wanting a mature, CNCF-graduated dependency for critical-path tooling — sandbox status means API and behavior are still shifting.
  • Anyone expecting this to replace observability (Prometheus/Grafana, tracing, log aggregation) — it's a triage assistant, not a monitoring stack.

Alternatives

  • kubectl-ai / kagent — similar LLM-assisted Kubernetes troubleshooting concepts, varying maturity and scope.
  • Botkube — rules- and integration-based cluster event notifier; less AI-driven explanation, more alert routing.
  • Robusta — automated runbooks and remediation for Kubernetes alerts, overlapping use case with more automation focus than explanation focus.

Pricing

Fully open source (Apache-2.0). No paid tier, no commercial edition publicly known. Cost exposure comes from whichever LLM backend you connect it to — API usage fees if using OpenAI/Bedrock/Azure, or compute cost if self-hosting a local model.

Frequently asked

What is K8sGPT?+
K8sGPT scans a Kubernetes cluster for problems — crashlooping pods, failed deployments, misconfigured services, pending PVCs, RBAC issues — and runs the raw error output through an LLM (OpenAI, Azure OpenAI, local models via Ollama, Amazon Bedrock, etc.) to produce a plain-English explanation and suggested fix.
Who builds K8sGPT?+
K8sGPT is a CNCF Sandbox project. It originated as an open-source effort to reduce the time engineers spend translating cryptic Kubernetes error states into root causes — the kind of triage work that's mechanical but requires tribal knowledge.
Is K8sGPT production ready?+
CNCF stage: Sandbox — the earliest CNCF maturity level. This signals active development and community interest, not proven production hardening at scale. Release cadence: v0.4.35 indicates a project still in pre-1.0 iteration; expect breaking changes between minor versions.
Who should use K8sGPT?+
Platform teams who want a fast, low-effort way to surface "what's actually wrong" summaries for common cluster failures without building custom tooling. SRE/on-call engineers who want AI-assisted triage to shorten the loop between "alert fires" and "root cause hypothesis," especially for less-experienced responders.
Who should not use K8sGPT?+
Regulated environments where sending pod logs, resource specs, or error text to a third-party LLM API is a compliance non-starter — unless you're running a fully local model and have validated the data path.
What are the alternatives to K8sGPT?+
kubectl-ai / kagent — similar LLM-assisted Kubernetes troubleshooting concepts, varying maturity and scope. Botkube — rules- and integration-based cluster event notifier; less AI-driven explanation, more alert routing.
How much does K8sGPT cost?+
Fully open source (Apache-2.0). No paid tier, no commercial edition publicly known. Cost exposure comes from whichever LLM backend you connect it to — API usage fees if using OpenAI/Bedrock/Azure, or compute cost if self-hosting a local model.