The Core Workflow #
KubeAura operates on a "zero-deployment" philosophy. It leverages your existing kubeconfig without requiring a database or server setup.
kubeaura # Reads current context and launches at http://127.0.0.1:7654
AI Model Flexibility: Local vs. Hosted #
This is where the tool gets interesting for those of us benchmarking LLM performance. KubeAura doesn't lock you into one provider; it's pluggable.
Local Deployment (Ollama): Best for privacy. You can run Llama 3.2, Mistral, or Qwen locally. No cluster data leaves your machine.Hosted Intelligence ( Best for complex root-cause analysis where deeper reasoning is needed.Claude/OpenAI):Compatibility: Works with any OpenAI-compatible endpoint, including vLLM and Groq.
Technical Breakdown #
The tool focuses on "detect-don't-install" integrations, meaning it automatically picks up existing cluster data rather than forcing you to install agents.
Observability: Includes a "Pulse" triage view for OOMKills and crashloops, plus a topology graph mapping Ingress → Service → Pod.Resource Tracking: Real-time CPU/Memory usage displayed alongside requests and limits for 19 different resource kinds.Voice Interface: Supports natural language queries (e.g., "Why is this API failing?") with spoken responses.Ecosystem Support: Native detection for Argo CD, Flux, Trivy, and cert-manager.
Deployment Guide #
If you want to test this AI workflow, the setup is nearly instant.
macOS / Linux:
curl -sSfL https://raw.githubusercontent.com/devganeshg/kubeaura/main/scripts/install.sh | sh
kubeaura
Docker:
docker run --rm -p 7654:7654 \
-v ~/.kube:/home/nonroot/.kube:ro \
ghcr.io/devganeshg/kubeaura
Connecting a Local LLM via Ollama:
ollama serve &
ollama pull llama3.2
export KUBEAURA_AI_PROVIDER=ollama
kubeaura
For those preferring Claude, just export your ANTHROPIC_API_KEY
before launching. It effectively turns your dashboard into an LLM agent with full visibility into your cluster's current state.
Next LLM Architecture: Lessons from Karpathy →