k8stalk: An Agentic LLM CLI That Finally Makes Kubernetes Troubleshooting Smart 🚀

Every Kubernetes user knows the pain: a pod is failing, and you're stuck in the endless manual loop

kubectl describe, logs, events, check ArgoCD/Flux, repeat.

I built k8stalk to fix that.

What is k8stalk?
GitOps-aware, conversational Kubernetes diagnostics agent.
It scans your clusters, diagnoses issues using agentic multi-step reasoning, and correlates failures with ArgoCD/Flux GitOps activity - all in plain English.
Backed by multi-model support: Ollama (fully local), OpenAI, Anthropic, Azure OpenAI, Google Gemini, Vertex AI, AWS Bedrock, and any OpenAI-compatible endpoint.

How it works
k8stalk combines deterministic static analysis (scanning 16+ resource types) with an agentic LLM loop. The model decides what to inspect next, gathers real Kubernetes data via API calls, reasons over findings, and delivers clear root cause + fixes. No hallucinations.

Key Features:
Native GitOps correlation (ArgoCD & Flux sync awareness)
Multi-turn conversational diagnosis
CLI + beautiful local browser-based UI (k8stalk chat)
Fast mode for smaller/local models
Fully local/offline with Ollama + complete data privacy

Real Example
A pod stuck in CrashLoopBackOff with noisy FailedToRetrieveImagePullSecret warnings.
k8stalk ignored the red herring, spotted OOMKilled (exit code 137) directly from container status, and recommended increasing memory limits.

Try It Yourself

`brew install naman833/k8stalk/k8stalk
k8stalk init     # interactive setup
k8stalk analyze  # quick diagnosis
k8stalk chat     # launch the nice local web UI`

GitHub: https://github.com/naman833/k8stalk
I’d love honest feedback from Kubernetes + GitOps users in production!

What’s the most frustrating part of your debugging workflow? Drop your thoughts below 👇``