k8sgpt

 Run Kubernetes AI Debugging Locally Using k8sgpt + Ollama (No OpenAI, 100% Free)

As Kubernetes clusters grow, debugging issues like ImagePullBackOff, CrashLoopBackOff, or scheduling failures becomes time-consuming.
k8sgpt solves this by analyzing your cluster and explaining issues in plain English using AI.

In this blog, I’ll show how to run k8sgpt locally with Ollama (no OpenAI key required) using Minikube on Windows.

This setup is ideal for:

• Kubernetes SREs

• DevOps Engineers

• Platform teams

• Anyone who wants AI-assisted debugging without cloud dependency

---

🧱 Architecture

Minikube (Kubernetes)
   |
k8sgpt (CLI)
   |
Ollama (Local LLM - llama3.1)

✔ Fully local
✔ No API key
✔ No cost
✔ Works offline

---

✅ Prerequisites

• Windows 10/11 (64-bit)

• Minikube installed and running

• kubectl configured

• Ollama installed

Verify:

kubectl get nodes
ollama list

Expected:

minikube   Ready
llama3.1

---

🔹 Step 1: Download k8sgpt (Windows)

Go to:
👉 https://github.com/k8sgpt-ai/k8sgpt/releases

Download:

k8sgpt_Windows_x86_64.zip

Extract it and move:

k8sgpt.exe → C:\Program Files\k8sgpt\

Add this directory to your PATH.

Verify:

k8sgpt version

---

🔹 Step 2: Verify Kubernetes Context

kubectl config current-context

Output:

minikube

---

🔹 Step 3: Remove OpenAI Backend (Important)

If OpenAI was previously configured:

k8sgpt auth remove --backends openai

This avoids quota and authentication errors.

---

🔹 Step 4: Configure Ollama as AI Provider

Add Ollama with explicit model name:

k8sgpt auth add --backend ollama --model llama3.1

Set Ollama as default provider:

k8sgpt auth default --provider ollama

Verify:

k8sgpt auth list

Expected:

Default:
> ollama
Active:
> ollama

---

🔹 Step 5: Verify Ollama Endpoint

curl http://localhost:11434/api/tags

You should see:

llama3.1

---

🔹 Step 6: Run k8sgpt Analysis

Basic analysis:

k8sgpt analyze

With AI explanation:

k8sgpt analyze --explain

For cleaner output:

k8sgpt analyze --explain --filter=Pod,Node,Deployment

---

🧪 Step 7: Test with a Real Failure

Create a broken pod:

apiVersion: v1
kind: Pod
metadata:
  name: broken-pod
spec:
  containers:
  - name: test
    image: nginx:doesnotexist

Apply:

kubectl apply -f broken.yaml

Now run:

k8sgpt analyze --explain --filter=Pod

✅ Output (Example)

• Detects ImagePullBackOff

• Explains root cause

• Suggests fix

• Generated locally using llama3.1

---

🧠 Why This Setup Is Powerful

Feature	Benefit
Local LLM	No internet required
No OpenAI	Zero cost
Minikube	Safe learning environment
k8sgpt	Fast RCA
Ollama	Production-grade local AI

---

🔐 Production Notes

• This setup works the same on large clusters (50+ nodes)

• In production, you can:

  • Run k8sgpt as CronJob

  • Integrate with Slack / MCP / ChatOps

  • Use with EFK / OpenSearch logs

  • Extend to Robin.io environments

---

🏁 Conclusion

By combining k8sgpt + Ollama, you get an AI-powered Kubernetes debugging assistant that:

• Runs locally

• Costs nothing

• Protects data privacy

• Scales from Minikube → Production

This is an excellent way for SREs to adopt AI safely.

---