Cluster Assistant for Kubernetes

Asking a prompt in the prompt box to Gemini
A logs output for a particular pod
Result of a prompt asking to scale a deployment
List of all pods and their status in a cluster
Architecture Diagram

Inspiration

This project is clearly inspired by kubectl-ai application which is created by the Google Cloud Platform team. It clearly joines a very tried, tested open source software like Kubernetes and using it with bleeding-edge, industry leading software like GenAI with Gemini.

It was my goal to combine these two approaches.

What it does

It is a cluster assistant. Imagine if there was a person who knew the cluster in-and-out including what logs are happening in a pod, what status a pod is at and more. On top of all this, this assistant can be ordered to do things in the cluster, such as scale deployments to a certain number of replicas.

How I built it

This was built using multiple tools and frameworks.

Firstly Gemini 2.5 Flash was used as the model creating commands out of natural language, by converting a prompt to a JSON format.

Next I used the Kubernetes Python Client to act on something based on the prompt. Such as scale deployment, fetch logs etc. ALL FROM JUST A PROMPT!

Finally all the outputs and displays was created using streamlit.

Challenges I ran into

Time constraint. Given that I was short on time, I had to speed-create and code a decent application and idea. On top of this I also was the only programmer in the hackathon which meant I had to code the frontend too.

Accomplishments that I'm proud of

Created something that combines GenAI and Kubernetes
Learned the Kubernetes API and the Pythone (and Go) client libraries for it.
Successfully created a working streamlit app that does this.

What I learned

How to get a proper output from AI models (such as JSON in my case) with perfect prompt engineering.
How to use GKE and cloud services to my advantage.
The k8s API and all the client libraries for it.