kagent

Before we install Microk8s, make sure you have kubernetes ready environment, since i have MacOS, i am using Microk8s here, similarly you can choose to install Depending on OS you are using here - Install Microk8s 

and while running the command , you might face error similar to this

brew install ubuntu/microk8s/microk8s

This is expected as sierra is deprecated and there was no replacement, here is supporting github issue

- Github issue

let us use mutlipass as suggested

Using multipass, you can run ubuntu VM in MAC or Windows and it developed by canoical

and now lets install mutlipass and run microk8s like we run in ubuntu

keep settings as default and install

Alternately, you can also use AWS EC2 like we did it in our previous project - microk8s previous project

Now, lets use the commands provided in Doc to launch our instance

since we are going to install microk8s, kagent and kmcp in same VM, lets increase resources for VM

command

multipass launch --name microk8s --cpus 4 --memory 8G --disk 80G 24.04

once launched enter inside VM

command: multipass shell microk8s

and now let us setup our kubernetes environment, the way we did for previous project and follow steps from here - microk8s deployment

command: sudo snap install microk8s --classic

command: sudo usermod -a -G microk8s ubuntu

mkdir ~/.kube

sudo microk8s kubectl config view --raw > ~/.kube/config

switch users for configuration change to take in effect

command:

sudo su - root

sudo su - ubuntu

command: 

microk8s kubectl get all --all-namespaces

command: alias kubectl="microk8s kubectl"

Now, let us do a basic nginx image deployment

kubectl create ns nginx-demo

command: kubectl create deployment nginx --image=nginx:latest -n nginx-demo

check pod status

command: kubectl get pods -n nginx-demo

Now let us also create service to expose our application

command: kubectl expose deployment nginx --port=80 --target-port=80 --type=NodePort -n nginx-demo

check microk8s service

command: kubectl get svc -n nginx-demo

since service is exposed via nodeport inside VM, we need to get IP of VM

command: hostname -I

Try accessing application now at nodeport inside you local

http://VMIP:NodePort

Before we install kagent, lets get our openai apikey and you can see step by step process of how to get OpenAI API Key from this Blog - Fine Tuning 

and Generate API Key

since our kubernetes setup is done, now lets go for installing kagent and see how it works

we can use official kagents documentation

paste the OpenAI API Key that generated

also install kubectl as kagent interacts mostly with kubectl

command: sudo snap install kubectl --classic

also install helm 

command

curl https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

command: 

curl https://raw.githubusercontent.com/kagent-dev/kagent/refs/heads/main/scripts/get-kagent | bash

let us check demo version which consists of few agents and mcp tools

since kagent profile takes lot of time to setup CRDs , we will use helm to install crds and kagent for faster setup 

kagent install --profile demo

Alternate install

helm install kagent-crds oci://ghcr.io/kagent-dev/kagent/helm/kagent-crds \

  --namespace kagent \

  --version 0.7.7 \

  --create-namespace

Now let us install kagent

helm install kagent oci://ghcr.io/kagent-dev/kagent/helm/kagent \

--namespace kagent \

--version 0.7.7 \

--set providers.default=openAI \

--set providers.openAI.apiKey=$OPENAI_API_KEY

Since we are running inside VM, let us do a port-forward for our kagent dashboard

kubectl port-forward --address 0.0.0.0 -n kagent service/kagent-ui 8082:8080

Now access kagent dashboard with http://VMIP:8082

Now clikc on skip wizard to enter into dashboard

and you can see few agents already configured

and now click on any agent 

and on the right, you can see tools like this agent can use

and for each question, it uses tools that is required for that use case, like when i asked how many pods are there in nginx-demo namespace, it has used k8s_get_resources tool

and when i asked it to increase pods to two resources, it did 

let us create an agent , that is specifically only for scaling deployments

and since it has already tools added, we will use same tool that is being used and Agent instructions must be clear

You are a Kubernetes deployment scaling agent.

You are only allowed to scale deployments up or down by changing replica counts.

Allowed actions:

- Get deployments

- List deployments

- Patch deployments/scale

- Update deployments/scale

Forbidden actions:

- Do not delete pods

- Do not delete deployments

- Do not create new workloads

- Do not read secrets

- Do not modify ConfigMaps

- Do not change images

- Do not restart deployments unless explicitly asked, and prefer scaling only

Before scaling, confirm:

1. Namespace

2. Deployment name

3. Current replicas

4. Requested replicas

After scaling, show:

1. Old replica count

2. New replica count

3. kubectl verification command