Workshop: Hands-on with the Agentic Layer

Requirements

• kubectl installed and configured
• GitHub account

Optional

• Local Kubernetes cluster (e.g., minikube, kind, Docker Desktop)
• Tilt for local development

Option 1: Use the remote cluster

Decrypt Kubeconfig

Use the included decryption script.

If using Windows, find another way to get the decoded file, e.g. via https://www.cryptool.org/en/cto/openssl/?tab=encryption Alternatively, use https://web.localsend.org/ and ask a friend.

1. Decrypt the kubeconfig (Kubernetes credentials file).

./decrypt-kubeconfig.sh kubeconfigs-encrypted/workshop-kubeconfig.yaml.enc <password-announced-during-workshop> kubeconfig.yaml

2. Connect to the cluster

   export KUBECONFIG=kubeconfig.yaml 
   
   kubectl get pods

Option 2: Use a local cluster (with Tilt)

Prerequisites

1. A running local Kubernetes cluster (e.g., minikube, kind, or Docker Desktop)
2. Tilt installed
3. A .env file in the project root with the following variables:

GOOGLE_API_KEY=<your-google-api-key>

Starting Tilt

tilt up

This will deploy the full stack locally, including:

• Agent Runtime — core runtime for the agentic layer
• AI Gateway — LLM proxy (LiteLLM-based)
• Agent Gateway — KrakenD-based API gateway
• Tool Gateway — MCP tool routing
• EC Schedule MCP Server — the MCP server for this project
• Presidio — PII guardrail
• LibreChat — chat UI for interacting with the agent
• Testbench — experiment runner
• LGTM Stack — observability (Grafana + Loki + Tempo + Mimir)

Port Forwards

Once Tilt is running, the following services are available locally:

Service	URL
LGTM (Grafana)	http://localhost:11000
AI Gateway	http://localhost:11001
Agent Gateway	http://localhost:11002
LibreChat	http://localhost:11003
Observability Dashboard	http://localhost:11004
Tool Gateway	http://localhost:11005
EC Schedule MCP Server	http://localhost:11020

Deploying Local Resources

Before you start

Replace <your-namespace> with your actual namespace in all files under deploy/local/.

Apply resources

kubectl apply -k deploy/local/

Apply a single resource

kubectl apply -f deploy/local/<filename>.yaml

Remove all resources

kubectl delete -k deploy/local/