Agent Sandbox

Hermes Agent in Kubernetes Sandbox Example

This project provides a complete, informational example of how to run Hermes Agent inside a Kubernetes cluster using the Sandbox CRD from agent-sandbox.

Prerequisites

1. A Kubernetes cluster (e.g., GKE, Minikube).
2. agent-sandbox installed in the cluster.
3. kubectl configured to access your cluster.
4. Python 3 (for running tests and chat scripts).

Files

• sandbox.yaml: The Kubernetes manifest for the Sandbox resource (core infrastructure).
• k8s-developer.md: The source file for the custom “Kubernetes Developer” skill.
• test_hermes.py: Automated Python script to verify the agent is running.
• chat_hermes.py: Interactive Python script to chat with the agent via CLI.

How to Use

1. Deploy the Sandbox

(Optional) Add Custom Skills

If you want to load the custom “Kubernetes Developer” skill, create a ConfigMap from the file before deploying the sandbox:

kubectl create configmap hermes-skills --from-file=k8s-developer.md

Apply the core manifest to your cluster:

kubectl apply -f sandbox.yaml

2. Verify the Deployment

Automated Verification

Run the provided test script. It will automatically find the pod, set up port-forwarding, and query the API:

python3 test_hermes.py

Manual Verification

Check if the Sandbox resource and Pod are created:

kubectl get sandboxes
kubectl get pods

You should see a pod named hermes-agent (matching the Sandbox metadata.name).

3. Interact with the Agent

Option A: Interactive CLI Chat

Use the provided chat script to talk to the agent.

1. First, find your pod name and start port-forwarding in a separate terminal:

   kubectl get pods
   kubectl port-forward pod/<your-pod-name> 8642:8642
   

2. Then run the chat script:

   python3 chat_hermes.py
   

Option B: Access the Dashboard

The dashboard runs on port 9119.

1. Start port-forwarding:

   kubectl port-forward pod/<your-pod-name> 9119:9119
   

2. Open http://localhost:9119 in your browser.

Configuration

To configure API keys or other environment variables, modify the sandbox.yaml file to include them in the env section or use a Kubernetes Secret.

Example using a Secret:

1. Create a secret with your API keys:

   kubectl create secret generic hermes-agent-api-keys \
     --from-literal=GEMINI_API_KEY="your_gemini_key" \
     --from-literal=OPENAI_API_KEY="your_openai_key"
   

2. The sandbox.yaml is already configured to use this secret:

   env:
   - name: GEMINI_API_KEY
     valueFrom:
       secretKeyRef:
         name: hermes-agent-api-keys
         key: GEMINI_API_KEY