Agents

Agents are the core execution units in Agents at Scale. They receive inputs, process them using AI models, and return outputs. Agents can be configured with specific prompts, models, and tools to perform specialized tasks.

An agent wraps an AI model with a specific prompt and configuration, creating a reusable component that can be queried by other agents or directly through queries. Agents can reference tools and models to extend their capabilities.

Specification


apiVersion: ark.mckinsey.com/v1alpha1
kind: Agent
metadata:
  name: sample-agent
spec:
  prompt: You're helpful agent. Always add amazing to the output.
  # If no model is defined, 'default' is used.
  # modelRef:
  #   name: gpt-4-model
  #   namespace: default

Creating Your First Agent

Let’s choose the Weather Agent defined in samples/agents/weather.yaml. Upon noticing, we can see that this agent uses two custom tools which are as follows:

get-coordinates (defined in samples/tools/get-coordinates.yaml)
get-forecast (defined in samples/tools/get-forecast.yaml)

Before we can actually query the agent for the forecast in a particular city, we have to apply these Tools as follows:


kubectl apply -f samples/tools/get-coordinates.yaml

and


kubectl apply -f samples/tools/get-forecast.yaml

After these, now’s the time to actually have the agent running. We can do so as follows:


kubectl apply -f samples/agents/weather.yaml

Now, let’s also have the query applied using


kubectl apply -f samples/queries/weather.yaml

After all the above steps, we can now query the agent using fark as below:


fark query weather-query

Modifying Agents

You can modify existing agents in several ways:

Update Agent Configuration


# Update the agent's prompt
kubectl patch agent weather --type='merge' -p='{
  "spec": {
    "prompt": "You are an expert weather forecaster. Provide detailed analysis."
  }
}'
 
# Add tools to an existing agent
kubectl patch agent weather --type='merge' -p='{
  "spec": {
    "tools": [
      {"type": "custom", "name": "get-coordinates"}
    ]
  }
}'
 
### Edit Agent Files Directly
 
```bash
# Edit agent configuration interactively
kubectl edit agent weather
 
# Or update from a modified file
kubectl apply -f samples/agents/weather.yaml

Managing Agent Resources

List and Inspect Agents


# List all agents
kubectl get agents
 
# Get detailed information about an agent
kubectl describe agent weather
 
# View the full agent configuration
kubectl get agent weather -o yaml
 
# Check agent status through ARK CLI
ark check status

Verify Agent Functionality


# Test the agent after creation or modification
fark agent weather "What's the weather in San Francisco?"
 
# Check recent queries
kubectl get queries
 
# View query details
kubectl describe query weather-query

Deleting Agents

Remove agents when they’re no longer needed:


# Delete a specific agent
kubectl delete agent weather
 
# Delete multiple agents
kubectl delete agent weather math-agent code-agent
 
# Delete all agents (use with caution!)
kubectl delete agents --all

Ready for multi-agent coordination? Learn about Teams.