Using Devtron Intelligence

What is Devtron Intelligence (AI Agent)

Devtron Intelligence is an AI assistant that helps you troubleshoot issues faster by analyzing your Kubernetes workloads. It offers smart and easy-to-understand suggestions using large language models (LLM) of your choice.

Check out the Results section to see where Devtron gives you AI-powered explanation for troubleshooting.

Tutorial

{% embed url="https://www.youtube.com/watch?v=WW7skAa0XAs" %}

---

Steps to Configure Devtron Intelligence

{% hint style="warning" %}

Who Can Perform This Action?

User must have permissions to:

• Deploy Helm Apps (with environment access)
• Edit the ConfigMaps of 'default-cluster'
• Restart the pods {% endhint %}

1. Get API Key from LLM

Devtron Intelligence supports all major large language models (LLM) e.g., OpenAI, Gemini, AWS Bedrock, Anthropic and many more.

You can generate an API key for an LLM of your choice. Here, we will generate an API key from OpenAI.

2. Create Secret in Devtron

There are 2 methods to create a secret in Devtron, follow the one you prefer:

• Method A: Using 'Create Resource'
• Method B: Using kubectl command

Method A: Using 'Create Resource'

1. Go to strings.devtron.ai and encode your API key in base64. This base64 encoded key will be used while creating a secret in the next step.
2. Go to Devtron's Resource Browser → (Select Cluster) → Create Resource
3. Paste the following YAML and replace the key with your base64-encoded OpenAI key. Also, enter the namespace where the AI Agent chart will be installed:

apiVersion: v1
kind: Secret
metadata:
  name: ai-secret
  namespace: <your-env-namespace>  # Namespace where the AI Agent chart will be installed
type: Opaque
data: 
  ## OpenAiKey: <base64-encoded-openai-key>           # For OpenAI
  ## GoogleKey: <base64-encoded-google-key>           # For Gemini
  ## azureOpenAiKey: <base64-encoded-azure-key>       # For Azure OpenAI
  ## awsAccessKeyId: <base64-encoded-aws-access-key>  # For AWS Bedrock
  ## awsSecretAccessKey: <base64-encoded-aws-secret>  # For AWS Bedrock
  ## AnthropicKey: <base64-encoded-anthropic-key>     # For Anthropic

Method B: Using kubectl command

{% hint style="success" %} Unlike Method A, this method doesn't require you to encode your LLM Key to Base64 format. {% endhint %}

1. Go to Devtron's Resource Browser and click the terminal icon next to the cluster where you wish to create the secret.
2. Use the following kubectl command to create a secret.
   • Replace my-namespace with the namespace where the AI Agent chart will be installed.
   • Use the correct LLM key-name and your key-value after --from-literal

kubectl create secret generic ai-secret \
  --namespace=my-namespace \
  --from-literal=OpenAiKey='openai-key-here' \
#  --from-literal=GoogleKey='google-key-here' \
#  --from-literal=azureOpenAiKey='azure-key-here' \
#  --from-literal=AnthropicKey='anthropic-key-here'

3. Deploy AI Agent Chart

{% hint style="warning" %}

Where should I install the Chart?

Deploy the chart in the cluster whose workloads you wish to troubleshoot. You may install the chart in multiple clusters (1 agent for 1 cluster). {% endhint %}

1. Go to Devtron's Chart Store.
2. Search the ai-agent chart and click on it.
3. Click the Configure & Deploy button.
4. In the left-hand pane:
   • App Name: Give your app a name, e.g. ai-agent-app
   • Project: Select your project
   • Deploy to environment: Choose the target environment (should be associated with the same namespace used while creating secret key in Step 2)
   • Chart Version: Select the latest chart version.
   • Chart Values: Choose the default one for the latest version.
5. In the values.yaml file editor, add the appropriate additionalEnvVars block based on your LLM provider. Use the tabs below to find the configuration snippet of some well-known LLM providers.

{% tabs %} {% tab title="OpenAI" %}

additionalEnvVars:
  - name: MODEL
    value: gpt-4o-mini       ## Examples: gpt-4o, gpt-4, gpt-3.5-turbo
  - name: OPENAI_API_KEY
    valueFrom: 
      secretKeyRef:
        key: OpenAiKey       ## Key of the secret created in Step 2
        name: ai-secret      ## Name of the secret created in Step 2
  - name: CLUSTER_NAME
    value: document-nonprod  ## Name of the target cluster (optional)

{% endtab %}

{% tab title="Google" %}

additionalEnvVars:
  - name: MODEL
    value: gemini-1.5-pro    ## Examples: gemini-2.0-flash, gemini-2.0-flash-lite
  - name: GOOGLE_API_KEY
    valueFrom: 
      secretKeyRef:
        key: GoogleKey       ## Key of the secret created in Step 2
        name: ai-secret      ## Name of the secret created in Step 2
  - name: CLUSTER_NAME
    value: document-nonprod  ## Name of the target cluster (optional)

{% endtab %}

{% tab title="Azure OpenAI" %}

additionalEnvVars:
  - name: MODEL
    value: azure/<DEPLOYMENT_NAME>   ## Replace with your Azure deployment name (keep "azure/" prefix)
  - name: MODEL_TYPE
    value: gpt-4o                ## Supported: gpt-4o, gpt-35-turbo, etc.
  - name: AZURE_API_VERSION
    value: <API_VERSION>    ## Replace with the version from Azure portal
  - name: AZURE_API_BASE
    value: <AZURE_ENDPOINT>  ## Your Azure endpoint e.g. https://my-org.openai.azure.com/
  - name: AZURE_API_KEY
    valueFrom:
      secretKeyRef:
        key: azureOpenAiKey      ## Key of the secret created in Step 2
        name: ai-secret          ## Name of the secret created in Step 2
  - name: CLUSTER_NAME
    value: document-nonprod  ## Name of the target cluster (optional)

{% endtab %}

{% tab title="AWS Bedrock" %}

additionalEnvVars:
  - name: MODEL
    value: bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0  ## Replace with your actual Bedrock model name
  - name: AWS_REGION_NAME
    value: us-east-1
  - name: AWS_ACCESS_KEY_ID
    valueFrom:
      secretKeyRef:
        key: awsAccessKeyId      ## Key of the Access Key ID created in Step 2
        name: ai-secret          ## Name of the secret created in Step 2
  - name: AWS_SECRET_ACCESS_KEY
    valueFrom:
      secretKeyRef:
        key: awsSecretAccessKey  ## Key of the secret created in Step 2
        name: ai-secret          ## Name of the secret created in Step 2
  - name: CLUSTER_NAME
    value: document-nonprod  ## Name of the target cluster (optional)

{% endtab %}

{% tab title="Anthropic" %}

additionalEnvVars:
  - name: MODEL
    value: claude-3-sonnet   ## Examples: claude-3-sonnet, claude-3-haiku
  - name: ANTHROPIC_API_KEY
    valueFrom: 
      secretKeyRef:
        key: AnthropicKey    ## Key of the secret created in Step 2
        name: ai-secret      ## Name of the secret created in Step 2
  - name: CLUSTER_NAME
    value: document-nonprod  ## Name of the target cluster (optional)

{% endtab %} {% endtabs %}

6. Click the Deploy Chart button.

4. Check Service Endpoint

1. In the App Details page of the deployed chart, expand Networking and click on Service.
2. Locate the service entry with the URL in the format: <service-name>.<namespace>:<port>. Note the values of serviceName, namespace, and port for the next step.

5. Update ConfigMaps

1. In a new tab, go to Resource Browser → (Select Cluster) → Config & Storage → ConfigMap
2. Edit the ConfigMaps:

   • devtron-cm

     Ensure the below entry is present in the ConfigMap (create one if it doesn't exist). Here you can define the target cluster and the endpoint where your Devtron AI service is deployed:

     CLUSTER_CHAT_CONFIG: '{"<targetClusterID>": {"serviceName": "", "namespace": "", "port": ""}}'

     

   • dashboard-cm

     To enable AI integration via feature flag, check if the below entry is present in the ConfigMap (create one if it doesn't exist).

     FEATURE_AI_INTEGRATION_ENABLE: "true"

     

6. Restart Pods

1. Go to Resource Browser → (Select Cluster) → Workloads → Deployment

2. Click the checkbox next to the following Deployment workloads and restart them using the ⟳ button:

   • devtron
   • dashboard

   

7. Perform Hard Refresh

Perform a hard refresh of the browser to clear the cache:

• Mac: Hold down Cmd and Shift and then press R
• Windows/Linux: Hold down Ctrl and then press F5

---

Results

Devtron supports Explain option at the following screens (only for specific scenarios where troubleshooting is possible through AI):

Pod Errors

Path: Resource Browser → (Select Cluster) → Workloads → Pod

Pod Last Restart Snapshot

Path: Resource Browser → (Select Cluster) → Workloads → Pod → Pod Last Restart Snapshot

Event Errors

Path: Resource Browser → (Select Cluster) → Events

App Details - Application Status

Path: Application → App Details → Application Status Drawer

App Details - K8s Resources

Path: Application → App Details → K8s Resources (tab) → Workloads