For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.

Learn how to configure Azure AI Foundry models in kagent.

Configuring Azure AI Foundry

Note: Foundry chat models and memory embeddings require the Go agent runtime (runtime: go). Chat models must be available through Foundry’s OpenAI-compatible chat completions API. Claude and other models are not yet supported.

The following steps use API key authentication. To authenticate without an API key, see Workload Identity.

1. Create a Kubernetes Secret that contains your Foundry API key.

export FOUNDRY_API_KEY="<your-foundry-api-key>"

kubectl create secret generic foundry-api-key \
  --namespace kagent \
  --from-literal=api-key="${FOUNDRY_API_KEY}"

2. Create a ModelConfig for your Foundry deployment.

apiVersion: kagent.dev/v1alpha2
kind: ModelConfig
metadata:
  name: foundry-chat
  namespace: kagent
spec:
  provider: Foundry
  model: gpt-4.1-nano
  apiKeySecret: foundry-api-key
  apiKeySecretKey: api-key
  foundry:
    endpoint: https://my-foundry-account.cognitiveservices.azure.com/
    deployment: gpt-4-1-nano
    apiVersion: "2024-10-21"

3. Reference the ModelConfig from an agent that uses the Go runtime.

apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
  name: foundry-agent
  namespace: kagent
spec:
  type: Declarative
  declarative:
    runtime: go
    modelConfig: foundry-chat
    systemMessage: "You are a helpful assistant."

4. Save the manifests from steps 2 and 3 as foundry-model.yaml and foundry-agent.yaml, then apply them to your cluster.

kubectl apply -f foundry-model.yaml
kubectl apply -f foundry-agent.yaml

Alternatively, apply only foundry-model.yaml, then select the ModelConfig from the Model dropdown when you create or update an agent in the kagent UI.

ModelConfig reference

Authentication

The runtime chooses a credential based on the fields in the ModelConfig.

Workload Identity

Omit apiKeySecret and apiKeyPassthrough to use Azure Workload Identity. For local development, the runtime tries to authenticate using your Azure CLI login.

Note: The runtime validates Azure credentials at startup. If credentials cannot be resolved, the agent does not become ready.

apiVersion: kagent.dev/v1alpha2
kind: ModelConfig
metadata:
  name: foundry-chat
  namespace: kagent
spec:
  provider: Foundry
  model: gpt-4.1-nano
  foundry:
    endpoint: https://my-foundry-account.cognitiveservices.azure.com/
    deployment: gpt-4-1-nano
    apiVersion: "2024-10-21"

See Configure the agent for Azure Workload Identity for the required agent settings.

Token passthrough

Set apiKeyPassthrough: true, then send the Foundry API key as the bearer token in each Agent2Agent (A2A) request. The runtime forwards that value to Foundry as the API key. Use Workload Identity, not token passthrough, for Microsoft Entra ID authentication.

apiVersion: kagent.dev/v1alpha2
kind: ModelConfig
metadata:
  name: foundry-passthrough
  namespace: kagent
spec:
  provider: Foundry
  model: gpt-4.1-nano
  apiKeyPassthrough: true
  foundry:
    endpoint: https://my-foundry-account.cognitiveservices.azure.com/
    deployment: gpt-4-1-nano

Endpoint from a ConfigMap

To use an endpoint from a ConfigMap, set foundry.endpointFrom to the ConfigMap name and key. For example, Azure Service Operator (ASO) can provision the account and write its endpoint to a ConfigMap.

apiVersion: v1
kind: ConfigMap
metadata:
  name: foundry-account
  namespace: kagent
data:
  endpoint: https://my-foundry-account.cognitiveservices.azure.com/
---
apiVersion: kagent.dev/v1alpha2
kind: ModelConfig
metadata:
  name: foundry-chat
  namespace: kagent
spec:
  provider: Foundry
  model: gpt-4.1-nano
  foundry:
    endpointFrom:
      name: foundry-account
      key: endpoint
    deployment: gpt-4-1-nano
    apiVersion: "2024-10-21"

The ConfigMap must be in the same namespace as the ModelConfig.

Note: Updating the endpoint value triggers a rolling update of agent pods that use this ModelConfig as their primary model.

Configure the agent for Azure Workload Identity

First, follow the AKS Workload Identity deployment guide to configure your cluster and managed identity. Grant the identity the Cognitive Services User role on the Foundry resource.

Azure Workload Identity requires the azure.workload.identity/use: "true" label on the agent pod and the managed identity client ID on its ServiceAccount. Configure the agent using one of the following options.

Let kagent create the ServiceAccount

apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
  name: foundry-agent
  namespace: kagent
spec:
  type: Declarative
  declarative:
    runtime: go
    modelConfig: foundry-chat
    systemMessage: "You are a helpful assistant."
    deployment:
      labels:
        azure.workload.identity/use: "true"
      serviceAccountConfig:
        annotations:
          azure.workload.identity/client-id: <managed-identity-client-id>

The ServiceAccount has the same name and namespace as the agent. Therefore, use system:serviceaccount:<namespace>:<agent-name> as the subject of the Azure federated identity credential.

Note: If you use Helm to configure a shared ServiceAccount with controller.agentDeployment.serviceAccountName, annotate that ServiceAccount and follow Use an existing ServiceAccount.

Use an existing ServiceAccount

Create or reuse a pre-annotated ServiceAccount, then reference it from the agent. The pod label is still required.

apiVersion: v1
kind: ServiceAccount
metadata:
  name: foundry-workload-identity
  namespace: kagent
  annotations:
    azure.workload.identity/client-id: <managed-identity-client-id>
---
apiVersion: kagent.dev/v1alpha2
kind: Agent
metadata:
  name: foundry-agent
  namespace: kagent
spec:
  type: Declarative
  declarative:
    runtime: go
    modelConfig: foundry-chat
    systemMessage: "You are a helpful assistant."
    deployment:
      serviceAccountName: foundry-workload-identity
      labels:
        azure.workload.identity/use: "true"

Use system:serviceaccount:<namespace>:<service-account-name> as the subject of the Azure federated identity credential.

Memory embeddings

Configure a memory embedding ModelConfig for a Foundry embedding deployment.

apiVersion: kagent.dev/v1alpha2
kind: ModelConfig
metadata:
  name: foundry-embeddings
  namespace: kagent
spec:
  provider: Foundry
  model: text-embedding-3-small
  foundry:
    endpoint: https://my-foundry-account.cognitiveservices.azure.com/
    deployment: text-embedding-3-small
    apiVersion: "2024-10-21"
  # No API key: use Azure Workload Identity.

For the complete memory and embedding configuration and model requirements, see Agent Memory.

Troubleshooting

• The ModelConfig reports Accepted=False: Check whether the required endpointFrom ConfigMap and key exist with kubectl describe modelconfig MODEL_CONFIG_NAME --namespace NAMESPACE.
• The agent fails to become ready with a Workload Identity credential error: Confirm the pod label and ServiceAccount annotation, and verify that the federated credential subject matches the ServiceAccount used by the pod.
• Foundry returns 401 Unauthorized or 403 Forbidden: Confirm that the managed identity has the Cognitive Services User role on the Foundry resource, or that the configured API key has access to the resource.