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

Define a Model resource and publish it to the registry catalog so harness agents can reference it at deploy time.

A model is a versioned catalog entry that declares which AI model a harness agent should call and how to authenticate with the model provider. When you deploy a harness agent, the registry reads the model configuration from the catalog and passes it to the harness runner so it knows which endpoint to call and which credentials to use.

Before you begin

1. Set up an OIDC provider so that users can authenticate with the registry.

2. Install Solo Enterprise for agentregistry in an AWS EKS cluster and the arctl CLI.

3. Port-forward the Solo Enterprise for agentregistry server on port 12121 if you have not done so already. The registry is not exposed externally by default, so you need port-forwarding to reach it from your local machine.

   kubectl -n agentregistry-system port-forward svc/agentregistry-enterprise-server 12121:12121

4. Save the AWS region where you plan to deploy the harness agent in an environment variable. Make sure that the Bedrock model you want to use is available in this region. This value is used in the model definition.

   export AWS_REGION=<aws-region>

Create and publish a model

Platform admins control which models are available to harness agent deployers by publishing them to the catalog. Deployers can then reference any published model by name in the harness agent deployment.

The example in this guide walks you through how to set up a model and publish it to the registry catalog. If you want to set an organization-wide default model, see Set an organization default model.

1. Create a model definition and push it to the registry catalog. The following example uses the claude-sonnet-4-6 model. Make sure that the model that you select is supported in your AWS region.

   arctl apply -f - <<EOF
   apiVersion: ar.dev/v1alpha1
   kind: Model
   metadata:
     name: claude-sonnet-4
   spec:
     title: Claude Sonnet 4
     provider: bedrock
     model: us.anthropic.claude-sonnet-4-6
     auth:
       strategy: runtime
     endpoint:
       region: $AWS_REGION
   EOF

   Field	Description
   spec.title	Human-readable display name that is shown in the catalog UI.
   spec.description	Optional description that is shown in catalog listings.
   spec.provider	The model provider. Currently only bedrock (AWS Bedrock) is supported.
   spec.model	The model ID to use. For AWS Bedrock, this is the inference profile ID, for example us.anthropic.claude-sonnet-4-6. For supported models and their availability by region, see the AWS Bedrock documentation.
   spec.auth.strategy	How the harness runner authenticates with the model provider. Use runtime for harness deployments on AWS Bedrock AgentCore. With runtime, the harness runner uses the AgentCore execution role’s IAM permissions to access Bedrock. No API key is required.
   spec.endpoint.region	The AWS region where Bedrock API calls are made. The model you select in spec.model must be available in this region. Use the same region that you plan to deploy harness agents in.

   Example output:

   ✓ Model/claude-sonnet-4 (latest) created
   

2. Verify that the model is created in the catalog.

   arctl get models

   Example output:

   NAME               TAG      PROVIDER   MODEL                              AUTH
   claude-sonnet-4    latest   bedrock    us.anthropic.claude-sonnet-4-6     runtime
   

Set an organization default model

Platform admins can set a default model that all harness agent deployments use by default by nameing the Model resource default. This way, harness agent deployments do not need to reference a Model config.

arctl apply -f - <<EOF
apiVersion: ar.dev/v1alpha1
kind: Model
metadata:
  name: default
spec:
  title: Organization default model
  provider: bedrock
  model: us.anthropic.claude-sonnet-4-6
  auth:
    strategy: runtime
  endpoint:
    region: $AWS_REGION
EOF

Next steps