Configure the Adapter

The Omnia adapter is configured through the deploy.config section of your arena configuration file. This guide covers every available option.

Minimal configuration

deploy:
  provider: omnia
  config:
    api_endpoint: "https://omnia.example.com"
    workspace: "my-workspace"
    providers:
      default: claude-prod

Full configuration

deploy:
  provider: omnia
  config:
    api_endpoint: "https://omnia.example.com"
    workspace: "my-workspace"
    api_token: "optional-inline-token"
    providers:
      default: claude-sonnet-4-20250514
      router-agent: gpt4-prod
    runtime:
      replicas: 2
      cpu: "1000m"
      memory: "1Gi"
    labels:
      team: platform
      environment: production
    dry_run: false

Field reference

api_endpoint (required)

The base URL of the Omnia Management API. The adapter appends /api/v1/workspaces/<workspace>/ to form the full API path.

api_endpoint: "https://omnia.example.com"

Must be a valid URI. Trailing slashes are stripped automatically.

workspace (required)

The Omnia workspace to deploy into. Must be a valid Kubernetes DNS subdomain: lowercase alphanumeric characters and hyphens, starting and ending with an alphanumeric character.

workspace: "prod-workspace"

Pattern: ^[a-z0-9][a-z0-9-]*[a-z0-9]$

api_token (optional)

Bearer token for authenticating with the Omnia Management API. If omitted, the adapter reads the OMNIA_API_TOKEN environment variable.

api_token: "your-token"

The environment variable approach is strongly recommended for CI/CD pipelines and shared configurations to avoid committing secrets:

export OMNIA_API_TOKEN="your-token"

One of api_token or OMNIA_API_TOKEN must be set. Validation fails if neither is provided.

providers (required)

A map of Arena provider names to Omnia Provider CRD names. At minimum, include a default entry. For multi-agent packs, you can add agent-specific overrides keyed by agent name.

providers:
  default: claude-sonnet-4-20250514
  router-agent: gpt4-prod
  specialist-agent: claude-sonnet-4-20250514

Resolution order:

1. Exact match on the agent name.
2. Fall back to the default entry.

runtime (optional)

Resource sizing for AgentRuntime CRDs.

runtime:
  replicas: 2
  cpu: "500m"
  memory: "512Mi"

Sub-field	Type	Description
replicas	integer	Number of runtime replicas. Must be >= 1.
cpu	string	CPU request/limit in Kubernetes resource format (e.g., "500m", "1").
memory	string	Memory request/limit in Kubernetes resource format (e.g., "512Mi", "1Gi").

If runtime is omitted, the Omnia platform applies its own defaults.

labels (optional)

Extra labels to apply to all created resources. These are merged with the adapter’s managed labels.

labels:
  team: platform
  environment: production
  cost-center: eng-42

User-supplied labels cannot override the adapter’s managed labels. See Resource Labels for details.

dry_run (optional)

When set to true, the Apply operation generates a deployment preview without making any API calls. Resources are planned but not created.

dry_run: true

Default: false. See Use Dry-Run Mode for details.

Validation

The adapter validates the configuration before any operation. Validation checks:

• api_endpoint is non-empty
• workspace is non-empty
• providers contains at least one entry
• An API token is available (from config or environment)
• runtime.replicas is >= 1 (if runtime is specified)

Validation errors are returned as a list of human-readable messages.