Enterprise Deployment

Enterprise deployment of catalog agents is the Inference Gateway Operator's job. This page is the registry-side blurb - it covers how a catalog entry gets into the cluster, not Operator setup itself.

The Operator owns production

The Operator ships four CRDs under core.inference-gateway.com/v1alpha1: Gateway, Agent, Orchestrator, and MCP. HPA tuning, telemetry, ingress, multi-provider config, RBAC, secret management - all of that lives in the Operator repo. Start there for installation and the full CRD reference.

Don't hand-write the Agent CR

The CR is generated from the ADL manifest. Add a spec.deployment.kubernetes block to the agent's agent.yaml:

spec:
  deployment:
    type: kubernetes
    kubernetes:
      image:
        registry: ghcr.io
        repository: inference-gateway/documentation-agent
        tag: 0.6.3

Then adl generate --deployment kubernetes writes the matching Agent CR (and any supporting manifests) under the project's deploy/ directory. kubectl apply -f deploy/ installs it, the Operator reconciles it into a Deployment + Service + HPA, and orchestrators discover it. See the ADL CLI docs for the full adl generate flow.

Picking a version from the catalog

When you browse Agents, the OCI Image row on each card shows the published tag (e.g. ghcr.io/inference-gateway/documentation-agent:0.6.3). Use that tag in your ADL spec.deployment.kubernetes.image block so the generated CR pins to the exact image you reviewed. latest works for experimentation but should not be used in production - it breaks the link between the manifest you reviewed in the catalog and the image actually running.

Where to go next

• Operator README - installation, RBAC, CRD reference.
• Operator examples - full Gateway + Agents + Orchestrator stacks.
• ADL CLI docs - adl generate flags and the full deployment field reference.
• Gateway repo - the underlying gateway image and provider config.