First model deployment

Add the first dn/* model deployment so chat, agents, and evaluations have something to run on.

A fresh install has no model deployments, and every model picker stays empty until you add one. Do this once, right after you create the platform owner. It is the same on Helm and Embedded Cluster — model deployments live in the product, not in the chart, so nothing here requires a redeploy.

Save a credential

Sign in as the platform owner and open Admin → Credentials → Add credential. For a self-hosted OpenAI-compatible endpoint, save:

Name:      gpu-cluster
API Key:   <endpoint-api-key>
API Base:  http://vllm.internal:8000/v1

Use the URL reachable from the LiteLLM pod. For a commercial provider such as OpenAI or Anthropic, save the key but leave API Base empty.

Add a deployment

Open Admin → Model Deployments, and select Add deployment:

Model Name:     dn/security-model
Provider Model: openai/<model-id>
Credential:     gpu-cluster

Model Name must start with dn/. A deployment without that prefix is invisible to every model surface in the platform.
Put the provider’s own model identifier in Provider Model, prefixed with the provider — openai/… for any OpenAI-compatible endpoint you host, bedrock/…, azure/…, and so on. LiteLLM’s provider documentation is the reference for the exact string and any extra parameters your provider needs; Configure any provider explains how to map it onto these fields.
Select Test connection and confirm it succeeds before choosing Create deployment.

If you are moving between environments, Actions → Import deployments loads a previously exported configuration instead. Exports never include API keys.

What’s next

Model deployments Configure any provider, credentials, api_base precedence, custom pricing, and routing.

Connect clients Point the TUI, CLI, and SDK at your deployment.