Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Create Alias Route

Use this form to connect a public Alias to a Provider Deployment. Together, the active Routes for an Alias define its ordered primary and fallback choices. Applications continue to send the Alias name; they never select the Deployment ID directly.

The selected Alias and Deployment must belong to the current host. The Deployment registration must use the Alias environment, and its model capabilities plus registration restrictions must satisfy the Alias’s required capabilities.

Fields

FieldDescriptionExample
Host IdRead-only tenant boundary supplied by the portal. Both referenced records must belong to this host.10000000-0000-4000-8000-000000000001
Public AliasNon-deleted Alias that clients use as their stable model name. Publication validates its routes and provider material.kb-index
Provider DeploymentNon-deleted Deployment that can serve the Alias. Its environment and embedding capabilities must match the Alias.nvidia-nemotron-3-embed-1b-loc
Route PriorityNon-negative ordering value. Lower values are evaluated first and must be unique within the Alias.0
Route WeightRead-only value fixed at 1 for the current MVP. Weighted selection is not supported yet.1
Fallback EnabledSelect when this Deployment should be used as a fallback rather than the preferred route.false
Canary PercentRead-only value fixed at 0 for the current MVP. Percentage-based canary routing is not supported yet.0
Residency ConditionsJSON or YAML governance object describing route residency constraints. Use Apply after editing. Use {} when no approved restriction applies. Current preview does not evaluate arbitrary conditions.{}

An Alias cannot contain the same Deployment twice. It also cannot contain two Routes with the same priority. A common convention is 0 for the preferred route and increasing values such as 10 and 20 for subsequent choices.

Primary Route Example

{
  "publicAliasId": "20000000-0000-4000-8000-000000000020",
  "providerDeploymentId": "30000000-0000-4000-8000-000000000030",
  "routePriority": 0,
  "routeWeight": 1,
  "fallbackEnabled": false,
  "canaryPercent": 0,
  "residencyConditions": {
    "regions": ["ca-central-1"]
  }
}

Fallback Route Example

For a second compatible Deployment, use another unique priority and enable fallback:

{
  "publicAliasId": "20000000-0000-4000-8000-000000000020",
  "providerDeploymentId": "30000000-0000-4000-8000-000000000031",
  "routePriority": 10,
  "routeWeight": 1,
  "fallbackEnabled": true,
  "canaryPercent": 0,
  "residencyConditions": {
    "regions": ["ca-central-1"]
  }
}

Creating a Route does not by itself make the Alias publishable. The Alias and Deployment must be present and internally consistent, and the Deployment needs an effective Credential and effective Pricing. The backend generates the Alias Route Id and aggregate version. The active state is backend-managed through soft delete and is not part of this form.

NVIDIA Knowledge Base routes

Submit this form twice: create one priority-zero Route for each Knowledge Base Alias. For the functional demo, both Routes may select the same hosted NVIDIA Deployment:

AliasDeploymentPriorityFallbackWeightCanary
kb-indexnvidia-nemotron-3-embed-1b-loc0false10
kb-querynvidia-nemotron-3-embed-1b-loc0false10

First submission:

{
  "publicAliasId": "select kb-index",
  "providerDeploymentId": "select nvidia-nemotron-3-embed-1b-loc",
  "routePriority": 0,
  "routeWeight": 1,
  "fallbackEnabled": false,
  "canaryPercent": 0,
  "residencyConditions": {}
}

Second submission uses the same values but selects kb-query as Public Alias. The values shown for the two selectors are dropdown labels; the form submits their UUIDs.

Use {} for Residency Conditions unless the Registration and Deployment carry an approved region restriction. Every routed Deployment must match the Alias’s complete embedding-space contract, not only dimension 2048.

Production-protected kb_index and kb_query lanes require genuinely separate runtime and capacity/quota domains. Routing both Aliases through one free shared NVIDIA Deployment is appropriate for this functional demo, but it is not production-isolation evidence.