Skip to main content
An alias is a stable, readable name for a deployment or an inference pool. Instead of hard-coding a raw deployment slug in your application, you call a friendly path that you control:
Valkyrie Alias Management screen

Alias Management. Sharing an alias never exposes the underlying instance ID.

Your account handle (the {account_slug} part of the URL) is assigned automatically when you create your first alias.

​
Every deployment already has a URL

You don’t have to create an alias. Every deployment is reachable at a default, non-guessable URL (a random public id), so it’s never addressed by a guessable sequential number. An alias is an optional, memorable name you add on top when you want something friendlier than the default id.

​
Why add an alias

  • Memorable. acme/support-bot is easier to share and reason about than the default public id.
  • Stable and re-pointable. Keep the same alias while you swap the deployment or pool behind it, so your applications never change their configuration.

​
Create or manage aliases

You’ll use this to add a custom alias (a name you choose) on top of a deployment’s default URL, or to manage an existing one. An alias is also offered as an optional field when you deploy a model.
1

Open Aliases

Go to Endpoints and click Aliases. From an endpoint’s page, you can also use the Create one link next to its URL.
2

Pick a target

Click New alias, enter a name, and choose a Target. The picker lists your deployments under Deployments and your pools under Inference pools. A disabled pool is marked disabled.
3

Call it

Use the alias path in place of the deployment slug in any OpenAI-compatible client.
The aliases table shows each alias’s URL and its Target: the deployment and its ID, or Inference pool. You can also manage aliases from the MCP server with list_aliases, create_alias, update_alias, and delete_alias. See MCP tools.

​
Re-point an alias

An alias name is fixed once created, but you can re-point it at a different target at any time (to rename, delete it and make a new one). Edit the alias and pick another deployment or pool. You can move an alias from a deployment to a pool and back. Point production traffic at an alias, then move the alias to a new target once you’ve verified it. Clients never need to change their configuration.

​
Point an alias at an inference pool

Use a pool alias to spread traffic over several deployments of the same model behind one memorable URL. Valkyrie sends each request to a pool member with a free request slot, and the pool’s queueing, retry, and health checks apply as they do on the pool itself.
  • Only the pool owner can target a pool. The pool must also be in the same workspace as the alias. Otherwise Valkyrie answers 404.
  • Call it with an account-wide API key. Keys scoped to one endpoint (vk_dep_ keys created for a single endpoint), the vk_mcp_ key, and dashboard sessions get 403. See API keys.
  • OpenAI-compatible routes only. A pool alias serves /v1/chat/completions, /v1/completions, and /v1/models. The Anthropic /v1/messages routes and /litellm/... routes return 400.
  • Session affinity. Send an X-Session-Affinity header (or X-Affinity-Key) to keep one session on one member and reuse its prompt cache. Without it, Valkyrie uses the request’s user field, then the start of the prompt.
Valkyrie sends the pool’s model name upstream, whatever model you pass. A pool alias adds two routes: On a deployment alias, both routes return 404.