model field in a request does double duty, naming both what the caller wants (“summarize this”) and which model answers (“OpenAI’s gpt-4o”). Tie those together and changing the model behind a task means changing application code.
A task alias separates the two. Your application calls a stable name; the model behind it is configuration. Swapping gpt-4o for gpt-4o-mini becomes a dashboard change, not a deploy, as long as the replacement’s provider speaks the wire format your application sends.
How you get this today
In the to11 dashboard, you express task aliases through named routes. When you create a routing rule in AI Gateway → Routing, you give it a routing identifier and your application calls it asroute::<name>:
route::summarize follows along, with no code change on your side. See Route by a stable name.
The gateway’s lower-level functions — task aliases with advanced per-endpoint variant configurations — are managed by to11 and don’t have a self-serve dashboard surface yet. For the routing customers configure themselves, use named routes and the strategies above.
Next steps
Route by a stable name
Create a named route and swap models without redeploying.
Routes
How routes map a name to a strategy across targets.