Skip to main content
Model Routes allows Coding Plan subscribers to create virtual model aliases that map to a sequence (chain) of real models. When the primary model encounters upstream capacity exhaustion or rate limits, the gateway fails over to the next configured model in the chain without client-side retry logic.
Model Routes requires an active Coding Plan subscription (paid or trial) and operates exclusively through dedicated coding endpoints with coding keys (nsk_code_*).

​
How It Works

  1. You create a virtual alias (for example my-coder) and assign an ordered sequence of models (such as ["claude-4.5-sonnet", "gemini-3-flash"]).
  2. When submitting requests with model: "my-coder" to the coding endpoints, the gateway attempts the primary model first.
  3. If the primary provider reports an upstream capacity error, the gateway dispatches the payload to the next model in your chain.

​
Tier Limits

​
Route Creation Rules

  • Virtual alias names must not conflict with native catalog model IDs (for example, you cannot name a route claude-4.5-sonnet).
  • Route names are case-insensitive and unique to your user account.
  • Every model included in a chain must belong to your active tier allowlist.
  • Maximum length for virtual route names is 100 characters.

​
Using Virtual Routes in Clients

Once defined in your dashboard, supply the virtual alias to your coding agent:

​
1. Jelma CLI

Terminal

​
2. Claude Code

Edit configuration in ~/.claude/settings.json:
JSON

​
3. Cline / OpenCode

Set the model field in your settings panel:
Configuration