Presets
A preset stores a fallback model list, provider preferences, sampling parameters and a system prompt under one name. Manage presets in your account; the gateway merges them into the request before validation, so limits, balance and PDP mode still apply.
Three ways to reference a preset
| Form | Meaning |
|---|---|
| "model": "@preset/slug" | The preset's first model becomes model; its model list becomes models, unless the request sends its own models. |
| "model": "vendor/model@preset/slug" | The named model is kept; every other setting comes from the preset. |
| "preset": "slug" | A separate field. The request's model is kept; without a model, the preset's models are used. |
Merge rules
- Request fields always win over preset values (temperature, max_tokens, stop and the rest).
- A request provider object replaces the preset's provider as a whole; fields are not merged one by one.
- The preset's system prompt is prepended only when the first message is not a system or developer message.
- The preset field is never forwarded to the provider.
- The merged request is validated like any other request, so a preset cannot bypass limits.
Example
A preset named hemat, called with cURL or the regular OpenAI SDK:
curl "$ROUTER_BASE_URL/chat/completions" \
-H "Authorization: Bearer $ROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"@preset/hemat","messages":[{"role":"user","content":"Halo!"}]}'Errors
| Form | Meaning |
|---|---|
| 400 preset_not_found | The preset does not exist, belongs to another account, or is archived. Nothing is charged and no provider is called. |
| 400 preset_without_model | The preset has no model list; use the vendor/model@preset/slug form. |