Presets

Bundle a model, system prompt, and parameters into one reusable name.

A preset bundles a model choice, system prompt, and generation parameters (temperature, max_tokens, top_p) into a single name you can reference from any client, instead of repeating the same configuration in every request.

Create a Preset

  1. Go to Presets in the dashboard.
  2. Select New preset.
  3. Set the model, system prompt, and parameters.
  4. Name the preset. This is the name you'll reference in requests.
  5. Select Save preset.

Reference the preset by name in the model field, using the @preset/ convention:

POST https://gateway.relixr.com/v1/chat/completions
Authorization: Bearer nf_live_...

{
"model": "@preset/support-bot",
"messages": [{"role": "user", "content": "My order is late"}]
}

// Gateway resolves @preset/{name} to the preset's model,
// system prompt, temperature, max_tokens, and top_p.
Tip:

Presets are the fastest way to change a model version for every caller at once — update the preset in the dashboard, and every integration referencing @preset/support-bot picks up the change on its next request. No redeploy needed.

Presets sync to the gateway as soon as you save. Assign a default preset to an inference key from the key detail page, so callers using that key don't need to specify one at all.

Next steps
Model routing
Route between models a preset might reference.
Agents
Bind a preset to an agent-scoped key.
← Bring your own keysAgents →