Presets

Save a model, system prompt, and default parameters under a name, then call it by reference. Presets keep request payloads small and let you change behavior without redeploying your app.

Create a preset

Under Dashboard → Presets, give it a name and slug, then set:

  • model — a model alias, tigergateway/auto, or left blank to defer to routing.
  • system_prompt — prepended to the conversation as a system message.
  • temperature, max_tokens, top_p — default sampling parameters.

Use a preset

Reference it in the model field as @preset/<slug>:

json
{
  "model": "@preset/support-agent",
  "messages": [{ "role": "user", "content": "My order hasn't arrived." }]
}

The preset expands into its model, system prompt, and parameters, then normal routing and fallback apply — so a preset whose model is tigergateway/auto composes with the auto-router.

Override behavior

  • Parameters sent on the request override the preset's defaults.
  • The preset's system prompt is prepended ahead of any messages you supply.
  • An unknown slug returns 404 preset_not_found.

Request logs show the preset reference (e.g. @preset/support-agent) alongside the model that actually served.