NUFI Docs

Models and presets

Switch between AI models, tune the response style, and save presets.

NUFI can offer several AI models. You choose which one answers each conversation, and you can fine-tune how it replies.

Pick a model

Click the model name at the top-left of the conversation (in the screenshot below it reads gemini).
A menu opens listing the model groups available to you — for example Nufi and Sun LLM. Use the Search models… box to filter when the list is long.
Click a group, then pick a model from it. The conversation now uses that model. A ✓ marks the one currently selected.

The model picker open, showing the available model groups

The list updates automatically — when your admin adds a new model, refreshing the page is enough to see it. Newer or larger models are usually more capable but slower and cost more against your budget; smaller models are faster and cheaper. Try a few and see what fits.

Tune the response style

Open Parameters from the left icon rail. A panel slides out where you can shape how the model behaves:

The Parameters panel with model settings

The settings you'll actually reach for:

  • Custom Instructions — what the model should know before your first message: a persona, a tone, standing rules. This steers the whole conversation. ("You are a concise assistant. Reply in Vietnamese.")
  • Temperature — low (near 0) for precise, repeatable answers (code, summaries, data). High (near 1) for creative, varied answers (brainstorming, copy-writing).
  • Max Output Tokens — caps how long a single reply can be.
  • Web Search — toggle live web lookups for this conversation (see Web search).

The advanced controls (Top P, frequency / presence penalty, reasoning effort) are sampling knobs — leave them alone unless you know what you want. Click Reset Model Parameters at the bottom to go back to defaults.

Settings apply to the current conversation only. Start a new chat to reset, or save your favourite combination as a preset.

Presets

A preset is a saved bundle of { model, parameters, instructions } under a name.

The Presets menu

  • See your presets — click Presets in the top bar. When you have none yet it says "No presets yet, use the settings button to create one."
  • Create one — set up a model and its parameters the way you like, then save it from the settings/parameters panel and give it a clear name (e.g. "Code review", "Translation EN→VN", "Friendly reply").
  • Use one — pick it from the Presets menu; new conversations start with all its settings applied.
  • Import / Clear all — bring in a preset someone exported, or wipe your list, from the same menu.

Saved prompts

A saved prompt is lighter than a preset — just a chunk of text you reuse often, with no model or parameters attached.

Manage them under Prompts in the left icon rail. Insert one mid-conversation by typing / at the start of the message box and picking from the list. Good candidates: frequently-used personas, tedious boilerplate, tone modifiers like "Rewrite in plain English, no jargon."

If the model menu is empty

  • Refresh the page once.
  • If it's still empty, the platform can't reach the AI provider — ask your admin; they have monitoring that tells them what's wrong.
  • If models appear but every reply errors with "unauthorised", open the console once. The first visit sets up your account automatically.