In Admin > Settings > Interface, configure chat interface automations and UI elements.
The screen has two main sections: Tasks (auto-generation features, task model) and UI (banners, prompt suggestions).
Task Model
Specify the model used for background automation like title generation, tag generation, and follow-up suggestions.
| Setting | Description |
|---|
| Local Models | Pick from models registered in the system |
| External Models | Enter model ID for the task model from a separate external API |
The task model needs fast response and low cost above all. Use lightweight models (e.g., GPT-4o-mini, Claude Haiku, Gemini Flash).
When the task model is unset, the currently selected chat model is used for background tasks too.
In a chat using a high-cost model (Opus, GPT-4o, etc.), the same model is invoked every turn for title/tags/follow-up generation, greatly increasing cost. Always set a separate task model in production.
Auto-Generation Features
Configure activation and prompts for chat-related auto-generation Tasks. Each Task can be toggled ON/OFF and have its prompt template customized.
| Task | Behavior | Default |
|---|
| Title Generation | Auto-generate title from chat content after the first turn | ON |
| Tags Generation | Auto-extract and assign tags from chat content | ON |
| Follow-Up Generation | Suggest 3–5 follow-up questions as a button list below the last assistant response | OFF |
| Autocomplete Generation | Suggest message autocomplete during user input | ON |
Follow-Up Generation ships OFF by default. Showing follow-up question buttons to users incurs additional LLM calls every turn — turn it on after considering cost and response latency.
When enabled, suggested questions appear as a vertical button list below the assistant response immediately after it ends.
Prompt Customization
Leaving each Task’s prompt empty uses the system default; entering a prompt overrides it.
| Prompt | Description |
|---|
| Title Generation Prompt | Length/style for generating chat titles |
| Tags Generation Prompt | Categories/language for tag extraction |
| Follow-Up Generation Prompt | Tone/perspective/count of follow-up questions (e.g., “3 short questions predicting the user’s next action”) |
| Value | Meaning |
|---|
-1 | No length limit (try autocomplete on every input) |
Positive integer ≥ 1 | Only call autocomplete when input is ≤ that length |
200) in production.
Prompt Suggestions
Configure default prompt suggestions shown to users when starting a new chat.
Each suggestion consists of Title, Subtitle, and Content (the actual prompt sent). Add with the + button and delete with each item’s × button.
For per-agent prompt suggestions, set Custom in the agent edit screen’s Prompt Suggestions section. Per-agent settings take precedence over this global setting.
Banners
Show announcement banners across the system. Use for maintenance schedules, policy changes, etc.
| Setting | Description |
|---|
| Banner type | info, warning, error, success |
| Content | Banner message (HTML supported) |
| Dismissible | Whether users can close the banner |
