AI Settings & Response Language
AI Settings & Response Language
Section titled “AI Settings & Response Language”The AI Settings page lets you configure your AI bot’s core behavior — response language, guardrail mode, system prompt, SLA targets, and auto-assign. Go to Settings → AI Settings in the left menu.
Response Language
Section titled “Response Language”By default, the AI auto-detects each visitor’s language and responds in the same language. Use these settings to override that behavior at the organization level.
Main Response Language
Section titled “Main Response Language”The Main Response Language dropdown sets the organization’s primary language. Options include all 15 supported locales:
| Locale | Language |
|---|---|
| Thai | ภาษาไทย |
| English | English |
| Chinese (Simplified) | 中文（简体） |
| Chinese (Traditional) | 中文（繁體） |
| Japanese | 日本語 |
| Korean | 한국어 |
| Vietnamese | Tiếng Việt |
| Indonesian | Bahasa Indonesia |
| Malay | Bahasa Melayu |
| Arabic | العربية |
| Hindi | हिन्दी |
| French | Français |
| German | Deutsch |
| Spanish | Español |
| Portuguese | Português |
When set to a specific language, the AI uses that language as its primary mode and the greeting defaults to that language in Org-Main greeting mode (see Widget Customization — Greeting Mode).
Leave this field blank (default) to keep full auto-detect — the AI reads each message and responds in the visitor’s language.
Force This Language
Section titled “Force This Language”The Force this language toggle is only active when a Main Response Language is selected.
| Toggle State | Behavior |
|---|---|
| OFF (default) | AI responds in the visitor’s detected or selected language; the main response language is a preference, not a rule |
| ON | AI always responds in the selected language, regardless of the visitor’s language or widget language selector choice |
When to use Force:
- Your business operates in one language only (e.g., a Thai-only support team)
- You want uniform AI output regardless of which language visitors type in
- You need to ensure human agents can always read the AI’s responses before handoff
When NOT to use Force:
- You serve customers in multiple languages
- You have the widget language selector enabled — forcing a language creates a confusing experience when visitors select a language but responses still come back in the org language
Guardrail Mode
Section titled “Guardrail Mode”Controls how strictly the AI stays on-topic:
| Mode | Behavior |
|---|---|
| Hybrid (default) | Handles on-topic questions fully; attempts short helpful answers to off-topic questions |
| Strict | Only answers questions directly supported by your knowledge base; refuses off-topic questions |
| Open | Answers any question using general knowledge, even without a matching document |
Custom System Prompt
Section titled “Custom System Prompt”An optional prompt that is prepended to every AI request. Use it to:
- Define a persona (“You are Aria, a friendly support agent for Acme Corp.”)
- Set tone and style (“Always be concise and professional.”)
- Add domain-specific context (“This bot supports Thai users only.”)
Maximum length: 5,000 characters.
SLA Targets
Section titled “SLA Targets”Set response and resolution time targets for human agent handoffs:
| Setting | Description |
|---|---|
| First Response SLA | Target minutes for the first agent reply after a conversation is handed off |
| Resolution SLA | Target minutes for a conversation to be marked resolved after handoff |
SLA metrics are tracked in Analytics → Dashboard and shown in the Conversations inbox.
Auto-Assign
Section titled “Auto-Assign”When enabled, handoff conversations are automatically routed to an available agent instead of waiting in the unassigned queue. Agents must be marked Available in the Conversations inbox to receive auto-assigned conversations.