AI provider, reply style and translation
Choose where your AI replies come from and how multilingual operators work.
Start with the source of the answer. The provider and model generate text, Knowledge supplies the facts in published articles, and Rules carry out configured actions such as tagging or handing off a conversation. A different reply style cannot repair a mistaken Knowledge article. Check the part of the workflow that produced the problem.
Set the response style
Open AI settings → Response style. The available controls are Warmth, Formality, Response length and Emoji use. For instance, TopSoft4U can choose brief, neutral replies for routine order questions. These options affect the tone and format, not which Knowledge articles are available.
Choose a model if your plan supports it
The Provider and Model selectors are available to organizations entitled to use their own AI providers. Open AI settings → AI providers to add a supported provider, configure its model and save credentials. This area is separate from response style. Keep real API keys out of public examples.
Translate for the operator
Go to AI settings → Translation to review how incoming and outgoing messages are translated and which language an operator uses. Test the behavior on a multilingual conversation before changing the workflow for your entire support team.
Decide when translation runs
For incoming messages, Automatically translate visitor messages can translate as conversations arrive, or operators can request translations when needed. For outgoing text, choose Operator approves the translation (recommended) to check wording before delivery, or Send the translation without a preview when that extra review is not part of your workflow.
Set the Team working language and review Terminology. Protected terms help keep brand names and product labels intact across languages. For a TopSoft4U support team, an incorrect product name can make a useful answer misleading. Watch the translation allowance if your plan includes usage limits, and test the complete receive-review-reply sequence before enabling automatic sending.
Trace a missing AI response
When AI doesn't answer, start with its availability and escalation settings. Then check whether a Rule sent a prepared reply or handed off the case. If AI reached Knowledge but had too little evidence, review the published articles and confidence policy in Knowledge. A different response style cannot repair a missing article.
This is different from choosing the languages of articles, Rules and the widget. For customer-facing translations, see Content languages. For answering hours and grounding rules, see AI availability.
Was this page helpful?