Configure the AI chatbot

The AI bot answers recurring questions from your knowledge base. When it kicks in and when it hands off to a human is decided per widget.

Activation modes

Three activation strategies — combinable:

First message: Bot answers the visitor's first message. Default for FAQ-heavy sites.
After delay: Bot only kicks in after X seconds without an agent reply — good for teams with live coverage.
Off-hours: Bot only takes over outside your business hours and writes a handoff note for the morning standup.

Knowledge-base wiring

The bot pulls answers exclusively from the connected knowledge base via vector search. Every answer includes the link to the cited article — visitors can read on, agents can audit.

Auto-handoff

The bot hands off to a human as soon as one of the conditions is met:

Confidence score on the answer drops below the configured threshold (default 0.7).
Visitor explicitly asks for a human or rates an answer negatively.
Sentiment analysis detects frustration, escalation, or legal language.

Suppression

Once a human agent starts typing in the conversation, the bot goes silent — no more double replies. If the conversation is handed back to the bot, it waits a beat before re-engaging.

Model selection

The default is the IONOS AI provider hosted in Germany, which keeps processing entirely in Germany. Plans that allow it can switch to Claude (Anthropic, US) or GPT-4o (OpenAI, US) instead; those process in the US (EU-US Data Privacy Framework / SCCs). Moving away from the EU default is recorded in the tenant’s audit log.