AI assistant
Choose per project how much the assistant may do, under Project settings → AI.
| Mode | What happens |
|---|---|
| Off | No AI. Your team answers everything themselves. |
| AI Assist | Agents get reply suggestions, rewrites and summaries. Visitors only talk to people. |
| Auto mode | The assistant answers visitors first and hands over to your team when needed. Available on Team and Organization plans. |
Briefing
Section titled “Briefing”The briefing is everything the assistant should know: who you are, what you offer, opening hours and when to hand over. Write it in plain words; Markdown works. For longer or changing content, use the knowledge base, which the assistant searches for every question.
Helping your team
Section titled “Helping your team”- Suggest: type a few keywords in the reply box and get ready-to-send replies, based on the conversation, your briefing and the knowledge base.
- Rewrite a draft: warmer and friendlier, shorter, simpler language, more formal, or translated into another language.
- Knowledge panel: the conversation sidebar shows matching knowledge base entries for the latest message. Insert one or draft a reply from it.
- Summaries: each closed conversation gets a short summary and topic tags. Summaries never contain names or contact details.
Replies written with AI help are marked AI-assisted for your team.
Auto mode
Section titled “Auto mode”In auto mode, the assistant answers web chat and WhatsApp visitors with streaming replies. It only uses facts from your briefing, the knowledge base and the conversation, never claims to be a person and never gives diagnoses or medication advice. It hands over to people:
- when the visitor asks for a person or is unhappy with the answers,
- when the question isn’t covered or it is unsure,
- on sensitive topics such as health deterioration, child safety, violence, legal or financial hardship and complaints,
- on signs of crisis or self-harm, after pointing to the emergency number.
On handover, the assistant can pick a fitting group based on the groups’ descriptions. Otherwise routing rules decide, and if none matches, the default escalation group is used. Your team is notified, sees why the conversation was escalated and can take over at any time.
You decide when the assistant answers: always, only outside opening hours, or whenever nobody is online. To keep costs predictable, set a daily auto-mode AI Credit cap and a maximum number of assistant turns per conversation. When a limit is reached, the conversation goes to your team.
The daily cap defaults to 100 AI Credits per project and resets at midnight UTC. Requests reserve a conservative estimate, so the assistant may hand over before the displayed cap is fully used.
Set the assistant name and a disclaimer shown in the widget, for example “I’m an AI assistant. In an emergency call 144.”
Privacy and usage
Section titled “Privacy and usage”- AI requests only go to providers that do not store or train on your data.
- Optionally, known personal data is replaced with placeholders before prompts are sent (see Privacy).
- AI usage is counted in AI Credits per organisation. The Team plan includes 1,000 AI Credits per month; see Settings → Plan & billing.