One size doesn't fit all. A technical support agent needs a different personality than a sales agent. Our flexible configuration system lets you create multiple specialized agents, each optimized for its specific role and audience.

Powerful personality customization

Name — appears prominently in the widget header and sets expectations
Tone — helpful, professional, friendly, concise, technical
Language default — ensures responses match your target market

Tone dramatically impacts customer perception

Friendly — "Happy to help!" with casual contractions and warm language, perfect for consumer-facing support
Professional — formal language, no contractions, projects authority, ideal for enterprise B2B
Technical — precise terminology, assumes domain knowledge, suited for developer docs or expert users

The system uses different AI prompt templates for each tone, ensuring consistency across all responses.

Guided quick-start tasks reduce friction

Default task suggestions appear when customers open the widget, guiding them to common actions. Examples:

"How do I reset my password?"
"Show me pricing options"
"Schedule a demo"

Customers click these to auto-fill the input, reducing effort and directing conversations toward high-value interactions.

Confidence-based answering protects your brand

Enable "don't answer if unsure" to prioritize accuracy over coverage. When confidence is low, the AI gracefully defers: I don't see this clearly addressed in our documentation. Would you like to contact our support team? You control the confidence threshold — higher thresholds mean fewer but more accurate responses.

Knowledge scoping creates specialized experts

Exclude specific knowledge sources from each agent:

A Technical Documentation agent that ignores marketing content
A Sales agent that focuses only on pricing and product benefits
An HR Agent that accesses only policy documents

Built-in testing accelerates optimization

The configuration page includes a test interface where admins can ask questions and evaluate responses before deploying changes. See which knowledge chunks were used, their relevance scores, the exact prompt sent to the AI model, and the raw AI response before formatting.