Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.inboxmate.psquared.dev/llms.txt

Use this file to discover all available pages before exploring further.

​
Write a good system prompt

The system prompt is the single most important factor in response quality. Do:
  • Name the company and what it does
  • Define the tone (formal, casual, friendly)
  • List topics the agent should and shouldn’t cover
  • Include fallback instructions (“If unsure, suggest emailing support@…”)
Don’t:
  • Leave it vague (“You are a helpful assistant”)
  • Give conflicting instructions
  • Try to cover every edge case — focus on the 80% of questions

​
Structure your knowledge

Rather than dumping your entire website into one bucket, create separate knowledge items for distinct topics: pricing, shipping, returns, product specs. This helps the agent retrieve the right context.
Knowledge items don’t need marketing language. Use clear, factual statements. Instead of “Our award-winning shipping is blazing fast!”, write “Standard shipping: 3-5 business days. Express: 1-2 business days. Free shipping on orders over €50.”
Stale knowledge leads to wrong answers. Review and update your knowledge items when prices, policies, or products change.

​
Set up quick questions

Quick questions appear as clickable suggestions in the chat widget. They:
  • Reduce friction for first-time visitors
  • Guide conversations toward topics your agent handles well
  • Increase engagement rates
Choose 3-4 questions that represent your most common customer inquiries.

​
Test before going live

Use the Widget tab in your agent editor to preview and test your agent with real questions before embedding it. Try:
  • Common customer questions
  • Edge cases and off-topic questions
  • Questions in different languages (if you support multiple)
Changes are saved and take effect immediately. Use the widget preview to verify everything looks correct before embedding on your site.

​
Monitor and improve

After launch, check your Analytics dashboard regularly:
  • Review conversations where customers weren’t satisfied
  • Look for questions the agent couldn’t answer — add knowledge to cover those gaps
  • Track response times and satisfaction scores over time