Knowledge base: what your bot knows

Your bot answers only from the knowledge you give it. The Knowledge tab of each bot manages three source types:

Website pages

Add a URL and the bot crawls it — sitemap first, falling back to following links. Each page becomes a knowledge source with its own status (processing → ready). Useful controls:

• Discover links — list a site's pages and pick which to index instead of adding one by one.
• Re-crawl — refresh a page after you change your website; Re-crawl All refreshes everything.
• Duplicate URLs are prevented automatically.

PDF documents

Upload price lists, manuals or policy documents — the content is extracted and indexed like a page. Document limits depend on your plan.

Q&A pairs

Hand-written question → answer pairs. The highest-precision source: when a visitor asks something close to the question, the bot answers with your exact text. Two places create them for you:

• Unanswered tab — questions the bot couldn't answer, one-click convertible into a Q&A.
• Training tab — review conversations that got a thumbs-down, correct the answer, and save the correction as knowledge.

Tips for better answers

• Prefer several focused pages over one giant page — the bot retrieves passages, and a passage that mixes topics dilutes every answer.
• After any big website update, hit Re-crawl All.
• If the bot must never guess, enable strict grounding in bot settings — off-script questions then always get your fallback message instead of an attempt.