Training sources: websites, documents, and Q&A

Crawl a website

Start with public pages that are accurate, current, and intended for customers. The API accepts a start_url and an optional max_pages value. Crawl a focused section rather than an entire domain when unrelated pages, navigation, or legacy content could confuse answers. Authentication walls, robots rules, and site structure can affect what is reachable.

After adding a source, wait until its status is ready and review any processing error before relying on it. Revisit the source after changing important policies or product details.

Prepare PDF, DOCX, and CSV files

Use the upload control in your account to confirm whether PDF, DOCX, or CSV is accepted in your environment and to see any current file-size or row limits. The project frontend identifies a generic document source, but does not publish a definitive file-extension allowlist, maximum file size, or extraction guarantee. Do not promise a format or quota until the current upload flow confirms it.

For PDF or DOCX, prefer text-based files with clear headings and selectable text; scanned images may require OCR that is not documented here. For CSV, use a header row and one complete record per row. Remove customer data, internal notes, and columns the chatbot should not disclose.

Write concise Q&A pairs

Add one clear question and its approved answer per pair. Include natural variants when useful, but keep answers complete, consistent, and safe to show to visitors. The current frontend does not publish a maximum number of Q&A pairs or character limit; check the account's current validation messages rather than assuming a quota.

Start a focused website crawl

curl --request POST "$PRIMEWEBKIT_API_URL/bots/$BOT_ID/knowledge-sources/website" \
  --header "Accept: application/json" \
  --header "Content-Type: application/json" \
  --header "Authorization: Bearer $PRIMEWEBKIT_ACCESS_TOKEN" \
  --data '{"start_url":"https://example.com/help/","max_pages":25}'

Example CSV structure for Q&A content

question,answer
"How do I contact support?","Email support@example.com during the hours listed on our contact page."
"Where can I find the return policy?","Read the current return policy at https://example.com/returns."

Add one Q&A source through the API

curl --request POST "$PRIMEWEBKIT_API_URL/bots/$BOT_ID/knowledge-sources/qa" \
  --header "Accept: application/json" \
  --header "Content-Type: application/json" \
  --header "Authorization: Bearer $PRIMEWEBKIT_ACCESS_TOKEN" \
  --data '{"question":"What are your support hours?","answer":"Our current support hours are listed on the contact page."}'

Frequently asked questions