Import and crawl
Bring existing content into the Knowledge Base from files or from your website.
Imported content always arrives as drafts, so nothing reaches customers or the AI until you review and publish it.
Import a file
Choose Import on the Knowledge Base page and pick a file (up to 5 MB).
| File | What happens |
|---|---|
| CSV | One draft per row. No AI is used, so it doesn't use your AI budget. |
| Word (.docx), Markdown, text, PDF, JSON | AI reads the file and splits it into articles. Headings and lists in a Word or Markdown file are kept. |
CSV format
Use a header row. Only a title and content are required.
| Column | Notes |
|---|---|
title (or question) | Required. |
content (or answer) | Required. Plain text becomes paragraphs. |
category | Optional. Defaults to Imported. |
type | Optional: article, faq or news. Rows with a question column default to FAQ. |
Up to 500 rows per file. Rows that can't be used (no title, or no content) are skipped and listed by their row number, so you can fix them and import again.
Crawl a website
Choose Crawl URL, enter the address of your help site, and Find pages. Select the pages you want (the first ten are ticked for you) and start the crawl. Kerabie reads each page and creates a draft article from it. Crawling runs in the background and uses your monthly AI budget and the page limit of your plan.
How imported articles are written
Imported and crawled text is rewritten into articles that are easy to read: a short answer first, then a heading for each section, steps as numbered lists, options as bullet lists and key terms in bold. Facts, numbers, names, prices and links are kept exactly as they are in the source, and nothing is invented. A long page becomes several focused articles instead of one wall of text.
Each page you select in a crawl is read in full on its own. If an article from an earlier import reads like plain text, open it and choose Refine with AI to restructure it the same way.