Skip to content
Elooo SystemHelp Center Go to website

Knowledge bases

How to train the AI on your website and PDF or Word files (and update them later)

In AI Agents › Knowledge Base, open the base and click Add Source. Choose Web Crawler, paste the address, pick the scope (that page only, all URLs with the path, or the whole domain), click Extract Data, select the pages and click Train. Or choose File Upload and add DOC, DOCX or PDF files.

Updated on

Before you start

  • You need permission to manage knowledge bases.
  • Have the website address (public, no login) or your files as DOC, DOCX or PDF with selectable text (a scanned, image-only PDF won't work).
  • If you don't have a knowledge base yet, create one first (see how to create a knowledge base).
Where to find it

AI Agents › Knowledge Base › (open the base) › Add Source

Step by step

Add your website

  1. Open AI Agents › Knowledge Base and open the base (or create a new one).
  2. Click Add Source › Web Crawler.
  3. Pick the scope:
    • Exact URL: that page only;
    • All URLs with the Path: every page under that path (for example, everything under /services);
    • All URLs in this Domain: the whole site.
  4. Paste the address and click Extract Data.
  5. When it finishes, click View All Pages, tick Select all or just the pages you need.
  6. Click Train.

Add files

  1. In the same base, click Add Source › File Upload.
  2. Drag in or browse for your DOC, DOCX or PDF files and click Upload Files.
  3. Wait until the status changes to Processed. You can open the file with the arrow icon or delete it with the bin icon.
  4. To write or fix text directly, use Add Source › Rich Text, write it and click Save.

Update when the site or file changes

  1. For a file: delete the old source with the bin icon and upload the new version. Rich-text entries don't change uploaded files; they're separate sources.
  2. For the site: run the Web Crawler again on the same address and train on the updated pages.
  3. Ask the agent a few test questions to confirm the answer comes from the new source.

Tips

  • The crawler opens tabs, accordions and sections that load on scroll, but it only sees public content: anything behind a login is left out.
  • Clear headings in documents (H1, H2…) give better answers, because the AI organizes the text by section.
  • Images inside documents are skipped: turn anything that's only in an image (a pricing table as a picture, for example) into text.
  • Don't repeat the same content across sources: the AI may pick the wrong version.

Common problems

  • The file upload fails → Check it's DOC, DOCX or PDF and not password-protected or corrupted. Export it again and upload.
  • The file is stuck processing → Wait, refresh the sources tab and, if it doesn't move, upload a smaller or cleaner version.
  • The AI can't find the PDF content → The PDF is scanned (image-only). Convert it to selectable text or add the information as rich text.
  • The AI answers with outdated information → There's an old source. Delete it, upload the new one and test again.
  • The AI answers from the wrong source → There's duplicate or conflicting content. Remove the extra and improve the headings.

Frequently asked questions

How long until the AI uses the new content?

It's usually available immediately or within about a minute, depending on file size.

Can I upload a Google Doc?

Export it as DOCX or PDF and upload that file.

Was this helpful?