NexaDesk

Knowledge Base

Build and manage the content that powers your AI chatbot

The Knowledge Base is the foundation of your AI chatbot's intelligence. It stores all the content your chatbot uses to answer visitor questions — from website pages and documents to custom Q&A pairs.

Content Sources

Provide a URL and NexaDesk crawls the page to extract content. You can crawl a single page or follow links to index multiple pages from the same domain (up to 50 pages per crawl).

Upload PDF, DOCX, or TXT files. NexaDesk extracts the text content, splits it into semantic chunks, and indexes it for retrieval.

Supported formats:

  • PDF (up to 10MB per file)
  • DOCX (up to 5MB per file)
  • TXT / Markdown (up to 2MB per file)

Paste custom content directly. Each entry gets a title and body. Useful for content that does not exist elsewhere — internal policies, specific instructions, or curated answers.

Define explicit question-answer pairs for precise control. When a visitor asks something similar to your question, the chatbot returns your exact answer.

Managing Sources

The Knowledge Base dashboard shows all your training sources with:

  • Status — Active (trained), Processing, or Failed
  • Type — Website, Document, Text, or Q&A
  • Chunks — Number of content chunks generated
  • Last Trained — When the source was last processed

When your source content changes (e.g., you update your pricing page), click Re-train to refresh the indexed content. The chatbot will use the updated content immediately after processing completes.

Removing a source also removes all its associated chunks and embeddings. The chatbot will no longer reference this content.

Content Chunking

NexaDesk splits your content into overlapping chunks of approximately 500 tokens. Overlapping ensures that context is not lost at chunk boundaries. You can view the individual chunks for any source to verify how your content was split.

Best Practices

  1. Quality over quantity — A few well-written pages produce better results than dozens of thin pages
  2. Update regularly — Re-train sources when your content changes
  3. Use Q&A pairs for precision — If the AI gets a specific question wrong, add a Q&A pair with the correct answer
  4. Check unanswered queries — Review the analytics to find gaps in your knowledge base
  5. Organize with titles — Give each source a clear, descriptive title for easy management
PreviousWorkflowsNextWidget Customization