Add Knowledge Sources
The Knowledge Base is what makes your agent useful. Without it, the agent can only reply using generic AI knowledge — with it, the agent answers specifically from your products, policies, and content. The more accurate and complete your knowledge base, the better every conversation.
How the agent uses your knowledge
AIChatVault uses Retrieval-Augmented Generation (RAG). When a visitor asks a question:
Q&A pairs are checked first
Semantic search runs across all sources
A grounded prompt is built
The AI generates a grounded answer
Source types at a glance
| Source | Best used for | Updates |
|---|---|---|
| Files (PDF, DOCX, etc.) | Product manuals, policy documents, price lists, SOPs | Manual re-upload |
| Websites & URLs | Help centres, product pages, blog posts, documentation | Manual or scheduled auto-sync |
| Manual Q&A | FAQs, pricing, return policies, definitive one-answer questions | Edit any time, instant effect |
| Notion | Internal wikis, onboarding docs, knowledge bases already in Notion | Manual or scheduled sync |
Uploading files
Go to Knowledge Base → Files. You can upload one or many files at a time by dragging them in or clicking to browse.
Supported formats and limits
| Format | Size limit | Notes |
|---|---|---|
| 50 MB | Text-based PDFs process best. Scanned/image-only PDFs have lower accuracy — use a PDF with selectable text where possible. | |
| DOCX / DOC | 50 MB | Microsoft Word. All body text, tables, and headers are indexed. |
| TXT | 10 MB | Plain text. Useful for exporting content from other tools. |
| CSV | 10 MB | Each row becomes a searchable chunk. Column headers are included for context. |
| XLSX / XLS | 20 MB | All sheets are indexed. Formulas are not evaluated — only their display values. |
What happens when you upload
Upload
Processing
Ready
Tips for better file quality
- •Use text-based PDFs, not scanned images. If you only have a scan, run it through OCR software first.
- •Structure documents with clear headings — they help the chunking algorithm group content logically.
- •Remove duplicate or outdated content from files before uploading to avoid conflicting answers.
- •For large files (>10 MB), consider splitting into topic-specific documents — targeted retrieval is more accurate than retrieving from one massive file.
- •CSV files work well for product catalogues — ensure each row has enough context (e.g. include the product name in every row, not just the header).
Adding websites and URLs
Go to Knowledge Base → Websites & URLs. The crawler visits the URL and indexes the readable text content from each page.
Enter the starting URL
https://help.yoursite.com).Choose the crawl scope
- •Single page — Only the URL you entered is indexed. Use this for a specific policy page or FAQ.
- •All pages on domain — The crawler follows internal links across the entire domain. Use this for a complete help centre or product site.
- •Custom depth — The crawler follows links up to N levels deep from the starting URL. Useful for crawling one section of a large site.
Add the URL
Enable auto-resync (optional)
What the crawler can and cannot access
| Can crawl | Cannot crawl |
|---|---|
| Public HTML pages | Pages behind login / authentication |
| Server-rendered content | JavaScript-only single-page apps where content loads asynchronously |
| Linked PDF files | Pages blocked by robots.txt |
| Sitemap.xml URLs | CAPTCHA-protected pages |
| Sub-pages linked from the starting URL | Content inside iframes from external domains |
Manual Q&A pairs
Q&A pairs are the most precise knowledge source. Unlike RAG retrieval which finds the best-matching passage from a document, Q&A pairs match on the question itself — so the answer is returned with near-perfect reliability when a visitor asks something similar.
When to use Q&A pairs
- •Your most frequently asked questions — the top 20 questions you get from customers.
- •Pricing information that needs to be 100% accurate.
- •Policies (return policy, shipping times, refund procedure) where paraphrasing is risky.
- •Questions where the document-based answer might be buried or retrieved inconsistently.
- •Any question where the agent has previously given a wrong or incomplete answer.
Adding Q&A pairs
Go to Knowledge Base → Q&A
Click Add Q&A
Write the question naturally
Write a complete, accurate answer
Save
Notion sync
If your business already maintains documentation in Notion, you can sync those pages directly into your agent's knowledge base instead of exporting and re-uploading them.
Go to Knowledge Base → Notion
Click Connect Notion
Select pages to sync
Click Sync Now
Enable auto-sync (optional)
Managing your knowledge base
Checking what's indexed
Each source tab shows all uploaded files, URLs, and Q&A pairs with their status (Ready, Processing, Error). You can see exactly what the agent has access to at any time.
Removing outdated content
Delete sources that are no longer accurate. Stale information is worse than no information — if the agent retrieves an old price or discontinued policy, it will give wrong answers. Delete first, then upload the updated version.
How many sources should I add?
- •There is no strict limit on the number of sources, but quality matters more than quantity.
- •Start with your top 10–20 Q&A pairs and your most important document or page.
- •Add sources gradually and test after each addition.
- •Too many low-quality or duplicate sources can dilute retrieval quality — the model may pull an irrelevant chunk from an unrelated document.
Was this page helpful?