Vaaad AI
Vaaad AI
Talk to SalesExperience Vaaad
Platform

Knowledge Base & RAG

Upload your documents once. Your AI agents read them during calls to answer customer questions accurately, without making things up.

What is Knowledge Base (RAG)?

RAG stands for Retrieval-Augmented Generation. In plain English: your AI agent reads your documents before answering any question, so it never makes things up. It only tells customers what is actually in your docs.

Without RAG, LLMs hallucinate. They confidently invent fake pricing, wrong policies, or made-up addresses. With RAG, your agent always grounds its answers in your real documents.

Video walkthrough: Setting up Knowledge Base

What Should You Upload?

Any information your agent needs to reference during a call. Common examples:

  • Product catalogues. Features, specs, SKUs, availability.
  • Pricing sheets. Tiers, discounts, custom quotes.
  • FAQs. Common customer questions and your answers.
  • Policies. Refund, shipping, warranty, cancellation.
  • Company info. Addresses, hours, team members.
  • Service menus. Clinics, salons, restaurants.

Two Ways to Add Content

Vaaad supports both file uploads and web crawling from the Knowledge Base page.

1. File upload

Drag and drop or browse for a file. Supported formats:

  • .pdf for PDFs (text-based, scanned PDFs need OCR first)
  • .docx for Microsoft Word documents
  • .txt for plain text
  • .md for markdown

2. URL crawl

Paste any public URL and Vaaad will fetch the page, strip the HTML, and ingest the text content. Great for company About pages, FAQ pages, pricing pages, or terms and conditions you keep on your website.

The crawl has a 10-second timeout. Pages that need JavaScript to render their content (single-page apps) may come back empty. In that case, copy-paste the text into a .txt file and upload it instead.

Uploading a Document

  1. Go to the Knowledge Base page
  2. Click + Upload for a file or + Add URL for a webpage
  3. Pick the file or paste the URL
  4. Give it a descriptive name like “April 2026 Pricing”, not “untitled”
  5. Optionally add tags (pricing, policies, faq) and a description
  6. Click Upload. Vaaad parses, chunks, and embeds the content in 10 to 30 seconds. The doc shows up with a status badge: pending, then processing, then ready.

Linking Docs to Agents

Uploading isn't enough. You have to link docs to the agents that should use them. This lets different agents have different knowledge.

  1. Open your agent in the dashboard
  2. Scroll to the Knowledge Base section near the bottom
  3. Check the boxes next to the docs you want this agent to use
  4. Save. The agent retrieves from these docs on every call.
💡
Smart linking: A real estate agent shouldn't see a healthcare clinic's pricing doc. Link only what's relevant. Smaller scoped indexes mean faster, more accurate retrieval.

How Retrieval Works

During a call, here is what happens every time the user asks a question:

  1. User says: “What's your 2BHK price?”
  2. Vaaad embeds the question (semantic search)
  3. Searches your linked docs for the most relevant chunks
  4. Adds those chunks to the agent's context
  5. Agent answers using only the info from those chunks

If no relevant chunk is found (similarity score below the threshold), the agent says something like “I don't have that information, let me have our team follow up with you.” It does not make something up.

Writing Documents That Work Well

Not all docs retrieve equally well. Tips for maximum accuracy:

  • Use clear headings. “## Pricing” is better than “details”.
  • Keep related info together. Don't split a product's specs across pages.
  • Write for Q and A. “Q: What's included in the Pro plan? A: ...” works really well because the question text matches what callers actually say.
  • Use full sentences. Avoid tables with no surrounding text. Table cells in isolation lose meaning during chunking.
  • Avoid contradictions. If two docs say different prices, retrieval gets confused. Keep one source of truth per topic.
  • Repeat key facts. Counter-intuitive but useful: if “our address is X” appears in 3 different sections, retrieval is far more likely to surface it.

Updating Documents

When prices or policies change, delete the old doc and upload the new one. Embeddings are tied to the doc record, so an in-place replace is the cleanest pattern. Any agents linked to the old doc need to be re-linked to the new one. We'll add proper in-place editing in a future update.

Plan Limits

PlanDocuments
Free1 doc
ProUnlimited

Each document can include thousands of chunks. There's no fixed page or word limit per doc. Very large docs just take a bit longer to process on upload.

Next: Track campaign performance with Live Analytics →