Knowledge Base

The knowledge base is how your AI assistant learns about your product. Feed it your website content, documents, or raw text — and it becomes an expert on your business.

How It Works

When a user asks a question, the SDK sends the message to the Produck backend which performs vector search against your knowledge base to find relevant content. This content is injected into the AI's context, enabling accurate, grounded answers.

User Question → Vector Search → Relevant Chunks → AI Response

Adding Sources

Via Dashboard

1. Go to SDK Projects → select your project
2. Navigate to the Knowledge Base tab
3. Click "Add Source"

Source Types

Website Crawling

When you provide a URL, Produck uses Firecrawl (opens in a new tab) to:

1. Fetch and render the page (including JS-rendered content)
2. Extract clean text content
3. Split into chunks (~500 tokens each)
4. Generate vector embeddings
5. Store in Qdrant vector database

URL → Firecrawl → Clean Text → Chunking → Embeddings → Qdrant

💡 Tip: Add your most important pages first — pricing, features, FAQ, and docs.

Retrieval

When the AI receives a user message, it:

1. Generates an embedding of the user's question
2. Searches Qdrant for the top-k most similar chunks
3. Includes those chunks as context in the AI prompt
4. Generates a response grounded in your actual content

This means the AI will never hallucinate about your product — it only uses information from your knowledge base.

Relevance Scoring

Each retrieved chunk has a similarity score (0–1). Only chunks above the relevance threshold are used. This prevents irrelevant content from polluting the AI's context.

Best Practices

What to Include

• ✅ Pricing pages and plan details
• ✅ Feature descriptions
• ✅ FAQ and help center articles
• ✅ API documentation
• ✅ Getting started guides
• ✅ Terms of service, refund policies

What to Avoid

• ❌ Extremely long pages (split into sections)
• ❌ Pages behind authentication (the crawler can't access them)
• ❌ Frequently changing data (use API schema for real-time data)

Content Quality

The AI is only as good as the content you feed it. Ensure your knowledge base sources are:

• Accurate — up to date with current product info
• Clear — well-written, unambiguous text
• Complete — cover the topics users ask about most

Limits by Plan

Next Steps

• API Schema — Let the AI execute actions on your behalf
• DOM Context — Let the AI read the current page
• Chat Widget — Customize the chat interface