Building Your Knowledge Base

The knowledge base is how your agent knows what to say. Without it, responses are generic. With it, your agent answers specific questions about your products, policies, pricing, and processes.

Website crawling

The fastest way to build your knowledge base:

1. Go to Dashboard > Knowledge Base > Website tab
2. Enter your website URL
3. Click Add Site to start the crawl

The crawler will:

• Discover and map your site structure
• Rank pages by relevance
• Extract content from each page
• Pick up your brand voice and terminology

Crawl status

Watch the progress as your site is processed:

• Queued - Waiting to start
• Analyzing - Discovering pages
• Ranking - Prioritizing content
• Scraping - Extracting text
• Pages Ready - Content available
• Completed - All done

Advanced options

Click Advanced Options to customize the crawl:

• Exclude Branches - Skip certain URL paths (e.g., /admin, /login)
• URL Filter - Include only or exclude URLs matching a pattern (e.g., /blog, /help)
• Max Depth - How many levels deep to crawl (1-5, default is 2)

Selecting pages

After crawling, you’ll see a tree view of discovered pages. Check the boxes next to pages you want to include, then click Save to add them to your knowledge base.

Re-crawling

If a crawl fails or your content changes, click Recrawl to run it again.

File uploads

For content that isn’t on your website:

1. Go to Dashboard > Knowledge Base > Other Files tab
2. Upload your documents
3. The system extracts text and adds it to searchable knowledge

Good candidates: SOPs, internal policies, product specs, training materials.

FAQ answers from customer questions

As your agent handles real email, Know Reply clusters similar questions from customer replies into FAQs automatically. You can review these in Insights > Knowledge Base FAQ.

The report shows:

• Common questions grouped by topic
• How confidently your AI is answering each one
• Which questions have gaps in your knowledge base

When you spot a question the AI isn’t handling well, you can add your own answer directly from the report. That answer gets saved to your knowledge base and used for future replies.

How retrieval works

When an email arrives, the system finds the most relevant content—not just keyword matches, but meaning matches. A customer asking “how do I return something?” finds your “refund and exchange policy” even if the exact words don’t match.

Embedding strategy

Each piece of content you add uses knowledge embeddings — your plan determines how many you have. Think of embeddings as your AI’s total knowledge capacity.

What to embed first

Start with content that directly answers customer questions:

1. Your website — crawl your main site for product info, pricing, and policies
2. FAQ content — add the questions your support team answers repeatedly
3. Key policies — returns, shipping, cancellations, warranties

When to use file uploads

Upload internal materials that aren’t on your website:

• SOPs and process guides — how your team handles common situations
• Product specs and data sheets — detailed technical information
• Training materials — onboarding docs that capture institutional knowledge
• Internal policies — HR, compliance, or operational guidelines

Supported formats: PDF, DOCX, TXT, Markdown, and images (JPEG, PNG with text extraction).

Organizing with document groups

Group related content together and control which agents access which groups. A sales agent might need product specs and pricing, while a support agent needs troubleshooting guides and return policies.

Scaling your knowledge

As your business grows, you’ll need more embeddings to cover new products, services, and topics:

Plan	Embeddings	Good for
Free	20	Testing with a few key pages
Starter	100	A small business website and core docs
Pro	1,000	Multiple product lines, detailed documentation
Premium	5,000	Large catalogs, extensive knowledge bases
Enterprise	Unlimited	Enterprise-scale content libraries

Tips

• Quality over quantity - 50 well-written answers beat 500 scraped pages of boilerplate
• Check the FAQ report - go to Insights > Knowledge Base FAQ to see which questions your AI struggles with, then add answers directly
• Use agent-level access control - don’t give every agent access to everything. A focused agent with relevant knowledge outperforms one drowning in irrelevant content

Keeping it current

Re-crawl your website after major content changes. Upload new documents as they’re created. Your agent always pulls from the latest content.