Knowledge Base

Give your AI the information it needs to answer accurately. Add your FAQs, product details, and policies — enuchat handles the rest.

How It Works

The knowledge base is the foundation of accurate AI responses. Here's the flow:

The pipeline

  1. You add text content — FAQs, policies, product information, anything your visitors might ask about
  2. enuchat chunks the content into smaller, meaningful segments
  3. Each chunk is converted into a vector embedding using OpenAI's text-embedding-3-small model
  4. Embeddings are stored in pgvector (PostgreSQL vector extension) for fast similarity search
  5. When a visitor asks a question, their question is also converted to an embedding
  6. The most relevant chunks are retrieved and included in the AI's prompt as context

This approach — called Retrieval-Augmented Generation (RAG) — means the AI answers from your actual content, not from its general training data.

Adding Sources

Sources are the individual pieces of content that make up your knowledge base. Each source has a title and text content.

Steps

  1. Open your widget and go to the Knowledge Base tab
  2. Click Add Source
  3. Enter a descriptive title (e.g. "Return Policy", "Pricing Plans", "Business Hours")
  4. Paste or type your content as plain text
  5. Click Save

After saving, the source status will show "Processing..." while enuchat chunks the content and generates embeddings. This typically takes a few seconds. Once complete, the status changes to "Ready" and the content is immediately available for AI responses.

Writing Good Content

The quality of your AI's answers depends directly on the quality of your knowledge base content.

Do

  • Write naturally — use the same language your customers use, not keyword-stuffed SEO text
  • Be specific — include actual prices, hours, addresses, and policy details
  • Answer real questions — look at your support inbox for the questions people actually ask
  • Organize by topic — one source per topic keeps things clean and easy to update
  • Keep it concise — 2-3 sentences per answer is ideal for retrieval accuracy

Don't

  • Don't dump your entire website into one source — it dilutes relevance
  • Don't use marketing fluff — the AI needs facts, not slogans
  • Don't leave out key details — if your return policy is 30 days, say "30 days", not "generous return window"

Example: good source content

Title: Shipping Policy

We ship to all EU countries. Standard shipping takes 3-5 business days
and costs €4.99. Orders over €50 qualify for free shipping.

Express shipping (1-2 business days) is available for €9.99.
We ship Monday through Friday. Orders placed before 2pm CET ship
the same day.

How Semantic Search Works

Unlike keyword search, semantic search understands meaning. A visitor asking "How much does delivery cost?" will match a source about "Shipping fees" even though the words are different.

The retrieval process

  1. The visitor's question is converted to a vector embedding (same model as your content)
  2. This embedding is compared against all your content embeddings using cosine similarity
  3. The top 5 most relevant chunks are selected
  4. These chunks are included in the AI prompt as context for generating the response
Similarity threshold: If no chunk is relevant enough (similarity score too low), the AI won't guess or make things up. Instead, it will tell the visitor "I don't have information about that" and offer to connect them with a human operator.

Grounding

enuchat's AI is grounded — it's instructed to answer only from the knowledge base content provided in its context. This is a critical safety feature.

What grounding means in practice

  • The AI will not make up information, prices, or policies
  • The AI will not answer from its general training data — only from your content
  • If the answer isn't in the knowledge base, the AI acknowledges this honestly
  • When the AI can't help, it offers to hand off to a human operator

This means your visitors get accurate, company-specific answers — or they get connected to your team. No hallucinated information, no wrong prices, no made-up policies.

Updating Content

Your knowledge base should evolve with your business. Updating is straightforward.

How updates work

  • Edit any source's title or content and save
  • The content is automatically re-chunked and re-embedded
  • Changes take effect within seconds — no manual reprocessing needed
  • Old embeddings are replaced, so the AI immediately uses the updated content

You can also delete sources you no longer need. Deleted content is removed from the vector store immediately.

Tips

Getting started

  • Start with 10-20 entries covering your most frequently asked questions
  • Check your email inbox, support tickets, and social media for real customer questions
  • Don't try to add everything at once — start small and expand based on what visitors actually ask

Ongoing maintenance

  • Review monthly — look at conversations where the AI couldn't answer and add that content
  • Update prices, hours, and policies whenever they change
  • Remove outdated information (discontinued products, expired promotions)

What not to add

  • Don't add your entire website — it reduces retrieval accuracy
  • Don't add content that changes daily (e.g. stock levels) — use the API for dynamic data instead
  • Keep entries focused — one topic per source

Ready to build your knowledge base?

Create your free account and start adding content in minutes.

Get Started Free