Should your AI agent use a knowledge base file or live data?

Most AI chatbots and voice agents start the same way. Someone exports the website, uploads it as a PDF, and the agent answers from that file. It works well on day one. By day ten the agent is quoting an old price, offering a product that sold out, or telling a caller their order hasn't shipped when it has.

The short answer

Split what your agent knows into two groups.

Things that rarely change go in the knowledge base: returns policy, opening hours, service descriptions, how to book, what's included in each package.

Things that change daily come from live data: stock levels, prices, order and delivery status, appointment availability, and anything a customer can change themselves.

If a wrong answer would cost you a sale or an angry customer, it belongs in the second group.

Why uploaded files fail for e-commerce and bookings

A knowledge base is a snapshot. Your store is not. Every time you add a product, change a price or run a promotion, the file is out of date until someone remembers to upload a new one. Nobody remembers.

Voice agents make this worse. On a chat window a customer can click through to check. On a phone call, they trust what the agent says.

How live data works in practice

The agent doesn't need to read your whole store. It needs a small set of actions it can call during the conversation, for example:

1. Look up a product by name and return price, stock and a link.
2. Look up an order by order number or email and return status and tracking.
3. Check availability for a date and return open times.

Each action is a short workflow. I usually build them in n8n: the AI agent (in Vapi for voice, or an n8n AI agent for chat) calls a webhook, n8n queries Shopify, WooCommerce, your booking tool or your CRM, and sends back only the fields the agent needs. The agent then answers in plain language.

This keeps answers current without retraining anything, and you can see every lookup in n8n's execution log.

What to keep in the knowledge base

Live data doesn't replace the knowledge base. It still holds:

• Your tone of voice and how the agent should introduce itself
• Policies: returns, refunds, warranties, privacy
• Service and product descriptions that don't change often
• When to hand over to a human, and how

A simple checklist before you launch

• List the ten questions customers ask most. Mark each one "rarely changes" or "changes daily".
• Build a live lookup for every "changes daily" question.
• Give the agent a clear handoff for anything it can't answer, to email, WhatsApp, a helpdesk like Zendesk or Gorgias, or a live call.
• Test with real order numbers and products that are out of stock.
• Review a week of conversations and add the questions you missed.

Frequently asked questions

Can an AI agent read my Shopify store in real time?

Yes. The agent calls a workflow (for example in n8n) that queries the Shopify API for the product or order the customer asked about, then answers using that result. It reads only what it needs, when it needs it.

Is live data slower on a phone call?

A well-built lookup returns in a second or two, which sounds natural on a call. Keep each workflow small and return only the fields the agent will say out loud.

Do I still need a knowledge base if I use live data?

Yes. The knowledge base gives the agent your policies, tone and background. Live data covers prices, stock, orders and availability. Most good agents use both.

Which tools do I need?

An AI agent platform (Vapi for voice, or a chat agent built in n8n or your helpdesk), an automation tool like n8n to run the lookups, and API access to your store, booking system or CRM.