You Changed Your Price List at 9am. When Does Your AI Voice Agent Know?

You put your prices up at 9am. By 9:05 a caller asks the new rate. Does your AI voice agent quote the old number or the new one?

A voice agent knowledge base update reaches callers faster than most people expect. The trick is knowing how the agent stores what it knows, and how to keep it current.

Most people assume an AI agent learns things the way a person does. So they think a change means re-teaching it. That model is wrong, and the wrong model gets expensive.

New to this? Our plain English guide to what an AI voice agent is covers the basics first.

How does an AI voice agent actually store what it knows?

A good AI voice agent does not memorise your price list. It reads the list the moment a caller asks. Picture it glancing at a printed sheet on the desk.

That approach has a name: retrieval augmented generation, or RAG.

Your hours, prices, policies and FAQs live in a knowledge base. A caller asks a question. The agent grabs the right passage and answers from it.

The knowledge does not live inside the model. The agent looks it up fresh.

This changes the whole refresh question. You are not retraining a brain. You are updating a document the agent reads from.

We cover the mechanics in our explainer on how voice agents retrieve company knowledge.

How fast does a voice agent knowledge base update reach callers?

Edit a knowledge document and a well built agent uses the new version on the very next call.

The reason is simple. The agent does not keep a fixed copy of your prices in its head. It looks the document up the moment a caller asks, so as soon as you save a change, the next lookup returns the new text.

You change the price, you save it, and the next caller hears the updated pricing. The model itself never changes. Only the document it reads from does.

This is what the industry calls retrieval augmented generation. If you want the technical version, Google Cloud's primer explains it clearly.

Does the agent keep your website up to date by itself?

Yes. Add your website as a source, and the agent re-crawls it automatically on a regular schedule. Change a price or a policy page, and it picks the change up without you re-uploading anything.

So you have two clean ways to keep the agent current. Edit a document in the knowledge base for an instant change. Or let the site crawl carry the slower, standing updates.

Most businesses use both. The crawl keeps the broad picture fresh, and a direct edit handles the price you changed this morning.

What files should you put in the knowledge base?

You can load many files into one knowledge base. Markdown, plain text, Word, CSV, spreadsheets, HTML and PDFs all work, up to 50MB each.

Markdown and plain text beat a heavy PDF every time. Retrieval reads words, not pictures, so a clean text file gets quoted accurately.

A glossy brochure PDF full of images often does not. The text is trapped in the design, and the agent cannot read it cleanly.

Split your knowledge across more than one knowledge base on the same agent. Products in one, policies in another, FAQs in a third. The agent draws on all of them.

Plain files, well named, beat one fat document. See how the agent then acts on what it finds in functions and extracted variables.

Can you push new products to the agent the moment they launch?

Yes. You do not wait for anyone. Our APIs let you push a new file or URL straight into the knowledge base as the product goes live.

Release a new price list, a new spec sheet, a new policy. A single API call adds it, and the agent can answer about it on the next call.

That matters most when your range changes often. The agent stays current with your catalogue, without anyone re-teaching it by hand.

How do you change what the agent does, not just what it knows?

Changing behaviour is a different lever from changing knowledge. Say you want the agent to ask a new qualifying question. Or take a different booking path.

Or change its tone. That lives in the agent's configuration, not the knowledge base.

A behaviour change takes a redeploy, so it is not quite next call the way a document edit is. It is the difference between updating a fact and rewriting part of the script.

This is why testing matters more for your behaviour changes than for a price edit. A new conversation path can bite you in ways a price number never will.

We run every change through structured checks before it reaches your callers. Our guide to batch testing voice agents walks through it.

The agents we run also tighten themselves up over time, separate from any edit you make. We explain that in how our voice agents get smarter every night.

Frequently Asked Questions

How quickly does an AI voice agent update when I change the knowledge base?

Your document edit lands on the next call. The agent reads the knowledge live, not from memory, so a price, an hours change or a new policy is current straight away.

Do I need to retrain the AI when my information changes?

No. Your retrieval based agent reads from a knowledge base at call time. A fact change needs an edit, not a retrain.

Fine tuning is only for changing how the model behaves or sounds, never for adding fresh facts.

What file formats work best in a voice agent knowledge base?

Markdown and plain text work best, because retrieval reads words, not pictures. Word, CSV, spreadsheets, HTML and PDFs also load, up to 50MB each.

Avoid PDFs packed with images, where the text is trapped in the design. A clean text file gets quoted far more accurately.

Can the agent read my website automatically?

Yes. Add your site as a source and the agent re-crawls it on a regular schedule. It picks up your changes without you re-uploading anything.

Can I update the knowledge base by API when I launch a product?

Yes. Our APIs push a new file or URL straight into the knowledge base. Release the product, make the call, and the agent can answer about it on the next call.

What is the difference between updating knowledge and changing behaviour?

Knowledge is what the agent knows: prices, hours, policies, FAQs. You update it by editing a document or letting the site crawl.

Behaviour is what the agent does: the questions it asks, the path it follows, its tone. You change that by configuration and a redeploy.

Keeping an agent current is not a technical chore. It is whether your callers hear the truth, on the call that counts.

Want an agent that updates on the next call, crawls your site, and takes new products by API? Book a setup conversation and we will set it up with you.