Skip to content
Technology: RAG (Retrieval-Augmented Generation)

Train the AI on
your business content

The knowledge base is how the optional AI receptionist answers factual questions from your own material instead of improvising. Paste a URL, upload a PDF, or add a short snippet such as hours, service area, or what you do not take on. It is part of the AI add-on, not the core plan or the free trial.

What is RAG?

Retrieval-Augmented Generation (RAG) is the process of giving an AI a "Library" to look at before it speaks.

Instead of the AI guessing your prices, it searches your Knowledge Base in real time and answers from the closest matching material you provided, or says it does not know when nothing is close enough.

01

Grounded Answers

The AI is restricted to ONLY answer from the data you provide. No made-up prices.

02

Real-Time Context

URL sources are re-read automatically, so edits on your site can reach the AI without a manual re-upload.

How to Build The Brain.

Follow these steps to train your AI in under 5 minutes.

๐ŸŒ

Website URLs

Simply paste your URL (e.g., your Pricing or FAQ page). Our system reads the text and adds it to your knowledge base instantly.

Setup: Paste & Scrape
๐Ÿ“„

PDF & Documents

Upload employee handbooks, technical manuals, or inventory lists. The AI processes the file and remembers every detail.

Setup: File Upload
โœ๏ธ

Manual Snippets

Have specific instructions like "If they ask for a manager, say X"? Paste them as direct text snippets to override any other data.

Setup: Instant Text
AI Manager

Global vs. Bot-Specific.

Our platform allows for Knowledge Isolation. You can add "Global" knowledge that every bot knows, or assign specific PDFs to specific bots.

Example Workflow:

"Upload a general company FAQ as GLOBAL, but give the 'Sales Bot' a specific pricing PDF that the 'Support Bot' can't see."

What to expect once it is switched on

What happens when it does not know

This is the question worth asking of any AI answering your phone, and the honest answer matters more than the clever one. When nothing in your knowledge base is a close enough match, the receptionist says it does not have that information rather than inventing something plausible.

That is a deliberate trade. An AI that guesses sounds more capable in a demo and quotes the wrong price to a real customer. A relevance threshold sits between the caller's question and the answer, and anything below it is treated as "not covered" instead of "close enough". You will occasionally hear the AI decline something it could have half-answered. That is the system working.

What is actually worth uploading

The instinct is to point it at your whole website. Resist that. The pages that make a receptionist useful are the ones a customer asks about on the phone: pricing, opening hours, service areas, what you do and do not take on, parking, appointment policy, how long a typical job takes.

Your About page and blog archive rarely answer a phone question. Adding them does not make the AI smarter; it gives the retrieval step more hay to search for the same needle. A tight knowledge base of genuinely useful pages beats a large one every time.

Keeping it current

Any URL you add is re-read automatically, so a price change on your own site reaches the receptionist without you touching anything. Pasted text and uploaded files stay exactly as you wrote them until you edit them, which is usually what you want for policies and scripts you have worded carefully.

The failure worth guarding against is the quiet one: changing a price on your website but leaving an old figure in a pasted note. The AI has no way to know which is right. When you update a price, search your knowledge base for the old number.

Does it slow the call down?

Barely, and only when it is used. The receptionist does not search your knowledge base on every sentence. It looks something up when a caller asks something factual, and skips it for greetings and small talk, so the pauses land where a human would naturally pause anyway.

Global or per-bot

Anything added globally is available to every bot on the account, which suits company-wide facts like hours and pricing. Anything attached to a single bot stays with it, which is how one number can answer as a sales line and another as support without the two bleeding into each other. Most accounts end up with a small global set and a handful of per-bot additions.

Deploy Your Brain Today.

Start training your AI and stop answering the same questions twice.

Open Knowledge Manager

Related AI features