Chatbots that answer from your own content

What businesses use it for

• The five questions that arrive every day on the website, answered at three in the morning too.
• A first reply in support, so people know they have been heard before you get to them.
• An internal search for your own team, over manuals, price lists and procedures.
• Pre-qualifying enquiries, so the ones that reach you are the ones worth your time.

Why it does not make things up

A plain language model answers from what it happened to read on the internet. That is exactly where invented answers come from. This one works differently: before it replies, it searches your documents, and it is only allowed to answer from what it found. If nothing fits, it says so and hands over to a human. Every answer can carry the source it came from, so you can check it.

How it runs

1. 1 We collect what he is allowed to know: website, price list, procedures, past support replies.
2. 2 We define the boundaries: what he answers himself, what he passes on, what he must never say.
3. 3 You throw your nastiest test questions at him and we correct until the answers hold.
4. 4 He goes live, and once a month you see what people actually asked. That list is worth money on its own.

Where he can live

On your website, in your existing chat tool, on WhatsApp or Telegram, or purely internally for your own team. If the content is sensitive, the whole thing runs on your own server without any outbound connection. That costs more hardware and a bit of quality, and for many internal use cases it is still the right call.

What it costs

Driven by three things: how much material has to be prepared, how many systems he needs to reach into, and whether he runs in the cloud or on your hardware. A chatbot that answers from a website and a price list is a small project. One that reads your stock levels and books appointments is a real one. You get a fixed price for a defined scope before anything starts.