An AI agent is only as useful as what it knows. If it knows your business, it saves you time. If it does not, it makes things up or sends every question to a person, and you are back where you started.
ATON Desk is live chat for websites, made by Atheron Network Labs. It is coming soon and not open to customers yet. This article explains how the AI agent inside it learns, so you know what to expect before you decide whether to join the waitlist.
Three sources, one agent
The agent is trained on three things you already have.
- Your website. The pages your customers can read are the first place the agent looks. Your pricing page, your FAQ, your product pages and your policies all count.
- Your files. Documents you upload give the agent what is not on the public site: internal notes, product details, process descriptions.
- Your answers. The replies you and your team send in chat become part of what the agent knows. Every good answer you give is a lesson.
You do not write a script. You do not build a decision tree. You point the agent at what you already have, and you keep answering customers the way you always did. The agent learns from the work you are already doing.
What the agent does with that knowledge
When a visitor opens chat, the agent replies first. It says it is an AI. After its first answer, it offers a person. That is a fixed part of how it works, not a setting you have to remember to turn on.
This order matters. Many questions are simple and repeat all day: opening hours, shipping times, whether a plan includes a feature. The agent handles those from your site and files. The visitor gets an answer in seconds and never needs to wait for you.
When the visitor wants a person, or when the question is beyond what the agent knows, an operator takes over in one click. The operator gets a short summary of what has been said so far, so nobody has to ask the customer to start again.
Gaps become draft articles
Here is the part that makes the agent better over time.
When a question comes in that the agent cannot answer, ATON Desk turns it into a draft article. The draft waits for you to approve it. You read it, correct it if needed, and publish it. From then on, the agent knows the answer.
Every unanswered question is a sign that something is missing from your knowledge. The draft is the shortest path from that gap to a fix.
This replaces the usual way of building a help centre, where someone sits down once a quarter and tries to guess what customers want to know. Instead, the questions themselves tell you what to write, and most of the writing is already started for you.
Nothing goes live without a person approving it. The agent proposes. You decide.
What this does not do
It is worth being plain about limits.
The agent cannot know what you have not given it. If your refund policy exists only in your head, the agent cannot explain it until you write it down, either in a file, on your site, or in a chat answer that it learns from.
The agent also cannot replace judgement. A customer with an unusual problem, an upset tone or a request that needs a decision should reach a person. That is why the offer of a person comes after the first answer, every time.
How we use it ourselves
Atheron's own support runs on ATON Desk. The same agent, the same handover to a person, the same draft articles for the questions it cannot answer yet. We are learning from our own gaps the same way you will.
When it opens
ATON Desk is not available yet. When it opens, prices will be in Canadian dollars per seat per month. Every plan will include AI conversations. There will be a free plan and a trial that does not ask for a card.
If you would like to be told when it is ready, you can join the waitlist at aton-desk.com.
