From a pasted address to a live assistant
Four steps. We do three, you do one — and yours is the one that carries the responsibility. None of them needs a developer.
You give us your website address
Sign up, create a project, enter your website address. You can also upload documents: price lists, terms, product sheets. You tell us the assistant's role — support, sales or documentation — because that determines what it looks for in your content.
We crawl it and build knowledge
We read your pages, filter out the boilerplate and build question-and-answer pairs. Price lists become tables. You watch it progress, and you see what became of each page.
You review and approve
This step is yours, and it cannot be skipped. You get an analysis report: what was produced, where it found contradictions, what is uncertain or time-sensitive. Where two of your pages disagree, you decide which one is right. What you rewrite is not overwritten by a later re-crawl.
You embed it on your site
You get one line of code. Paste it into your site template and it is live. No installation, no plugin, no server-side work. The widget can only be called from the domain you registered.
What it looks like inside
Three screens that make up the daily work: deciding contradictions, the return report, and what the assistant could not answer.
Review queue
The machine puts contradicting pages side by side. You decide which one is right — or write the correct answer yourself. What you write is not overwritten by a re-crawl.
Return report
How many questions it handled, how many it could not, how many leads stayed with you. Broken down by week, in numbers — not in feelings.
Missing knowledge
What people asked but it could not answer — grouped, by frequency. The top row is the most expensive gap: that is what most people ask.
The three images above are DIAGRAMS, not screenshots: they show the structure, without customer data. You see the real interface in the free trial.
And then?
It refreshes itself weekly
We re-read your site. If a page has not changed, reprocessing it does not consume your processing allowance — only genuinely changed pages go back to the model. Your monthly fee is unaffected.
You learn what it could not answer
Unanswered questions are collected by frequency. You write the answer, and from then on it knows.
The lead stays with you
Where the bot could not help, the visitor can leave their details. That becomes your work queue.
What your developer will want to know
- The embed code is a single <script> tag loaded asynchronously. It does not block page rendering.
- The widget uses no cookies and stores no messages in the browser. It writes exactly one entry, bound to the tab: the identifier of the conversation in progress, for at most 30 minutes — so the conversation does not restart after a page change.
- Calls are only allowed from the domain registered for the project (origin check), and rate limiting guards against abuse.
- Your knowledge base, the conversations and the leads stay on our own servers in the European Union. We call a language model to produce answers; the details are listed in the privacy notice.
See it on your own site
Give us the address, run the crawl, and see what it could answer from your content. If it does not convince you, you have paid nothing.