Everything you need in order to actually trust it
From getting the knowledge in to seeing what it brought back. Every plan includes every feature — no modules, no add-ons.
So it knows your business
Website crawling
You do not write a knowledge base: you hand over the address and we build it from the pages you already have. Header, footer and cookie bar are recognised as boilerplate and left out. The crawl never leaves your own domain.
Document upload
Whatever is not on your site, you upload: PDF, Word, Excel, plain text. A price list becomes a table, not a wall of text — that is how it quotes an actual price instead of "please request a quote".
Automatic re-crawl
Nobody has to tell it that a price changed. We re-read your site weekly, and only genuinely changed pages go back to the model: on a real 830-page knowledge base, a weekly round reprocessed 1 page instead of 830. Unchanged pages do not consume your allowance.
Review queue
When two of your pages give different delivery times, today it is your customer who finds out. Here you find out first: contradictions and duplicates are laid side by side, you decide which is true, and what you write is never overwritten by a later crawl.
So it never speaks out of turn
Source citation
Visitors do not have to take the bot's word for it: every answer shows the page it came from, checkable in one click. In our own measurement, 100% of those links resolved to live pages.
Honest refusal
The most important feature, and the least impressive to look at. With no coverage in the knowledge base, it does not invent: it says it does not know and offers a human. On the 12 unanswerable questions in our test set, it did this all 12 times.
The visitor's language, your material
Your foreign visitors do not have to switch languages — and they do not need to. The chat interface ships in six languages (Hungarian, English, German, French, Spanish, Polish), and the answer arrives in the language they asked in, drawn from your single-language knowledge base. Prices and contractual terms are quoted in the form the source uses, never converted.
So you stay in charge
Live chat — you answer
The visitor can ask for a colleague and gets one in the same window, in the same conversation — you take it over from the panel. A work queue shows who is waiting, who is with you and who is with someone else. While a human is answering, the conversation uses NO credit: live chat is part of the subscription, not a paid add-on and not a per-agent fee. You decide when it is offered and during what hours — outside those, the visitor does not wait in vain, they get the contact form.
Handover to a human
The visitor the bot could not help simply leaves today. Here they can leave contact details, and the lead lands in a work queue with an owner and a status — in the morning you open a task list, not a log.
Missing-knowledge list
Unanswered questions, grouped and ranked by frequency. The top of the list tells you which missing paragraph is costing you the most: write it once, and it knows from then on. This list is often worth more than the bot itself.
Conversation log
When a customer says "your bot told me X", you can read it back: what was asked, what was answered, from which source, and how long it took.
Return report
You do not decide by gut whether it paid off. How many questions it handled, how many it missed, how many leads stayed with you — broken down by week.
Roles and permissions
A colleague can work on the knowledge base without being able to burn your allowance: a reader cannot spend. Owner, editor, reader.
Multiple projects in one account
Two sites, two separate assistants, one account and one invoice. Knowledge never crosses over — one client's content cannot surface on another client's site.
Two-factor authentication
Your account holds customer data. App-based (TOTP) sign-in with recovery codes, so a leaked password is not enough on its own.
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.