How chatting with a website works
Paste a URL and we fetch the page the way a browser would, strip the navigation and boilerplate, and split what is left into passages of roughly 500 tokens. Each passage becomes a vector embedding stored next to the original text.
When you ask a question we embed the question too, retrieve the passages closest to it, and hand only those to the language model. That is retrieval-augmented generation, and it is why the answer quotes your page instead of inventing something plausible. If your site does not cover a topic, the bot says so rather than guessing.
It is the same pipeline that powers a real ChatbotGen chatbot. The demo simply throws the index away after 24 hours.
What people use it for
Support teams paste their help centre and check which questions it already answers well — the gaps are the articles worth writing next. Founders point it at a competitor's pricing page and interrogate it in plain language instead of reading it twice.
Agencies use it as a five-minute proof for a client: here is your own site answering your own customers, live, before anyone signs anything. Writers drop in a long guide and pull out the three facts they needed.
You can mix sources in one demo — a URL, a PDF and a block of pasted text all land in the same index, so the bot answers across them.
Limits, cost and privacy
The demo is free and needs no account: up to three URLs, three files and ten messages per session. We delete the content and the index after 24 hours, and we do not use it to train anything.
Only pages a normal visitor can reach are readable — anything behind a login is out of scope. When you want the chatbot to stay, sign up free and it moves to a permanent bot with a stronger model, an embed script for your site, and WhatsApp and Telegram.