Your data
Train an AI Chatbot on Your Own Data
Upload documents, add website pages, or paste YouTube links. The chatbot answers from those sources and says so when the answer is not there. No fine-tuning, no model training, no waiting.
Sources you can train on
Files, pages, and videos, alone or combined in one chatbot.
Documents
PDF, DOCX, TXT, and CSV. Product guides, policies, syllabi, manuals, price lists, anything you already hand to people as a file.
Website pages
Point it at the pages that answer questions: FAQs, shipping and returns, getting-started guides, pricing. Leave out the blog archive and the careers page.
YouTube videos
Paste a link to a lecture, tutorial, or product walkthrough. The transcript is indexed, so a two-hour recording becomes something you can question.
All of them together
One chatbot can hold documents, pages, and videos at once. A question is answered from whichever source actually covers it.
How training works here
Retrieval, not fine-tuning. That single choice explains everything about speed, cost, and accuracy.
Indexed, not fine-tuned
Your content is split into passages and indexed. No model is trained on it, which is why adding a document takes seconds rather than hours.
Retrieved per question
Each question pulls the handful of passages most likely to answer it. The model writes its reply from those passages, not from memory.
Grounded by design
When nothing in your sources covers a question, the chatbot says so. That is the behaviour you want on a support page or a syllabus.
Updated by re-uploading
Change a price or a policy, replace the document, and the next answer reflects it. Keeping it current is a content job, not an engineering one.
Choosing what to feed it
The quality of the answers is decided here, before anyone asks a question.
Start from your inbox
The questions you answer by email every week tell you which pages to add first. Sitemaps tell you nothing about what people actually ask.
Leave out what is stale
A retired plan page or an old announcement will be retrieved and quoted with full confidence. If a page is wrong today, do not feed it.
Prefer the page that answers
A dedicated returns policy beats a homepage that mentions returns in passing. Specific pages produce specific answers.
Test as a stranger
Ask in the words a customer would use, not the words your team uses. The gaps you find are pages to write or upload next.
Questions
What people ask before handing their content to a chatbot.
Put it to work
Once the chatbot answers well, these are the places it goes.
Upload one document and ask it a question
The free plan is enough to see whether the answers hold on your content.
Start with the page you send people most. Add it, ask the questions you get by email, and judge the answers before you pay for anything.