Walk through the website first setup flow: paste your URL, let the crawl and indexing run in the background, answer the generated setup questions, then test and publish.
Let AI handle the repetitive questions. Free your team to focus on real conversations.
Get Started FreeSetup in minutes · No credit card required
Most chatbot projects stall in the same place. Someone opens a builder, sees an empty canvas, and has to decide what the bot should say before knowing what visitors will ask. OrhanAI starts from the opposite end: your website already contains the answers, so setup begins with your URL.
Paste a website address on the signup or new chatbot screen. OrhanAI creates a draft chatbot immediately and returns control to you. Nothing blocks on a long crawl.
In the background the crawler walks working pages on the same site, skips pages it cannot fetch, and records progress on the training source. You can watch discovered pages, crawled pages, and skipped pages while it runs. Local and private network addresses are rejected, responses are size capped, and one failing page does not stop the rest of the crawl.
Crawled pages are split into knowledge chunks and embedded so the chatbot can retrieve them later. Once there is enough content, an AI pass reads what your site is about and produces setup questions specific to your business.
Those questions are not a generic wizard. They use four control types so you answer in whatever form fits the question:
Single choice, for picking one option
Multiple choice, for picking several
Short text, for names, hours, or a service list
Long text, for policies and tone
When setup finishes you land in Playground. This is the chatbot with your real trained knowledge, not a preview mockup.
Ask the questions you know visitors ask. Where the answer is thin, use Improve with AI to get proposals you can review and apply, or add a training source to close the gap directly.
Treat Playground as a rehearsal. Every weak answer you find here is one a visitor will not run into later.
Publishing is deliberately gated. A chatbot stays unpublishable until it has either an allowed domain list or an explicit setting to allow any domain. An empty allowlist does not quietly mean open access.
That guard is enforced on every public widget request, not just at publish time. Draft and archived chatbots are never served publicly.
The last step is the smallest. Copy the embed snippet from the Widget tab and paste it into your site template:
Open the chatbot's Embed tab
Copy the 1 line script tag
Paste it before the closing body tag on your site
Reload your page and open the launcher
The widget renders inside Shadow DOM, so your site styles and the widget styles stay out of each other's way.
Once the chatbot is live, the useful work shifts to reading what people actually ask. Check the conversation archive for questions that fell back instead of answering, and turn those into new training sources. That loop, not the initial build, is what makes a website chatbot good.
Everyone assumes you need a developer to build an AI agent. You don't. If you have a website, a no-code AI platform can turn it into a working conversational AI agent in minutes, one that handles support, captures leads, and never clocks out. Here's exactly how to train, test, and launch it right.
