From website crawl to trained agent in under 10 minutes

The fastest path to a working Zenchat agent is also the one most people underestimate: paste your website URL and let the crawler do the first pass. Here’s what happens in that window.

1. Crawl

We discover pages from your sitemap first, then follow internal links to catch anything missing from it. Marketing pages, docs, and help center articles all get pulled in; login-gated pages and anything blocked by robots.txt are skipped automatically.

2. Clean and chunk

Raw HTML gets stripped of navigation, footers, and cookie banners — the parts of a page that would otherwise pollute retrieval with boilerplate. What’s left is split into chunks along heading and section boundaries.

3. Embed

Each chunk is embedded and indexed alongside a hybrid keyword index, so the agent can find both “what does this mean” matches and exact-term matches like plan names or error codes.

4. Set guardrails

Before it goes live, you set the boundaries: what the agent should never claim, when it should hand off to a human, and what tone matches your brand. This is the one step we don’t automate — it’s worth the two minutes.

5. Embed the widget

One script tag. It works on any site, loads asynchronously, and starts capturing conversations immediately.

Most teams have something live well inside ten minutes, then spend the following days refining sources as they watch real conversations come in — which is exactly the order we’d recommend doing it in.