Crawl any website, then search it by meaning — not just keywords.
Enter the site’s address. It reads sitemap.xml if there is one,
otherwise it follows links from the homepage.
Ask a question the way a visitor would. The words need not appear on the page.
The same retrieval, answered in prose with citations. Every stage is reported as it happens, and follow-up questions keep the thread.
Call the API and render results with your own markup. Every highlighted
field ships twice — plain text, and <mark> HTML — so you never
have to parse or trust server HTML.
And the chat endpoint, which streams its
progress and then the answer. GET /api/chat?siteId=…&q=… works too
for a single turn, if you would rather use EventSource or
curl -N.
Browsers are checked against each site's allowedOrigins, which defaults
to the crawled domain. To call it from another origin, add it in
sites.config.json — or set ["*"] to allow any.