Customize the Assistant per Page
When an assistant is embedded across a multi-page site, you can tailor its description and suggested questions for each page by adding <meta> tags to the page's <head>. The assistant reads these tags at startup and overrides the global configuration for that page only — no code changes, no extra API calls.
Page-Specific Description
Add a skdwn-description meta tag to replace the assistant's description on a particular page:
<meta name="skdwn-description" content="I can help you find cables, connectors, and network accessories." />The visitor sees this description instead of the one configured in General Settings, making the assistant feel context-aware — a product page shows product-related text, a support page shows support-related text.
Page-Specific Hints
Add a skdwn-hints meta tag to replace the global hints with questions relevant to the current page. The value is a JSON array of objects, each with a title (the label shown to the visitor) and an assistantMessage (the query sent when clicked):
<meta name="skdwn-hints" content='[
{"title": "What lengths are available?", "assistantMessage": "What lengths are available for this cable?"},
{"title": "Is it suitable for outdoor use?", "assistantMessage": "Can I use this cable outdoors?"},
{"title": "Compare with Cat6A", "assistantMessage": "What are the differences between Cat6 and Cat6A cables?"}
]' />When the assistant loads on this page, it shows these three hints instead of the ones configured in Configure Hints. On a different page with its own skdwn-hints tag, the hints adapt automatically.
Pages without either meta tag keep the assistant's global description and hints.
Page Preview Images in Highlighted Contents
When the assistant shows reference cards (highlighted contents), it can display a preview image for each source page. The image comes from the standard Open Graph tag in the crawled page:
<meta property="og:image" content="https://example.com/images/product-photo.jpg" />The data capture job captures this tag during execution and stores it with the content. When the assistant cites that page in an answer, the reference card shows the image alongside the title and summary — giving visitors a visual preview of the source before they click through.
No assistant configuration is needed — just make sure your pages have og:image tags and the data capture job will pick them up automatically.