Adding a Website Resource

A Website Resource points your Agent at a web page. The Agent reads the page and keeps it up to date for you, automatically. With two optional settings, Content Selectors and Extraction Patterns, you can fine-tune exactly what the Agent takes from the page so its knowledge stays focused. This article shows you how.

Before You Start

Resources live inside a Place, so open the Place you want to add knowledge to, then add a new Resource and choose Website as the type.

A Website Resource is the best choice when your information already lives on a web page, because it refreshes automatically on a regular schedule. When the page changes, your Agent catches up on its own.

Adding the Resource

1. Open the Place and add a new Resource.
2. Choose Website as the type.
3. Give the Resource a clear Name.
4. Enter the URL of the page.
5. Optionally choose an Extraction Pattern. The default, No Extraction, is usually the right place to start (see below).
6. Save.

Content Selectors aren't in this dialog. You add them by editing the Resource after you create it (see Content Selectors below).

Why Focus Matters

The more focused your knowledge, the better your Agent answers. It's tempting to pull in a whole page and let the Agent sort it out, but every extra bit of noise (navigation menus, footers, cookie banners, unrelated sidebars) makes it harder for the Agent to find the right information. Trimming that noise is worth the effort. Content Selectors and Extraction Patterns are the two tools that help you do it.

Content Selectors

Content Selectors let you keep only the part of a page you care about. By default, the Agent reads the whole page. Add one or more selectors, and the Agent keeps just the matching parts and drops the rest.

You set Content Selectors by editing a Website Resource after you've created it. Open the Resource to edit it, and you'll find the Content Selectors field.

These are CSS selectors. To add one, type it and press Enter. Each becomes a chip, and you can add several. You name the parts to keep, so to leave out a navigation bar you simply select the main content instead.

This is the technical part, and there's no way around that. To find the right selectors, you (or a developer) look at the page's HTML to see how it's structured. It takes a little effort, but well-tuned selectors are powerful. For example, on a help center page you might select only the article body, so the menus and navigation aren't pulled in every time the page refreshes.

If you're not sure what to use, leave Content Selectors empty to start. The Agent reads the full page, and you can tighten it later.

Extraction Patterns

Here's the part people often get wrong: an Extraction Pattern is not how content gets pulled off the website. The page content is already fetched (and trimmed, if you used Content Selectors). An Extraction Pattern is an extra AI step that reshapes that content into a particular kind of knowledge before it's stored.

For a Website Resource you can choose:

• No Extraction: store the content as it is. This is the default.
• Hotel Facts Extraction: pull out structured facts about a property.
• Images Extraction: focus on images from the page.
• Knowledge Extraction: distill the content into general knowledge.
• Facts Extraction: pull out facts from the content.

Our recommendation: if your page already holds well-structured or condensed knowledge, choose No Extraction. Start there, see how your Agent does, and only reach for another pattern if you have a clear reason. Because an Extraction Pattern rewrites your content with AI, treat it as fine-tuning and use it deliberately, not by default.

The Best Combination

The strongest setup is well-tuned Content Selectors plus No Extraction. You trim the page down to exactly the content you want, then feed it to your Agent as is. Focused input in, better answers out.

Good to Know

• Website Resources refresh automatically. You don't re-add them when the page changes.
• Content Selectors only add what you select. Leave them empty to use the whole page.
• Leave the Extraction Pattern on No Extraction to use the content as it is.

What's Next

With your Website Resource added, here are a few ways to get more from your Agent's knowledge:

• Adding Document and Markdown Resources -- add knowledge that isn't on a web page
• Understanding Your Agent's Knowledge: Places and Resources -- the mental model behind how it all fits
• Creating Your First Agent -- see Resources in the full setup flow
• Creating and Managing Places -- organize your Resources into Places and assign them to Agents