Skip to main content
Agents search the web and read pages directly, without opening a full browser. Their answers and research draw on current information.

​
Two ways to pull from the web

Search

Query the web and get back relevant results with their content inline.

Fetch

Given a URL, read the page’s content directly.
These are the agent’s first choice for finding information. They are quick and lightweight. The agent escalates to the full browser only when it needs to interact with a page.
  • Answering questions that need current facts or data
  • Researching companies, people, products, or topics
  • Pulling documentation or reference material
  • Checking something the agent isn’t confident about, instead of guessing
A typical request:

​
Grounded answers

The agent fetches real sources, so its answers reflect what is on the page right now, not what a model memorized. For research tasks, it gathers across multiple sources and synthesizes a result.
When accuracy matters, ask the agent to cite its sources: “research X and include the links you used.”

Go deeper with a research session

For big topics, start a background learning session.