Web & browser

Read the open web and, under governance, act in a web application. Search to find, fetch to read, and a real browser only when a page has to be driven.

web_search

Searches the web and returns titles, URLs and snippets. Use it to find sources for anything current, then read the ones that matter with web_fetch.

How an agent gets it. Declare it in the agent's tools: list.

• Needs a search provider configured for the organisation under Admin → Web Search; without one it answers web_search is not configured.
• depth: "thorough" runs a deeper, costlier search — for niche or very recent topics. The default is fast.

Arguments

web_fetch

Reads a URL. A web page comes back as its main text, with navigation and boilerplate removed, plus its title, publication date and source; a PDF comes back as text with [page N] markers.

How an agent gets it. Declare it in the agent's tools: list. Needs outbound web access in the deployment (tools_web_enabled); without it the tool answers that it is disabled.

• Long documents come back in sections: the reply says which characters you have and the offset to pass to read on.
• Fetched images and PDFs are also saved as run artifacts and placed in /work for a later code_execution call.
• A page's list of images ("Images on this page") is how the Document Author finds a company's real logo.

Arguments

browser_use

Drives a headless browser: navigate and extract read a page; click, type, submit and download act on it.

How an agent gets it. Declare it in the agent's tools: list.

• Which sites it may reach is decided by the browser_egress policy, which is deny by default.
• Every write action pauses for approval before it happens; extracted text is DLP-screened; every session is recorded as an artifact.
• See Browser actions and the IT Fulfilment agent, its first user.

Arguments