Skip to main content
The orq:web_fetch tool retrieves text from a URL chosen by the model. The AI Gateway checks the URL before fetching it and returns the extracted content to the model.

Quick start

The examples use the client configuration from the Server tools overview.

Configuration

Domain entries must be bare hostnames such as example.com. An entry with a scheme, a path, or a *. wildcard returns 400. Subdomains match by suffix, so example.com also covers docs.example.com.

URL restrictions

The tool accepts public http and https URLs. It rejects localhost, hostnames ending in .localhost, .local, or .internal, and any address that is loopback, link-local, private, multicast, unspecified, or otherwise not globally routable. The hostname must resolve to a public address, and the resolved addresses are checked as well. Domain rules are checked before the fetch runs. When a URL is rejected, the model receives the reason and can choose another URL. Rejected URLs do not use a max_uses slot.

Usage

Fetches that pass URL validation and the usage limit appear at usage.server_tool_use.web_fetch_requests. There is no separate server-tool charge for web fetch in this release. orq:web_scraper remains accepted as a legacy alias. The alias does not support the configuration fields above. Use Web search when the URL is not known in advance.