Implements Web fetch and search through scoped network capabilities. fetch and search own content policy; settings manages provider configuration; provide registers the Host factory. It can run in a Host composition without an Agent engine.
probe supports process health checks: send HEAD, fall back once to GET only for 405/501, never follow redirects or read the response body, and wait for response cleanup. It requires its own network authorization and is unavailable in incognito mode; retrieval settings and search credentials do not control endpoint probes. Permission wait is included in elapsed time. HTTP 2xx is healthy, 3xx is unknown, and 4xx/5xx is unhealthy; denied or failed requests report an unknown endpoint.
Fetch preserves complete extracted text. Redirects stop on an actual repeated target. Documents outside the recursive readability adapter‘s supported shape, or whose readability extraction fails, return the complete original HTML as text. Parser workers retain their native permits through completion; scoped response cleanup and transport idle/connect deadlines remain. Search retains complete returned titles/snippets and the caller’s selected result count within Tavily's documented 20-result endpoint range.