Web Search and Page Reading, Only When You Ask
Search the open web, read a page as clean text, or crawl a site a few hops deep — the only features here that use the network, and offline mode switches them off.
Web search is the one part of MyBenAI that is genuinely online, and it is treated as the exception it is.
Three tools, not one
Search returns results for a query. Read a page fetches a URL and extracts it as clean text, without navigation, ads or cookie banners. Crawl a site follows a site's own links a few hops deep and reads what it finds, for when the answer is spread across a documentation set rather than sitting on one page.
What leaves your phone
The query, and nothing else. Not your conversation, not your documents, not your memories — the model decides what to search for and only that string is sent. Every one of those requests is written to the network log, with the time, the destination and the reason, so you can check this claim rather than believe it.
Turning it off
Offline mode disables it entirely, and the model is told the tool is unavailable rather than being left to fail on it. A persona can also simply omit it from its allowlist, which is a good default for anything personal.
What it is for
Anything after the model's training cutoff, anything local to you, and anything you want a source for. It is also the engine underneath sub-agents and deep research, which search and read many pages in parallel and cite what they used.
Looking for something else? Every page on this site.