web_search · web_fetch

web_search and web_fetch are two tools an AI agent uses when it needs to check the actual internet instead of relying only on what it already knows. As the names suggest, web_search means typing keywords into a search box and getting back a list of relevant pages, while web_fetch means opening one specific page at that address and reading its full content.

Think of it like doing research at a library. Pulling up a list of where relevant books are located at the search desk is web_search, and taking one of those books off the shelf and actually reading it is web_fetch. When you ask an AI something it couldn't have known at training time — like today's exchange rate or recent news — the agent uses these two tools in sequence to put together an answer.

However, letting an agent open just any site can be risky. So developers sometimes set up rules in advance that define which sites the agent is allowed to search or open. It's a kind of safeguard, similar to telling someone to only do research using trustworthy sources.

In articles, this appears as "domain allow/block parameters for web_search·web_fetch." This doesn't mean the two tools are newly created — it means that a setting has been added to the existing search-and-read functions to allow or block certain sites. In other words, the key point of the update isn't the tools themselves, but how much control you now have over them.

You can get a feel for both actions by asking an AI chatbot with web search enabled the following:

1. "Search for today's weather in Seoul and tell me" — the process of the AI summarizing an answer from multiple results resembles web_search.
2. "Summarize the content of this link (paste an actual URL)" — the process of the AI reading the entire page and summarizing it resembles web_fetch.

Trying these two steps separately makes it clear that searching and reading are distinct actions.