Dipper

Fetch a web page and return its title and main content as Markdown. Use it to read a specific page in full: a URL the user gave you, or a promising result from web_search whose snippet was not enough. Only http and https URLs are supported. Pages may be served from a shared cache; set refresh to read the live page (billed at the full rate) when the page changes often or the cached copy looks stale.

Search the web and get a ranked list of results with title, URL, date and a short summary. Use it when you need current or external information: recent events, documentation, prices, facts you are not sure about. Prefer one well-chosen query over many similar ones; if the summaries are not enough, call web_fetch on the most relevant URL instead of searching again. Query rules: - At most 128 characters. Write concise keywords, not a long question. - Operators (combine freely): site:example.com only results from that site -site:example.com exclude that site lang:en results in a language (e.g. lang:zh, lang:en) after:2025-01-01 published on or after a date (YYYY-MM-DD) before:2025-12-31 published before a date - Example: "cloudflare durable objects pricing site:developers.cloudflare.com after:2025-06-01" max_results: number of results (1-50, default 10). Asking for more does not cost more but uses more context. include_content: include the full page text of every result. This is very large (often thousands of tokens per result) and rarely needed — leave it false and use web_fetch on one or two chosen URLs instead. Results may come from a shared cache. refresh: skip the cache and search again now; the new results also replace the cached copy for everyone. Billed at the full (uncached) rate, so only set it when the user needs the latest information or a cached result looks out of date.