Module web_fetch_tool 

web_fetch built-in tool.

Companion to web_search: takes one or more URLs the agent already knows (from a prior web_search hit, a user message, a link_understanding summary, …) and returns the cleaned body text + title for each. Reuses the runtime’s existing LinkExtractor so:

• The fetch budget, deny-host list, max-bytes cap, timeout and host blocklist are exactly the same as the auto-link pipeline. There’s no second copy of the config to drift.
• The LRU cache is shared, so a web_fetch of a URL the user already pasted earlier in the session is free.
• Telemetry (nexo_link_understanding_fetch_total, nexo_link_understanding_cache_total, nexo_link_understanding_fetch_duration_ms) covers web_fetch calls as well — operators don’t need a second dashboard.

Distinct from web_search because the agent often knows the URL up front (skill output, RSS poll, calendar attachment) and would otherwise have to either hallucinate a search query or shell out to a fetch-url extension.

Structs

WebFetchTool