ebbe2d66d0
The web_search tool now accepts an optional `url` parameter that fetches a specific page directly (skipping the search step). This lets the model follow URLs it found in earlier search results instead of re-searching with increasingly desperate queries. Also updated the tool description to mention that page content is returned, so models know they will get more than just snippets. |
||
|---|---|---|
| .. | ||
| data_recipe | ||
| export | ||
| inference | ||
| training | ||
| __init__.py | ||