Search Tool

Applies to: v4.0.8.2

Search is built on ddgs, and the source requires ddgs>=9.10.0.

1. Recommended Search + Browse flow

How to read this diagram

• Search discovers candidates; it does not guarantee high-quality body text.
• Browse turns those candidates into readable evidence, which is why the v4.0.8.2 mental model is now “Search + Browse”, not “Search + Playwright”.

2. Available methods

• search(query, timelimit=None, max_results=10)
• search_news(query, timelimit=None, max_results=10)
• search_wikipedia(query, timelimit=None, max_results=10)
• search_arxiv(query, max_results=10)

3. Initialization

from agently.builtins.tools import Search

search = Search(
    proxy=None,
    timeout=20,
    backend="auto",
    search_backend=None,
    news_backend=None,
    region="us-en",
    options=None,
)

Notes:

• backend is the default fallback backend
• search_backend and news_backend can override it separately
• options is passed through to ddgs

4. Use it as an Agent tool

from agently import Agently
from agently.builtins.tools import Search

agent = Agently.create_agent()
search = Search(region="us-en", backend="google")

agent.use_tools([
    search.search,
    search.search_news,
    search.search_wikipedia,
    search.search_arxiv,
])

5. Return shape

search / search_news / search_wikipedia commonly return:

• title
• href or url
• body / snippet
• source
• date

search_arxiv returns:

• feed_title
• updated
• entries[]

6. Recommended composition

The recommended pipeline is now:

• Search + Browse

because:

• Browse already encapsulates Playwright / PyAutoGUI / bs4 fallback
• workflow code should not decide which browser backend to call

7. Daily News Collector usage

Agently-Daily-News-Collector primarily uses:

• search_news(...)

and then feeds results into Browse for body extraction. That is the most recommended “search first, read second” pattern in v4.0.8.2.