Web Search

Use these tools to give agents general web search, news search, answer-style search APIs, Google- or Baidu-oriented results, or a self-hosted SearXNG backend.

Tools On This Page

Setup

Add the tool to an agent's tools list; see Per-Agent Tool Configuration for inline options and include_tools/exclude_tools. duckduckgo, googlesearch, and baidusearch work without any setup. searxng needs only a reachable host URL. The API-backed tools need an api_key, set through the dashboard or credential store (see Security Restrictions). Each API-backed tool also reads its key from the environment when no api_key is stored:

Missing Python dependencies install on first use; see Automatic Dependency Installation. Every toolkit that has an all option enables all of its functions when all: true, regardless of the individual enable_* flags, except that [tavily] still provides only one search function.

[duckduckgo]

duckduckgo searches the web and news through DuckDuckGo with no API key. It provides web_search(query, max_results=5) and search_news(query, max_results=5), both returning DDGS results as JSON.

agents:
  researcher:
    tools:
      - duckduckgo:
          fixed_max_results: 8
          timelimit: w

[googlesearch]

googlesearch provides the same web_search() and search_news() functions, options, and output as [duckduckgo], except that it has no backend option and requests the Google engine. It is not an official Google API. DDGS currently falls back to automatic engine selection for Google requests, so results and ranking are not guaranteed to come from Google. Use [serper] or [serpapi] when you need real Google results.

agents:
  researcher:
    tools:
      - googlesearch:
          modifier: site:docs.mindroom.chat
          fixed_max_results: 6

[baidusearch]

baidusearch provides baidu_search(query, max_results=5, language="zh"), which returns a JSON array of results with title, url, abstract, and rank.

Only the query and result count reach Baidu, so language, fixed_language, headers, proxy, timeout, and debug do not change the search.

agents:
  cn_research:
    tools:
      - baidusearch:
          fixed_max_results: 8

[tavily]

tavily searches current information through the Tavily API and can extract page content from URLs. It provides one search function, web_search_using_tavily(query, max_results=5) by default or web_search_with_tavily(query) when enable_search_context: true, plus extract_url_content(urls) when enable_extract: true. web_search_using_tavily() returns results and, with include_answer, a synthesized answer, as Markdown or JSON according to format. web_search_with_tavily() returns one compact context block instead of a result list. extract_url_content() accepts one URL or a comma-separated list and returns page content as Markdown or plain text according to extract_format.

The filters from topic through chunks_per_source apply only to web_search_using_tavily(), not to context mode.

agents:
  newsdesk:
    tools:
      - tavily:
          enable_extract: true
          topic: news
          time_range: week

[exa]

exa is the research-oriented search toolkit. It provides search_exa(query, num_results=5, category=None), get_contents(urls), find_similar(url, num_results=5), exa_answer(query, text=False), and, only when enable_research: true, the long-running research(instructions, output_schema=None). Search results include title, author, published date, URL, and page text truncated to text_length_limit.

Publication-date filters cannot be combined with the company and people categories.

agents:
  analyst:
    tools:
      - exa:
          enable_research: true
          type: auto
          category: news
          include_domains:
            - matrix.org
            - element.io

[serpapi]

serpapi provides search_google(query, num_results=10) and, when enabled, search_youtube(query) through SerpApi. search_google() returns organic results plus recipes, shopping, knowledge graph, and related questions. search_youtube() returns video, movie, and channel results.

agents:
  researcher:
    tools:
      - serpapi:
          enable_search_youtube: true

[serper]

serper provides search_web(query, num_results=None), search_news(query, num_results=None), search_scholar(query, num_results=None), and scrape_webpage(url, markdown=False) through the Serper API. All functions return Serper's raw JSON, and scrape_webpage(markdown=True) also requests the page as Markdown.

agents:
  analyst:
    tools:
      - serper:
          location: de
          language: de
          enable_scrape_webpage: false

[searxng]

searxng searches through your own SearXNG instance. It provides search_web, image_search, it_search, map_search, music_search, news_search, science_search, and video_search, each taking (query, max_results=5). The instance must allow JSON output (json in SearXNG's search.formats).

The tool sends no credentials, so put any authentication or access policy at the network or reverse-proxy layer.

agents:
  privacy_research:
    tools:
      - searxng:
          host: https://search.example.com
          engines:
            - duckduckgo
            - wikipedia

[linkup]

linkup provides web_search_with_linkup(query, depth=None, output_type=None), which returns Linkup's response as a result list or a sourced answer. The configured depth and output_type apply when a call does not pass its own.

agents:
  briefings:
    tools:
      - linkup:
          depth: deep
          output_type: sourcedAnswer

Related Docs

• Tools Overview
• Research Sources
• Web Scraping & Browser