Provides three web information retrieval tools: general search, academic search, and page scraping.
Tools
tavily_search— AI-powered web search that returns refined, high-quality results and can include answer summaries. RequiresTAVILY_API_KEY。arxiv_search— Searches arXiv papers and returns titles, authors, abstracts, and links. Using English keywords works best.web_fetch— Scrapes the page content of a specified URL and extracts it as Markdown or plain text.
Examples
General web search:
tavily_search(query="2024年诺贝尔物理学奖")
Deep search, returning more results:
tavily_search(query="transformer architecture explained", num_results=8, search_depth="advanced")
Search academic papers:
arxiv_search(query="large language model reasoning", max_results=5)
Search first, then scrape a specific page:
web_fetch(url="https://arxiv.org/abs/2301.00001")
Scrape as plain text and limit the character count:
web_fetch(url="https://example.com/article", extract_mode="text", max_chars=10000)
Tips
- Search first, then scrape: First use
tavily_searchto search, then useweb_fetchto read the result pages of interest in depth. - Prefer arxiv for academic questions: When it involves papers or cutting-edge research, use
arxiv_searchdirectly; it is more precise than general search. - Optimize keywords: If the results are not ideal, try different or more specific keywords.
- Cite sources: Always include the URL of the information source in your answer.