Tavily Researcher
Web research via Tavily's LLM-optimized APIs: search, extract, sitemap, and research.
Setup
Requires TAVILY_API_KEY environment variable. Get a free key (1000 credits/month) at https://app.tavily.com
Choosing the Right Tool
| Need | Script |
|---|---|
| Web search results | tavily_search.sh |
| Content from specific URLs | tavily_extract.sh |
| Discover URLs on a site (fast) | tavily_sitemap.sh |
| AI-synthesized report with citations | tavily_research.sh |
Quick Start
Search
scripts/tavily_search.sh "What is retrieval augmented generation?" --include-answerNews Search
scripts/tavily_search.sh "AI regulation updates" --topic news --time-range weekExtract Content from URLs
scripts/tavily_extract.sh https://example.com/article1 https://example.com/article2Discover URLs on a Site
scripts/tavily_sitemap.sh https://docs.example.com --max-depth 2 --limit 100AI Research Report
scripts/tavily_research.sh "Compare LangGraph vs CrewAI for multi-agent systems" --model proScripts
tavily_search.sh
Standard web search. Returns titles, URLs, snippets, and optional AI answer.
Key options:
--depth ultra-fast|fast|basic|advanced— Trade speed for relevance--topic general|news|finance— Optimize for content type--include-answer— Get AI-generated summary--include-raw-content— Get full page content--time-range day|week|month|year— Filter by recency--chunks-per-source N— Chunks per source (1-5, advanced/fast)--include-domains/--exclude-domains— Filter sources
tavily_extract.sh
Extract full content from specific URLs (up to 20 at once).
Key options:
--depth basic|advanced— Use advanced for JS-heavy pages--query TEXT— Rerank chunks by relevance--chunks-per-source N— Return only relevant chunks (1-5, requires --query)--format markdown|text— Output format
tavily_sitemap.sh
Discover URLs on a website without extracting content. Faster than crawl — use to understand site structure first.
Key options:
--max-depth N— Crawl depth 1-5 (start with 1)--limit N— Total URLs cap (default: 50)--instructions TEXT— Focus on specific types of pages
tavily_research.sh
AI-synthesized research with citations. Takes 30-120 seconds.
Key options:
--model mini|pro|auto— mini for focused queries, pro for comprehensive analysis--citation-format numbered|mla|apa|chicago--output-schema JSON— Get structured JSON output--output-file PATH— Save results to file
Cost Optimization
| Task | Recommended Approach | Credits |
|---|---|---|
| Quick fact check | tavily_search.sh --include-answer | 1 |
| Deeper search | tavily_search.sh --depth advanced | 2 |
| Full article content | tavily_extract.sh <urls> | 1 per 5 URLs |
| Discover site URLs | tavily_sitemap.sh | 1 |
| AI research report | tavily_research.sh | varies |
Domain-Specific Search
# Only search specific sites
scripts/tavily_search.sh "async python" --include-domains docs.python.org,realpython.com
# Exclude unreliable sources
scripts/tavily_search.sh "health advice" --exclude-domains reddit.com,quora.comAPI Reference
See references/api_reference.md for full parameter documentation.