tavily（Tavily 网络搜索）

Tavily

Purpose

Provide a curl-based interface to Tavily’s REST API for web search, extraction, mapping, crawling, and optional research. Return structured results suitable for LLM workflows and multi-step investigations.

When to Use

• Use when a task needs live web information, site extraction, mapping, or crawling.
• Use when web searches are needed and no built-in tool is available, or when Tavily’s LLM-friendly output (summaries, chunks, sources, citations) is beneficial.
• Use when a task requires structured search results, extraction, or site discovery from Tavily.

Required Environment

• Require TAVILY_API_KEY in the environment.
• If TAVILY_API_KEY is missing, prompt the user to provide the API key before proceeding.

Base URL and Auth

• Base URL: https://api.tavily.com
• Authentication: Authorization: Bearer $TAVILY_API_KEY
• Content type: Content-Type: application/json
• Optional project tracking: add X-Project-ID: <project-id> if project attribution is needed.

Tool Mapping (Tavily REST)

1) search → POST /search

Use for web search with optional answer and content extraction.

Recommended minimal request:

curl -sS -X POST "https://api.tavily.com/search" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TAVILY_API_KEY" \
  -d '{
    "query": "<query>",
    "search_depth": "basic",
    "max_results": 5,
    "include_answer": true,
    "include_raw_content": false,
    "include_images": false
  }'

Key parameters (all optional unless noted):

• query (required): search text
• search_depth: basic | advanced | fast | ultra-fast
• chunks_per_source: 1–3 (advanced only)
• max_results: 0–20
• topic: general | news | finance
• time_range: day|week|month|year|d|w|m|y
• start_date, end_date: YYYY-MM-DD
• include_answer: false | true | basic | advanced
• include_raw_content: false | true | markdown | text
• include_images: boolean
• include_image_descriptions: boolean
• include_favicon: boolean
• include_domains, exclude_domains: string arrays
• country: country name (general topic only)
• auto_parameters: boolean
• include_usage: boolean

Expected response fields:

• answer (if requested), results[] with title, url, content, score, raw_content (optional), favicon (optional)
• response_time, usage, request_id

2) extract → POST /extract

Use for extracting content from specific URLs.

curl -sS -X POST "https://api.tavily.com/extract" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TAVILY_API_KEY" \
  -d '{
    "urls": ["https://example.com/article"],
    "query": "<optional intent for reranking>",
    "chunks_per_source": 3,
    "extract_depth": "basic",
    "format": "markdown",
    "include_images": false,
    "include_favicon": false
  }'

Key parameters:

• urls (required): array of URLs
• query: rerank chunks by intent
• chunks_per_source: 1–5 (only when query provided)
• extract_depth: basic | advanced
• format: markdown | text
• timeout: 1–60 seconds
• include_usage: boolean

Expected response fields:

• results[] with url, raw_content, images, favicon
• failed_results[], response_time, usage, request_id

3) map → POST /map

Use for generating a site map (URL discovery only).

curl -sS -X POST "https://api.tavily.com/map" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TAVILY_API_KEY" \
  -d '{
    "url": "https://docs.tavily.com",
    "max_depth": 1,
    "max_breadth": 20,
    "limit": 50,
    "allow_external": true
  }'

Key parameters:

• url (required)
• instructions: natural language guidance (raises cost)
• max_depth: 1–5
• max_breadth: 1+
• limit: 1+
• select_paths, select_domains, exclude_paths, exclude_domains: arrays of regex strings
• allow_external: boolean
• timeout: 10–150 seconds
• include_usage: boolean

Expected response fields:

• base_url, results[] (list of URLs), response_time, usage, request_id

4) crawl → POST /crawl

Use for site traversal with built-in extraction.

curl -sS -X POST "https://api.tavily.com/crawl" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TAVILY_API_KEY" \
  -d '{
    "url": "https://docs.tavily.com",
    "instructions": "Find all pages about the Python SDK",
    "max_depth": 1,
    "max_breadth": 20,
    "limit": 50,
    "extract_depth": "basic",
    "format": "markdown",
    "include_images": false
  }'

Key parameters:

• url (required)
• instructions: optional; raises cost and enables chunks_per_source
• chunks_per_source: 1–5 (only with instructions)
• max_depth, max_breadth, limit: same as map
• extract_depth: basic | advanced
• format: markdown | text
• include_images, include_favicon, allow_external
• timeout: 10–150 seconds
• include_usage: boolean

Expected response fields:

• base_url, results[] with url, raw_content, favicon
• response_time, usage, request_id

Optional Research Workflow (Deep Investigation)

Use when a query needs multi-step analysis and citations.

create research task → POST /research

curl -sS -X POST "https://api.tavily.com/research" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TAVILY_API_KEY" \
  -d '{
    "input": "<research question>",
    "model": "auto",
    "stream": false,
    "citation_format": "numbered"
  }'

Expected response fields:

• request_id, created_at, status (pending), input, model, response_time

get research status → GET /research/{request_id}

curl -sS -X GET "https://api.tavily.com/research/<request_id>" \
  -H "Authorization: Bearer $TAVILY_API_KEY"

Expected response fields:

• status: completed
• content: report text or structured object
• sources[]: {title, url, favicon}

streaming research (SSE)

Set "stream": true in the POST body and use curl with -N to stream events:

curl -N -X POST "https://api.tavily.com/research" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TAVILY_API_KEY" \
  -d '{"input":"<question>","stream":true,"model":"pro"}'

Handle SSE events (tool calls, tool responses, content chunks, sources, done).

Usage Notes

• Treat search, extract, map, and crawl as the primary endpoints for discovery and content retrieval.
• Return structured results with URLs, titles, and summaries for easy downstream use.
• Default to conservative parameters (search_depth: basic, max_results: 5) unless deeper recall is needed.
• Reuse consistent request bodies across calls to keep results predictable.

Error Handling

• If any request returns 401/403, prompt for or re-check TAVILY_API_KEY.
• If timeouts occur, reduce max_depth/limit or use search_depth: basic.
• If responses are too large, lower max_results or chunks_per_source.