@tavily/ai-sdk weekly downloads statistics

@tavily/ai-sdk

AI SDK tools for Tavily, compatible with Vercel's AI SDK v5 and v6.

Features

• Search: Real-time web search optimized for AI applications
• Extract: Clean, structured content extraction from URLs
• Crawl: Intelligent website crawling at scale
• Map: Website structure discovery and mapping

Installation

npm install @tavily/ai-sdk
# or
pnpm install @tavily/ai-sdk
# or
yarn add @tavily/ai-sdk

Prerequisites

You need a Tavily API key to use these tools. Get one at tavily.com.

Set your API key as an environment variable:

export TAVILY_API_KEY=tvly-your-api-key

Or pass it directly to the tool configuration.

Usage

Tavily Search

Search the web for real-time, AI-optimized information:

import { tavilySearch } from "@tavily/ai-sdk";
import { generateText, gateway, stepCountIs } from "ai";

const result = await generateText({
  model: gateway("openai/gpt-5-mini"),
  prompt: "What are the latest developments in quantum computing?",
  tools: {
    tavilySearch: tavilySearch({
      searchDepth: "advanced",
      includeAnswer: true,
      maxResults: 5,
    }),
  },
  stopWhen: stepCountIs(3),
});

Configuration Options:

• apiKey?: string - Tavily API key (defaults to TAVILY_API_KEY env var)
• searchDepth?: "basic" | "advanced" - Search depth (default: "basic")
• topic?: "general" | "news" | "finance" - Search category (default: "general")
• includeImages?: boolean - Include images in results (default: false)
• includeAnswer?: boolean - Include AI-generated answer (default: false)
• maxResults?: number - Maximum results to return (default: 5)
• includeImageDescriptions?: boolean - Include AI-generated image descriptions
• includeRawContent?: false | "markdown" | "text" - Include raw content in specified format
• includeDomains?: string[] - List of domains to include in search
• excludeDomains?: string[] - List of domains to exclude from search
• maxTokens?: number - Maximum number of tokens in response
• timeRange?: "year" | "month" | "week" | "day" | "y" | "m" | "w" | "d" - Time range for results
• chunksPerSource?: number - Number of content chunks per source
• country?: string - Country code for localized results
• startDate?: string - Start date for results (YYYY-MM-DD format)
• endDate?: string - End date for results (YYYY-MM-DD format)
• days?: number - Number of days back to search (for news/finance topics)
• autoParameters?: boolean - Enable automatic parameter optimization
• timeout?: number - Request timeout in milliseconds
• includeFavicon?: boolean - Include favicon URLs in results
• includeUsage?: boolean - Whether to include credit usage information in the response.
• proxies?: { http?: string, https?: string } - HTTP/HTTPS proxy configuration
• apiBaseURL?: string - Custom API base URL

Tavily Extract

Extract clean, structured content from URLs:

import { tavilyExtract } from "@tavily/ai-sdk";
import { generateText, gateway, stepCountIs } from "ai";

const result = await generateText({
  model: gateway("openai/gpt-5-mini"),
  prompt: "Extract and summarize the content from https://tavily.com",
  tools: {
    tavilyExtract: tavilyExtract(),
  },
  stopWhen: stepCountIs(3),
});

Configuration Options:

• apiKey?: string - Tavily API key (defaults to TAVILY_API_KEY env var)
• includeImages?: boolean - Include images in extracted content (default: false)
• extractDepth?: "basic" | "advanced" - Extraction depth (default: "basic")
• format?: "markdown" | "text" - Output format (default: "markdown")
• timeout?: number - Request timeout in milliseconds
• includeFavicon?: boolean - Include favicon URLs in results
• includeUsage?: boolean - Whether to include credit usage information in the response. NOTE:The value may be 0 if the total successful URL extractions has not yet reached 5 calls. See our Credits & Pricing documentation for details.
• proxies?: { http?: string, https?: string } - HTTP/HTTPS proxy configuration
• apiBaseURL?: string - Custom API base URL

Input Parameters (for AI agent):

• urls: string[] - Array of URLs to extract content from (required)
• extractDepth?: "basic" | "advanced" - Override extraction depth per call

Tavily Crawl

Crawl websites to discover and extract content from multiple pages:

import { tavilyCrawl } from "@tavily/ai-sdk";
import { generateText, gateway, stepCountIs } from "ai";

const result = await generateText({
  model: gateway("openai/gpt-5-mini"),
  prompt: "Crawl tavily.com and tell me about their integrations",
  tools: {
    tavilyCrawl: tavilyCrawl(),
  },
  stopWhen: stepCountIs(3),
});
```…