Skip to main content

​
Overview

The webSearch tool performs web searches using the Firecrawl API, filtering results to prioritize trusted news outlets. It’s designed to find high-quality, diverse sources for argument analysis.

​
Function Signature

​
As Genkit Tool

export const webSearch = ai.defineTool(
  {
    name: 'webSearch',
    description: 'Searches the web using Firecrawl for maximum coverage. Returns results from trusted news sources.',
    inputSchema: WebSearchInputSchema,
    outputSchema: WebSearchOutputSchema,
  },
  async (input) => { ... }
)

​
As Standalone Function

export async function searchWeb(
  query: string
): Promise<{ title: string; link: string; snippet: string }[]>
Source: src/ai/tools/web-search.ts:54

​
Input Schema

​
query
string
required
The search query.

​
Input Type

const WebSearchInputSchema = z.object({
  query: z.string().describe('The search query.'),
});

​
Output Schema

​
results
WebSearchResult[]
required
An array of search results from trusted news sources.Each result contains:
  • title: The title of the search result
  • link: The URL of the search result
  • snippet: A brief summary of the content

​
Output Type

const WebSearchResultSchema = z.object({
  title: z.string().describe('The title of the search result.'),
  link: z.string().describe('The URL of the search result.'),
  snippet: z.string().describe('A brief summary of the search result.'),
});

const WebSearchOutputSchema = z.array(WebSearchResultSchema);

​
Trusted News Outlets

The tool filters results to prioritize these trusted sources: 
const TRUSTED_NEWS_OUTLETS = [
  "opindia.com",
  "swarajyamag.com",
  "thehindu.com",
  "indianexpress.com",
  "hindustantimes.com",
  "timesofindia.indiatimes.com",
  "ndtv.com",
  "bbc.com",
  "reuters.com",
  "aljazeera.com",
  "cnn.com",
  "nytimes.com",
  "washingtonpost.com",
  "theguardian.com",
  "businesstoday.in",
  "livemint.com",
  "economictimes.indiatimes.com",
  "firstpost.com",
  "wionews.com",
  "apnews.com",
  "bloomberg.com",
  "cnbc.com",
  "pib.gov.in",
];

​
How It Works

  1. Site Filtering: Constructs a search query with site: filters for trusted outlets
  2. Firecrawl Request: Sends request to Firecrawl API with 20 result limit
  3. Fallback: If filtered search returns few results, performs general search
  4. Deduplication: Removes duplicate URLs before returning
  5. Formatting: Returns up to 20 results with title, link, and snippet

​
Example Usage

​
Standalone Function

import { searchWeb } from '@/ai/tools/web-search';

const results = await searchWeb('climate change policy');

console.log(`Found ${results.length} results`);

results.forEach(result => {
  console.log(`${result.title} - ${result.link}`);
  console.log(result.snippet);
});

​
As AI Tool

import { ai } from '@/ai/genkit';
import { webSearch } from '@/ai/tools/web-search';

const response = await ai.generate({
  prompt: 'Find the latest news on climate change',
  tools: [webSearch],
});

​
Configuration

Firecrawl API Settings:
  • Limit: 20 results per request
  • Language: English (en)
  • Scrape Options: Markdown format
  • Endpoint: https://api.firecrawl.dev/v1/search

​
Environment Variables

​
FIRECRAWL_API_KEY
string
required
Your Firecrawl API key. Get one at firecrawl.dev.
FIRECRAWL_API_KEY=your_api_key_here

​
Error Handling

The tool includes multiple fallback mechanisms:
  • If trusted source search fails, performs general search
  • If Firecrawl returns < 5 results, adds general search results
  • Returns empty array if all searches fail (no throw)
  • Logs warnings for debugging

​
Performance

  • Average response time: 1-3 seconds
  • Returns up to 20 results
  • Automatically retries with general search if needed
  • Optimized for diversity and quality over quantity

Build docs developers (and LLMs) love

Get started for freeTalk to us