# Tavily Extract - Core.Today AI API

> Extract clean, LLM-ready content (markdown or text) from up to 20 web pages in one call, optionally reranked to a query. Billed per successfully extracted URL - about 4 credits per URL (basic), failed URLs are refunded.

- **Provider**: Tavily
- **Model ID**: tavily/extract
- **Category**: Web Search
- **Credits**: 4 per extracted URL (3.72/URL basic, 7.44 advanced; failed URLs refunded)
- **Speed**: Fast
- **Quality**: High

## Features
- Up to 20 URLs per request, processed together
- Clean main content as markdown or plain text
- Optional query reranks the extracted chunks by relevance
- advanced depth extracts tables and embedded content
- Per-URL failures reported in failed_results and not billed

## Use Cases
- Feeding the pages a search returned into an LLM context (RAG)
- Turning documentation or articles into markdown for indexing
- Monitoring known pages for content changes

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Required
- **urls**: array - Web page URLs to extract (1-20).

### Optional
- **query**: string - Optional intent to rerank the extracted chunks by relevance. With a query, raw_content holds the top chunks joined by [...].
- **chunks_per_source**: integer - Chunks per URL when query is set (1-5, default 3).
- **extract_depth**: string (default: basic) - advanced also extracts tables and embedded content and succeeds on more pages, at 2x price and higher latency.
  Options: basic, advanced
- **format**: string (default: markdown) - Format of raw_content.
  Options: markdown, text
- **include_images**: boolean (default: false) - Add the image URLs found on each page.
- **include_favicon**: boolean (default: false) - Add each page's favicon URL.
- **timeout**: number - Per-request timeout in seconds (1-60). Default 10s for basic, 30s for advanced.

## Examples

### Extract a page as markdown
Returns the main content of one page.
```json
{
  "model": "tavily/extract",
  "input": {
    "urls": [
      "https://docs.tavily.com/documentation/api-credits"
    ],
    "extract_depth": "basic",
    "format": "markdown"
  }
}
```

### Rerank several pages to a question
Top 3 chunks per page that answer the query.
```json
{
  "model": "tavily/extract",
  "input": {
    "urls": [
      "https://en.wikipedia.org/wiki/Retrieval-augmented_generation",
      "https://en.wikipedia.org/wiki/Large_language_model"
    ],
    "query": "how does retrieval reduce hallucination",
    "chunks_per_source": 3
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- The result is Tavily's JSON as-is: results[] (url, raw_content), failed_results[] (url, error) and usage
- Check failed_results even on success - pages that block bots or need login fail and are refunded
- Use advanced for pages with tables or heavy JavaScript; it costs twice as much per URL
- Pair with tavily/search: search first, then extract the pages you want in full

## Documentation
https://docs.tavily.com/documentation/api-reference/endpoint/extract