Perplexity launches Context API to extract clean web page data for AI agents
·2026.11.17 09:00
Key point
The API reads up to 10 URLs to return clean markdown and query-based snippets, outperforming competitors like Firecrawl and Exa in a multi-provider evaluation.
Details
- Perplexity launches the Context API, which processes up to 10 URLs to return clean Markdown, HTML, links, and query-relevant snippets for AI agents.
- In a benchmark against Firecrawl, Exa, Parallel, Keenable, and Context.dev across 715 sites, the Context API returned usable content for 92.3% of URLs, recovered 87.9% of key facts, and achieved 80.3% content precision.
- The service uses a model to score page blocks as main content or boilerplate, filtering out navigation and ads while preserving original wording for selected passages.
- Pricing is $1 per 1,000 successful URL fetches, with failed requests not billed; developers can control freshness via max_age_seconds and cache_only parameters.
Technical Implementation
- Supports parallel fetching where each URL has its own time budget and status, preventing one slow page from failing the entire batch.
- Returns metadata including title, description, language, author, published date, and Open Graph tags for each result.
- Default behavior checks Perplexity's web index for copies fetched within the last two days before triggering a live fetch.
Use Cases
- Enables coding agents to retrieve full API documentation including parameters and error cases missing from search snippets.
- Allows support agents to access complete policy pages to account for regional or product-specific exceptions.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.