Perplexity launches Context API so AI agents can read entire web pages
IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds
Perplexity is offering developers Context API, a service that hands AI agents the useful text of several web pages in one go, without menus or advertising. Context API also returns each page's links and the fragments that best answer a question, taken from its index or read on the spot. Perplexity charges $1 per 1,000 pages read without errors.
On Tuesday, November 17, Perplexity introduced Context API, a developer tool that reads up to 10 web addresses in a single request and returns, for each one, the clean content in Markdown or HTML, the page's links and the passages most relevant to a query.
According to the company, this means an AI agent does not have to rely on search engine snippets or on raw HTML packed with menus, cookie notices and ads. The passages are not a summary: they are pieces of the page in its original wording, picked for how they relate to the query and kept within the token limit the developer sets. Each result also carries metadata such as the title, description, language, author and publication date, plus the page's Open Graph tags.
The content comes from Perplexity's web index or from a live read. By default, the API looks for a copy of the page taken in the last two days and, if none exists, reads it right then. Developers can demand the live version every time, for prices, scores or breaking news, or widen the window for material that rarely changes, and each result states where it came from and when. Addresses are read in parallel, each with its own status, so one slow or broken page does not make the rest fail.
Perplexity compared Context API with Context.dev, Keenable, Parallel, Exa and Firecrawl on thousands of addresses taken from real agent requests, across 715 sites, and a judge model scored the results without knowing which provider they came from. According to the company, its API returned usable content for 92.3% of the addresses, captured 87.9% of the key facts and reached 80.3% precision. This is a measurement made by Perplexity itself.
The price is $1 per 1,000 addresses that return a correct result; failed ones are not charged. The announcement gives no details on usage limits or countries.
Why it matters · analysis and opinion
Reading a web page properly is one of the weak spots of AI agents: they either settle for the snippet a search engine gives them or load an entire page full of noise, which makes each answer more expensive and less reliable. With this service, Perplexity starts competing with companies that specialize in preparing web content for models, and it does so with a price per page read that makes the cost easy to calculate. For those building agents, the advantage lies in being able to choose between speed and freshness in each request. The quality figures, however, come from a test designed by the company itself, so the sensible step is to measure it with the pages each project really needs before switching providers.
Official source: Perplexity · Written with the help of AI: how we make the news


