Perplexity Context API adds page fetching for agents
Perplexity announced its Context API on 18 November 2026, offering developers a way to retrieve web-page content for an AI workflow. Its launch post quotes a price of $1 per 1,000 fetched URLs and…
Perplexity announced its Context API on 18 November 2026, offering developers a way to retrieve web-page content for an AI workflow. Its launch post quotes a price of $1 per 1,000 fetched URLs and describes Markdown, HTML, links and excerpts as outputs.
For a commerce app, the useful distinction is that this service reads URLs the developer already knows. It is a building block for a product-page audit or supplier-page review, rather than a finished merchant monitoring app.
Check freshness before checking prices
The official quickstart documents a default maximum cache age of 172,800 seconds, or 48 hours. Setting max_age_seconds to zero requests a live fetch. A promotion monitor built without that distinction could report yesterday’s offer as current.
The launch post mentions batches of ten URLs. Current documentation is more specific: Tier 0 allows ten per request, Tiers 1–3 allow fifteen, and Tiers 4–5 allow twenty-five. Developers should use their account’s tier limit rather than treating ten as a universal ceiling.
Keep extraction separate from interpretation
A sensible first test is a small set of your own product pages. Compare the extracted price, currency, variant and availability with the visible page. Then decide which fields are reliable enough to feed into an assistant. Missing text should produce an exception for review, not an invented value.
Perplexity reports 92.3% usable-content coverage, 87.9% fact recall and 80.3% content precision. Those are vendor benchmark claims, not results from a Shopify catalogue test conducted by The Founder’s Inn.
At the advertised rate, 500 billed URL fetches a day for 30 days would be $15 in fetch charges. That arithmetic excludes model usage, storage and the work of handling failures. The API can return an overall HTTP 200 response while individual URLs fail; the documentation says to inspect each result’s status before using its content.
For app builders, the opportunity is a simpler retrieval layer. The product work remains: choosing authorised sources, tracking when evidence was fetched, testing field accuracy and presenting uncertainty to the merchant.
More from News
See all →