Web Search Research x402 offerings

Page change watch: reports what changed on a web page since your last check (lines added and removed). The first call saves a snapshot; each later call compares and updates it. Snapshots are private to the paying wallet.

Crawl a website into clean Markdown: follows same-site links from a start page (optionally under a path prefix), up to 100 pages, respecting robots.txt. Returns each page's URL, title and text. $0.005 + $0.001 per page. Add store=true (+$0.001) to get a 7-day share link instead of the data inline.

Cited answers: ask a question, get a concise answer written from live web pages with numbered [n] citations and the source list. Searches the web, reads the top pages and writes only from what they say (gpt-oss-120b).

Context packer: from long texts or URLs, picks the passages most relevant to your question and packs them into your token budget, with source markers. Fit more useful context into a prompt for less. Also callable as GET with the same fields as query parameters.

Ask any website or documentation a question: the site is crawled (up to 50 pages), indexed for 7 days and searched; the answer cites the pages it used. Later questions about the same site reuse the index. $0.03, or $0.05 above 25 pages.

Latest news headlines matching a query from 12 major outlets (BBC, NPR, Guardian, Al Jazeera, CBC, CNBC, NYT, TechCrunch, Verge, Ars Technica, CoinDesk, ESPN) plus Hacker News. Titles, links, source and time.

Since-my-cutoff briefing: everything that happened on a topic after a given date (for models whose knowledge stops there), newest first with dates and citations, plus what changed versus the old state.

Summarize any web page or long text. A headless browser renders the URL, an LLM writes a faithful summary: short (2-3 sentences), medium (one paragraph) or bullets. Also callable as GET with the same fields as query parameters.

Citation checker: for each claim and the URL cited for it, reads the source and says whether it really supports the claim (supported, partial, not supported, contradicted) with the deciding quote. Also callable as GET with the same fields as query parameters.

Ask a document: give a web page, PDF, Word, Excel or CSV URL (or raw text) and a question; get an answer backed by exact quotes from the document. Long documents are searched with embeddings first. Also callable as GET with the same fields as query parameters.

List every page on a website: reads robots.txt and all sitemaps (including sitemap indexes), with last-modified dates; falls back to on-page links. Up to 5000 URLs. Also callable as GET with the same fields as query parameters.

Scrape text and attributes from a rendered web page using CSS selectors. Returns every match per selector. Also callable as GET with the same fields as query parameters.

Shared answer cache: a recent cited answer to a question similar to yours (from /v1/answer) for a quarter of the price. Not charged when nothing close enough is cached.

Fact lookup for grounding: best-matching Wikipedia article summary plus structured Wikidata facts (website, founding date, birth/death dates, population, coordinates) and the source link. Any language edition.

Deep research report: plans several searches, reads up to 8 sources and writes a structured Markdown report (summary, key findings, details, open questions, numbered sources) with inline citations. Takes 20 to 60 seconds.

Bulk reader: up to 20 web pages read and cleaned to text in one call (instead of 20 separate calls), each trimmed to your character limit. Unreadable pages are flagged, not failed. Also callable as GET with the same fields as query parameters.

Rent a real Chrome browser for up to 60 minutes: opens a live session (optionally at a URL) and returns a session id, a screenshot, the page text and the clickable controls. Drive it step by step with /v1/browser/act (click, type, press, scroll, select, go back). The session stays open 10 minutes after each action. Also callable as GET with the same fields as query parameters.

Full-page PNG screenshot of any public URL rendered by a real browser. Returns PNG bytes. Also callable as GET with the same fields as query parameters.

Web archive: the Internet Archive snapshot of a page closest to a date, with the archived page text and a list of recent snapshots. See what a page said before it changed.

Fetch any public web page with a real headless browser (JavaScript rendered) and return clean Markdown. Use to read articles, docs, product pages or SPAs that a plain fetch cannot render. Also callable as GET with the same fields as query parameters.

Extract structured JSON from any web page. Give a URL plus a JSON schema or a plain-English prompt describing the fields you want; a headless browser renders the page and an LLM fills the schema. Also callable as GET with the same fields as query parameters.

Web search for AI agents. Send a natural-language query, get ranked results with title, url and snippet. Use when you need fresh information from the live web. Also callable as GET with the same fields as query parameters.

Scores how relevant a source is to a claim for AI fact-checking.

Check answer. Splits an AI answer into claims and checks each against Wikipedia, Wikidata and (for current claims) news. Returns per-claim verdict, confidence, quote, source URL and sentence, a source-quoted suggested_correction for wrong claims, and an answer summary with problem_sentences. $1.00 USDC.