Web & Data

Fetch a public web page and return clean readable text with its title and metadata - the form a model can actually use, with navigation, scripts and styling removed. Follows up to 3 redirects, caps the download at 2000000 bytes and gives up after 15000ms. Reaches the PUBLIC internet only: the hostname is resolved and every redirect hop re-checked against private, loopback, link-local and CGNAT ranges, so this cannot read internal services or cloud metadata. Returns raw HTML too on request.

Search the healthcare vocabularies loaded on this deployment and get codes back with a MEASURED confidence and full provenance: which dataset, loaded from where, under which licence, and whether it is synthetic. The scoring function is documented on every response (exact code 1.00, exact display 0.95, exact synonym 0.90, prefix 0.60-0.85, token overlap 0.30-0.75) rather than being an opaque number.

Emit the actual files that fix a site's AI legibility, ready to deploy: a real /llms.txt, a /robots.txt that stops excluding AI crawlers (with a unified diff against the current one), and a JSON-LD block for the page . Nothing is invented: every link in the llms.txt is discovered on the site and then FETCHED to confirm it answers 200, with failures dropped and counted, and every JSON-LD field is copied from something the page actually declares — anything that cannot be grounded is…

Full-text search over what machine-payable services actually publish — their descriptions, input schemas, parameter names, prices and settlement networks — across ~18,000 x402 endpoints. Built from their own payment challenges, which a general crawler discards as a dead page: ask for a parameter name, a schema field or a network and you match services whose one-line blurb never mentions it. Filter by price ceiling, by whether the service publishes a call spec, and by payment protocol.

Give it up to 10 URLs and get back the readable text of each — titles, clean body text, navigation and scripts removed — in a single call and a single payment. The half of every retrieval pipeline an agent already has links for: search results, a sitemap, an RSS feed, a user's reading list. Pages are fetched concurrently under a deadline.

Audit whether a website can actually be read, quoted and cited by AI agents, and get a prioritised fix list. Probes the homepage as each real AI crawler (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot, CCBot, meta-externalagent, Amazonbot) to catch the 429s, 403s and soft-404s that make a live site invisible; parses robots.txt per agent including the robots-only training tokens Google-Extended and Applebot-Extended;

Get a page's title, description, canonical URL and social preview tags. Send a URL and receive its metadata: title, description, Open Graph and Twitter card fields, canonical address, language and favicon. The page is rendered by a real browser, so tags injected client-side are captured too.

Reveal the server_seed for a round so anyone can verify past draws. ?round=<id>

Provably-fair round commit: sha256(server_seed) before any draw. ?round=<id>

Extract the title, description and Open Graph metadata of any web page URL — link previews and page metadata. ?url=https://example.com

Provably-fair random integer from a committed seed. ?client_seed=s&nonce=1&min=1&max=100&round=1

Sitemap lookup: locates a site's sitemap from robots.txt or same-origin /sitemap.xml, fetches it server-side and returns up to 100 normalized http/https URLs with the sitemap URL, document kind, count, truncation flag, a timestamp and a tamper-evident receipt. Single GET query parameter (url). One document only; child sitemaps are not fetched.

Give a URL and a question; get an answer grounded in that page, WITH the passages the answer was drawn from. We fetch the page, chunk it, embed the chunks and your question, retrieve the closest 4 passages and answer from those only. Nothing is stored — it is a one-shot pipeline, not an index. The retrieved passages come back with the answer so you can check it rather than trust it, and when nothing on the page is relevant the product says so instead of letting the model improvise.

Render any web page in a real headless browser and return a PNG screenshot. Supports viewport width/height (320-1920) and full-page capture. Ideal for agents doing visual verification, UI monitoring, archiving or content review.

Keyword & Entity Extraction: extract keywords, entities and topics from text Executed deterministically by the AmanChain node against live consensus state (real blocks, pools, balances) — paid per call via x402, settled in USDC on Base. Input: JSON body {"serviceId":"aman-keywords","request":" "}. No account, no API key.

List a Person's external profiles. Returns every known third-party profile (LinkedIn, Wikipedia, social) for the Person.

Reveal the server_seed for a round so anyone can verify past draws. ?round=<id>

Provably-fair round commit: sha256(server_seed) before any draw. ?round=<id>

Current weather conditions for any coordinates: temperature, wind, humidity, precipitation and weather code for a city or location. ?lat=52.52&lon=13.41

Fetch any public web page or PDF and return clean main-content Markdown for AI agents / LLMs (boilerplate, nav and ads removed) plus title & metadata. CJK-aware; optional Simplified<->Traditional (incl. Hong Kong / Taiwan) conversion. Respects robots.txt. $0.002 per call, USDC on Base (x402). Failed extractions are not charged.

Read any public web page as clean readable Markdown for RAG and LLM pipelines — main article text with nav/ads stripped, plus title. The reader/scraper building block for agents. ?url=https://example.com&max_chars=20000

Domain registration lookup via RDAP (modern WHOIS). Returns registrar, registrar_url, created_at, updated_at, expires_at, days_until_expiry, name_servers array, status array, and registrant_country. Clean JSON — no raw WHOIS text to parse. Covers all major TLDs.

List every link on a page, resolved to absolute URLs. Send a URL and receive every hyperlink found on the page, resolved to absolute addresses and returned with its anchor text. The page is rendered by a real browser first, so links injected by JavaScript are included rather than missed.

Verify a research claim against available evidence