DeFi yields now ($0.002): best supply and borrow APY of every Aave v3, Spark and Compound v3 market on Ethereum, Base, Arbitrum, Optimism and Polygon, read on chain every 10 minutes, highest APY first, filter by asset, chain or protocol. Use to find where stablecoins or ETH earn most.

GPT Image 2 Low - GPT Image 2, low quality: text to picture. Sharp detail, faces and readable text. The all-rounder. Any shape up to 3:1, transparent PNG or JPEG. $0.008, about 20 s.

Fresh random identifiers from a secure generator: UUID v4 (up to 50) plus random hex bytes.

Markdown to HTML with GitHub-flavoured tables and strikethrough, heading ids, the list of headings and links, and a word count. Raw HTML in the input is escaped unless allowHtml is true; links and images with javascript:, vbscript: or data: addresses are removed. Pure code, no outbound request. Input up to 200 KB.

Multilingual text embeddings (BGE-M3, 1024 dimensions) for up to 100 texts per call. For search, RAG and clustering.

Open Commerce Pathways session: a budget from 0.50 to 35 USDC, a flat 0.01 USDC service fee, delegated planning and one report within 1 hour. Returns a session token. Your wallet pays every seller directly. Unused sessions expire; no refunds.

Bounded low-cost inference with a signed delivery receipt

x402 payment readings on the XRP Ledger from our own scan of validated ledgers: payments and volume per asset for 1h/24h/7d, hourly and daily series, unique payers and merchants, top and new merchants, sample transactions with explorer links, and caveats. Each reading lists its source and time. JSO…

Olas-style yes/no forecast: { question } in, { p_yes, p_no, confidence, info_utility } out with reasoning and sources. BTC/ETH/SOL price questions with a threshold and a date are modelled from the Taifoon venue tape (sources replayable); any other question is a small open model's prior with no web search (no sources, confidence at most 0.3). Checked before you are charged. 0.01 USDC a forecast.

Call before hiring, paying or subcontracting a Québec (Canada) contractor. One call checks: active RBQ licence (status, categories, restrictions), RENA listing (banned from public contracts), environmental penalties, public contracts awarded on SEAO (count, total, buyers). Verdict ok / review / blocking with reasons and source dates. Input: name, NEQ or RBQ licence number. Official open data.

Source-returning web search with normalized results and signed delivery

Combined record in one call: AI provider health readings (public status pages and incidents), x402 payment readings on the XRP Ledger (payments, payers, merchants) and XRPL server info, plus GitHub, npm and MCP Registry readings for agent tooling. Raw readings first, each with its source and timest…

Current time from a trusted server: UTC ISO timestamp, Unix seconds and milliseconds, and the time in an optional IANA time zone.

Validates JSON data against a JSON Schema (draft 2020-12): valid true or false, and up to 50 errors with the instance path, the keyword and the message. Schema and data are each limited to 200 KB. Remote $ref targets are not fetched. No network call.

Cryptographically secure random integers in a range you choose (unbiased), up to 1,000 per call.

Decode a JWT without verifying it: header (algorithm, key id), payload claims, and issued, not-before and expiry times with whether it is expired now.

Dutch statutory extrajudicial collection costs (buitengerechtelijke incassokosten, WIK scale, Besluit vergoeding voor buitengerechtelijke incassokosten art. 2): 15/10/5/1/0.5% over the principal in brackets, minimum EUR 40, maximum EUR 6,775, optional 21% VAT when the creditor cannot reclaim VAT. Use it to compute the legally allowed collection fee on a Dutch claim.

Hashes of a text or base64 input (up to 1 MB): MD5, SHA-1, SHA-256, SHA-384, SHA-512 and CRC-32, and HMAC-SHA-256 or HMAC-SHA-512 when a key is given. Hex and base64 output. MD5 and CRC-32 are implemented here and checked against RFC 1321 and the standard CRC test vector. Body { input, encoding, algorithms, key }. Pure code and WebCrypto, no outside request.

Top-tier research in one call: the question plus 3 sub-questions (from Google's people-also-ask and related searches, or your own), up to 16 pages read in full with prompt-injection checks, the key quoted passages overall and per sub-question, Google's direct answer, and a papers block (Google Scholar) or news block when the topic needs it. Extractive, no LLM: every passage is quoted from its source.

Opens each URL in a cloud browser and returns one tab per URL, in input order, with the title, content type, extracted text, the accessibility tree and a link map. Results over the inline size limit return a preview per tab plus a download reference. Each call is one request.