Returns Brazilian company data from BrasilAPI (CNPJ lookup, MIT licence): legal name, trade name, status, opening date, legal nature, CNAE activities, address, share capital, and partners (QSA). CNPJ check digits are validated before any network call. Rate limit: BrasilAPI allows ~3 req/s.

Open positions of a Polymarket wallet (proxy wallet address): market, outcome, size, average entry price, current price, value in USDC, cash and percent PnL, plus totals of the listed positions and Polymarket's own portfolio value. The largest 500 positions are listed; truncated says when there are more. Information only, not financial advice.

Tone analysis of a text (max 10,000 characters): formality 0-1, sentiment, emotion scores, tone labels such as friendly, aggressive or urgent, and short suggestions. The reading level (Flesch-Kincaid grade) is computed in code with a simple syllable count, so treat it as an estimate. Model judgements are not measurements.

Searches OpenAlex (CC0) for scholarly works: title, authors with institutions, year, venue, DOI, open-access PDF link, citation count, topic and abstract (max 2,000 characters). Filters: publication year range, open access only, minimum citations; sort by relevance, citations or date (citations and date search the words as one phrase). Limit max 25.

A regular expression from a description, with examples that must match and examples that must not. The pattern is returned only when every example passes when tested in the service (one retry with the failures). Patterns with nested quantifiers are refused to avoid catastrophic backtracking. Examples are up to 10 strings of 300 characters each.

Fetches latest trade executions for a Polymarket condition ID or token. Returns timestamp, side (BUY/SELL), outcome label, execution price, size, value in USDC, and taker proxy wallet address. Limit up to 100 trades. Information only, not financial advice.

SQL from a question and a schema (CREATE TABLE statements or a JSON description of tables and columns), for postgres (default), mysql, sqlite or bigquery. The tables and columns used are checked against the schema; a query that does not match it is retried once, then refused. Writes (INSERT, UPDATE, DELETE, DDL) are refused unless allowWrite is true. The query is not run.

Searches the World Bank Listing of Ineligible Firms and Individuals for debarred and ineligible firms and individuals. Each hit: list, name, country, grounds, from/to dates, source URL. If a list is unreachable, returns listsUnavailable and 503 when none are checked.

Returns UK Global Tariff data for a 6–10-digit commodity code via the UK Trade Tariff API (OGL v3): description, third-country duty rate, VAT, measures, and preferential rates for a given origin country. Text search (`q`) returns matching commodity headings. Coverage: UK only, post-Brexit schedule.

Searches the US trade.gov Consolidated Screening List (Entity List, SDN, Denied Persons, Unverified, ITAR Debarred, etc.) by name and optional country. Each hit includes list name, aliases, addresses, programs, federal register notice, and match score.

US HTS lines from the USITC HTS REST API: HTS number, description, general, special and column 2 rates, units and footnotes. Look up by number (hts, e.g. 8471.30) or search words (q; every word must appear in the description). Chapter 99 lines (extra duties such as Section 301 or 232) come back separately. A blank rate on a statistical line means the rate is set on its parent heading. Up to 25 lines. Public domain.

Builds an outline (title, sections with headings and bullet points, optional word targets) for a topic, or from a text you give. Depth 1-3, at most 12 sections. Model output: review it before use; word targets are estimates. Input text max 10,000 characters.

Answers a question from the given text only (max 30,000 characters), with supporting quotes and their character offsets. Each quote is checked against the text and dropped if it does not occur in it. answerable is false when the text does not contain the answer. Model output: verify before relying on it.

Up to 15 question and answer pairs written from a text (max 30,000 characters). Each answer carries a supporting quote that is checked against the text; pairs whose quote does not occur in the text are dropped. Model-written: review the questions before publishing.

Up to 10 headlines for a text or topic in one style: news, seo (under 60 characters, enforced in code), social or email subject. Each headline comes with its character count. Model output: check claims before publishing. Input text max 10,000 characters.

Finds token contract addresses by symbol or name across chains using DexScreener search and verified lists. Returns matching tokens with chain, contract address, symbol, name, liquidity of largest pool, and verified list membership. Limit up to 25. Information only, not financial advice.

Fetches a site's security.txt from /.well-known/security.txt (fallback /security.txt) and parses the RFC 9116 fields: contacts, expiry (and whether it has expired), encryption, policy, acknowledgments, languages, canonical, hiring. Reports whether it is PGP-signed and lists RFC problems found. Public http(s) hosts only.

Token facts read on chain: for an ERC-20 on 22 EVM chains the name, symbol, decimals, total supply and EIP-1967 proxy implementation; for a Solana SPL mint the decimals, supply, mint and freeze authority. Adds price and liquidity from the token's deepest DexScreener pool on that chain, its logo and websites, and whether the Uniswap default list carries it (EVM). Information only, not financial advice.

Wayback Machine snapshots of a public URL: the capture closest to a date, the list of captures in a date range (one per day, max 500), first and last capture, and mode=diff, which compares the latest snapshot's text with the live page (added and removed lines, max 200 each). Diff is line-based and does not parse JavaScript-rendered pages.

Daily or monthly page views (user agent, all access) for up to 5 articles in one Wikipedia edition over a date range of at most one year, with totals, peak day and the change between the first and second half. mode=top lists the most viewed articles of one day. Data lags by about a day; bots and unknown agents are excluded.