Validate a Factur-X / ZUGFeRD invoice: send PDF bytes, JSON { url } or { pdfBase64 }, or raw CII { xml }. Returns a rule-based report — PDF/A-3 structure, EN 16931 CII anchors, arithmetic rules (BR-CO) and French PA « Flux 2 » checks (BR-FR) — with the detected profile. Structural + heuristic, not a certified conformance service. ?attest=1 → Ed25519-signed audit trail.

Extractive summary: selects the N most salient sentences via TextRank (TF-IDF cosine + PageRank) — a pure algorithm, no LLM, so deterministic and free of inference cost. Returns the summary and each chosen sentence with a salience score and index. Extractive (selects existing sentences), NOT abstractive/generative — that is the point: reproducible and model-free. Pre-condense a page before an LLM call. Provided text only. Over 1 MB truncated.

De-duplicate a list of records or strings. Exact: group by a `key` field (dot-path) or, by default, the whole item (key-order-insensitive). Fuzzy (`fuzzy`: a field name, or true for scalar strings): group by normalized Levenshtein similarity ≥ `threshold` (default 0.85) — e.g. 'Jean Dupont' ≈ 'jean dupont' but 'Jean Martin' stays separate. Returns unique[] and duplicates[] (kept + dropped indices). Fuzzy ≤ 1000 items.

Row-level diff of two CSVs (a→b) — reconcile two exports. With `key` (a column name, or a column index when header:false), rows pair by key → added / removed / changed (per-field old→new). Without a key, a whole-row multiset diff gives added/removed only. Delimiter auto-detected (, ; tab |); tolerant RFC-4180 parsing. First row is the header unless header:false. Max 5 MB per input; reported buckets cap at 10000.

Mask PII in free text before you send it to an LLM or write it to a log. Pattern-based detection (not ML): emails, phone numbers (Google libphonenumber), IBANs (mod-97), payment cards (Luhn), URLs and IPs. Choose types[] and mode: mask (j***@d***.com, readable), remove, or label ([EMAIL_1], re-identifiable by index). Names and postal addresses are out of scope. The input text is never logged or stored.

Extract a page's structured data: JSON-LD (schema.org), Open Graph tags, and best-effort microdata. Returns the parsed JSON-LD objects, detected schema.org types (Product, Offer, Review, Event, Recipe, Article…), microdata items and the og:* map. Send inline {html} or a public {url} (SSRF-checked). Malformed JSON-LD yields a {parseError} entry, never a crash. Microdata is best-effort; JSON-LD is the reliable path. Max 5 MB.

Test a regex against one or many inputs, safely. Send {pattern, flags?, input|inputs[], mode?}: 'match' returns the first match + capture groups, 'matchAll' every match, 'replace' the rewritten string (needs {replacement}). A mandatory ReDoS guard statically rejects catastrophic patterns (nested quantifiers, `.*.*$`, backrefs) BEFORE running — on Workers no timeout can stop a runaway regex. For agents that generate and dry-run regexes.

Extract values from a JSON document with a JSONPath expression or an RFC 6901 JSON Pointer, without returning the whole document. Send {data, path}; get {matches, count}. Path starting with $ = JSONPath ($.items[*].price, $..author, [0], [1:5], [*]); starting with / = JSON Pointer (/items/0/price). No-eval tree walk with bounded traversal; filter expressions [?(…)] are not supported by design.

Probe up to 10 x402 endpoints in parallel and get each one's reachability and decoded price — for an agent comparing several sellers at once. Send {urls:[…]}; returns [{url, reachable, isX402, price?, priceUsd?, error?}]. Each URL is SSRF-validated independently and NO payment is ever sent; one unreachable URL never fails the batch (it is reported per-URL). Reuses the same decoder as x402-probe.

Apply a jq-style filter to JSON — the JSON swiss-army knife. Supports identity, fields, index/slice/iterate, pipe |, comma, //, and/or/not, arithmetic, comparisons, array/object construction and builtins (select, map, keys, length, add, sort_by, group_by, unique, to_entries, has, range…). A SAFE pure-JS subset (no eval, bounded): no variables, def, reduce, regex or assignment. Returns the whole jq output stream as results[].

Extract salient keywords/key-phrases from text via TextRank (word co-occurrence graph + PageRank), merging adjacent high-ranked words into phrases. Pure, deterministic, model-free — extractive, NOT ML/NER or an LLM (the point: reproducible, no inference cost). English + French stopwords removed. Returns keywords[] with normalized scores (top ≈ 1). For content tagging/indexing and pre-RAG. Over 100k tokens is truncated.

Reshape JSON declaratively — predictable, not a language. `mapping` mirrors the output shape; its leaf strings are dot-paths into `data` (a.b.0.c; negative index ok; 'contacts.*.email' projects over an array; '=text' escapes a literal). Objects/arrays nest; number/boolean/null are literals. Missing paths → null. Field rename/flatten/projection to fit a payload to a target schema. Distinct from /v1/jq (declarative vs query language).

Resolve a website's best icon: parses <head> for link rel=icon / apple-touch-icon / mask-icon, reads a linked web app manifest's icons, and adds the /favicon.ico fallback, then ranks them (scalable/large first). Returns the best iconUrl plus all ranked candidates with rel/sizes/type/source. The URL is SSRF-checked; fetches are capped. For agents building directories or UIs.

Segment text into sentences — robust to the classic traps: abbreviations (Dr., e.g.), decimals ($3.14), URLs/emails (www.x.com), initials (J. R. R.) and ellipses. A period is a boundary only when followed by whitespace and a sentence-like start; !, ? and … are stronger; blank lines split paragraphs. Rule-based (EN/FR tuned), deterministic. A fine-grained chunking primitive, complements /v1/chunk. Over 1 MB truncated.

Historical daily exchange-rate series from the European Central Bank (via Frankfurter), with computed summary statistics. Pass base = 3-letter ISO 4217 currency (default USD), start = YYYY-MM-DD, optional end = YYYY-MM-DD (default = latest available), optional symbols = comma-separated target codes (default all), optional amount to scale rates (default 1). Range is capped at 366 days. Returns, per target currency, first/last/min/max/mean rate plus absolute and percentage change over the window, alongside the full daily series. ECB publishes on business days only; weekends and holidays are omitted.

Federal counterparty due-diligence dossier on one name, in a single call. Give name (a company or person) and optional state; get back six independent federal/authoritative screens composed into one verdict: SAM.gov registration (UEI/CAGE, active status), SAM federal exclusions (debarment/suspension), OFAC SDN sanctions, GLEIF Legal Entity Identifier, USAspending federal contract-award history, and FARA foreign-agent registration. Returns headline riskFlags (e.g. federally_debarred, sanctions_high_confidence_match, registered_foreign_agent), a cleared boolean (true only when both debarment AND sanctions sources answered and were clean — FARA status is disclosure context and does NOT affect cleared), a summary of booleans, and a per-source found/error/data block so one slow or empty source never fails the whole call. Composition of /api/gov/entity + /api/gov/exclusions + /api/law/sanctions-check + /api/business/lei + /api/gov/usaspending-awards + /api/gov/foreign-agents — all free, public-domain US data. Name matching is probabilistic: review flagged matches manually and confirm with a hard identifier (UEI, LEI) before acting. For vendor onboarding, KYC/AML triage, grant eligibility, and procurement integrity. The federal counterpart to /api/business/entity-screen (which covers state registries).

Live US severe-weather alerts from the National Weather Service. Pass a point ("lat,lon") to get every active watch, warning, and advisory for that exact location, or an area (2-letter US state or marine zone code) for area-wide alerts. Optionally filter by severity or urgency. Each alert returns the event type, severity, urgency, certainty, headline, affected-area description, issuing office, recommended public response, onset / expiry / end times, and the full hazard description plus protective-action instructions. Results are sorted most-severe first and capped by limit, with the true total reported. Public-domain NOAA / National Weather Service data, refreshed in real time.

NHTSA NCAP 5-Star crash-test ratings for a US vehicle by (make, model, modelYear). Returns one item per crash-tested body style with the overall star rating, front/side/rollover sub-ratings, modeled rollover probability, crash-avoidance tech flags (electronic stability control, forward-collision warning, lane-departure warning), and complaint/recall/investigation counts. A vehicle with no crash testing returns an empty list (a valid result, not an error). Backed by NHTSA's public 5-Star Safety Ratings program; data is public-domain US government records.

Resolve a MAC address (or bare OUI prefix) to its IEEE-registered hardware vendor. Accepts any common format — FC:FB:FB:01:02:03, fc-fb-fb-01-02-03, fcfb.fb01.0203, fcfbfb, or a 9-hex MA-S prefix. Uses longest-prefix matching across the IEEE MA-L (24-bit), MA-M (28-bit) and MA-S (36-bit) registries so blocks IEEE has subdivided resolve to the actual manufacturer. Returns the vendor name, the matched OUI prefix and which registry it came from, plus decoded address bits: whether the address is multicast/group, locally administered, or a randomized (privacy) MAC — for which no vendor exists by design. Authoritative bundled IEEE data.

Find the right 2s endpoint for a task using natural language. Pass q (e.g. 'check if a domain can be spoofed', 'is this CVE being exploited', 'screen a company for sanctions', 'decode a VIN') and get back the catalog's most relevant endpoints, ranked, each with its id, API path, method, group, description, and price. Deterministic keyword + synonym matching over the live endpoint registry (no LLM, no external calls) — a self-routing index so an agent (or its planner) can discover which paid call answers a question instead of reading the whole 290+ endpoint catalog. Returns the path you then call directly.