PrepLayer.ai

Near-duplicate detection across a batch of documents. Exact Jaccard similarity over word-3-shingle sets decides duplication; a 64-bit SimHash is returned per document as a compact content fingerprint. Up to 200 documents per call.

Deterministic text normalization: Unicode NFKC, zero-width/bidi stripping, smart-quote and dash folding, whitespace collapse, case folding. Returns a stable content hash so two texts can be compared for exact equality.

Validate and canonicalize one identifier: email, URL, E.164 phone, IBAN (mod-97), payment card (Luhn), UUID, SHA-256, git SHA, EVM address (EIP-55) or domain. Returns {valid, canonical, fingerprint}.

Rule-based multi-label text classification over a fixed taxonomy (urgency, financial_promise, credential_request, crypto_mention, cta, legal_offer, medical_claim, seo_spam, url_shortener, press_release, question, has_contact). Deterministic - identical text always yields identical labels.

Pull caller-defined structured fields from text using regex patterns with deterministic type coercion. Supported types: string, bool, date, url, email, number. Returns {fields: {name: value|null}, found, missing}. Pure deterministic — no LLM call.

Content-addressable hashing for caching and dedup. Three algorithms: sha256 (via node:crypto / Web Crypto), xxhash64 and murmur3 (inline pure-JS, deterministic across runs). Returns hex digest + algorithm + byte count.

URL-safe slug from arbitrary text. NFKD-decomposes unicode (é -> e), lowercases, replaces every non-[a-z0-9] run with a separator, strips leading/trailing separators, truncates on separator boundaries (never mid-word).

Split long text into pieces for RAG retrieval. Strategies: 'sentence' packs whole sentences into a size budget and never splits mid-sentence; 'paragraph' prefers \n\n boundaries; 'fixed' is a hard char window with overlap. The highest-value endpoint for agents that build retrieval pipelines.

Line-level LCS diff of two text blobs. Returns a structured edit script with insertions, deletions, unchanged line counts and per-operation metadata; with format='unified' also returns a unified-diff string.

Extract structured fields from unstructured text with byte offsets and normalized values: emails, URLs, IPv4, phones, money amounts, ISO dates, UUIDs, EVM addresses, SKUs, semver.

Redact PII from text before it is stored or logged. Returns redacted text plus exact byte spans and counts per class: credit_card, iban, email, private_key, evm_address, phone, ipv4, url.

Deterministic JSON-Schema subset validator. Catches malformed JSON before it breaks a pipeline. Supports: type, required, properties, items, minLength, maxLength, pattern, enum, minimum, maximum, additionalProperties. No external dependencies.

Run up to 10 primitive calls in one round trip. One x402 payment of $0.40 (=$0.04 per call, 20% off the single-call price). Each request specifies an `endpoint` (normalize, validate, extract, redact, classify, dedupe, diff, chunk, extract_custom, validate_json, slugify, hash) and the input for it; the response carries 10 results in the same order. Single-call batch — no streaming.