GLINR Studios

GLIN PROFANITY

Profanity, PII, and secrets filter for real text.

MIT-licensed. Runs offline. 24 languages, leetspeak, and homoglyph detection. Drop into Node, Next.js, Express, or Cloudflare Workers in minutes.

HOW IT WORKS

One method. Full context.

Call filter.check() and get a structured result. Homoglyphs and leetspeak are normalized before matching.

ts
import { Filter } from "glin-profanity";

const filter = new Filter({ allLanguages: true });

const result = filter.check(
  "This was a c@$tle move you sc#ndrel."
);

console.log(result.containsProfanity);
// true

console.log(result.flaggedWords);
// ["c@$tle", "sc#ndrel"]
Filter result

{

containsProfanity: true,

flaggedWords: ["c@$tle", "sc#ndrel"],

severity: "medium",

}

Homoglyph and substitution-aware. c@$tle is normalized before matching.

TRY IT

See the filter in action.

Type or paste any text. Click an example chip to load a curated test case.

output
// Run the scanner to see results here

Demo uses a subset of words. pnpm add glin-profanity for 24-language full coverage.

0 / 2000

Runs in your browser. Nothing is sent anywhere.

Categories:CleanFlaggedEdge case

80.6

F1 score (highest of 5 libs tested)

0%

False-positive rate

1039

ops / sec throughput

~1 ms

average latency

Source: glin-profanity shootout benchmark, 2026-04-20, 60-case torture set (Node v22.22.2).

BENCHMARKED

CAPABILITIES

Everything a guardrail needs.

Built for production moderation pipelines, not just demo word lists.

24 languages

English, Spanish, French, German, Japanese, Chinese, Hindi, Arabic and more. One unified API for multilingual moderation pipelines.

Leetspeak detection

Catches f4ck, sh1t, b1tch and aggressive Unicode homoglyph substitutions.

Custom word lists

Add domain-specific terms or allowlist false positives per use-case.

Offline by default

No cloud calls. Runs in Node, Bun, Deno, and Cloudflare Workers with no Node dependencies.

Configurable severity

Per-word severity scores. Build tiered moderation workflows without rewriting detection logic.

MCP server included

Ships glin-profanity-mcp: 24 tools for Claude, Cursor, and Windsurf out of the box.

PIPELINE

How the filter pipeline works.

Each input passes through normalisation, detection, and redaction before the result is returned.

Input
Tokenise
Normalise
Match
Redact
Output

24 LANGUAGES

DROP INTO YOUR STACK

Pick your runtime.

Works in Node, Next.js, Express, Cloudflare Workers, and anywhere JavaScript runs.

import { Filter } from "glin-profanity";

const filter = new Filter({
  allLanguages: true,
  detectLeetspeak: true,
  leetspeakLevel: "aggressive",
  normalizeUnicode: true,
});

const { containsProfanity, flaggedWords }
  = filter.check(userInput);

if (containsProfanity) {
  throw new Error(
    `Flagged: ${flaggedWords.join(", ")}`
  );
}

COMMON QUESTIONS

Frequently asked questions.

GitHub sourcenpm package