Documentation · API · Guides · Changelog · Contributing · Bahasa Indonesia

---

Indonesian Badwords

Keep chat, comments and usernames civil in Bahasa Indonesia, and stay in control of what counts as "bad".

People type g0bl0k or ＧＯＢＬＯＫ (both caught by default), or spaced-out g-o-b-l-o-k (caught with the opt-in evasion policy), to slip past a filter. Your team also needs to let a name like Baju Bodo through, treat a mild word differently from a severe one, and send doubtful cases to a person. Indonesian Badwords does all of that, and it reports what it found instead of guessing what the writer meant.

npm install indonesian-badwords

import { Badwords } from 'indonesian-badwords';

const checker = new Badwords();

checker.flag('terima kasih'); // => false
checker.flag('dasar g0bl0k!'); // => true
checker.censor('dasar g0bl0k!'); // => 'dasar ***!'

Requires Node.js 22+ (browser bundles need ES2022 and Intl.Segmenter). ESM, CommonJS and TypeScript types are included, with zero runtime dependencies.

What it can do

Capability	How to use it
Whole-word matching with Unicode, case, width and invisible-character normalization, plus leetspeak (g0bl0k)	On by default (mode: 'word', normalization: 'standard')
A conservative default list, an expanded list, and your own words	dictionary, expandedDictionary, new Badwords({ dictionary: [...] })
Reviewed lists, selected by category, severity and region	selectDictionary() from indonesian-badwords/metadata
How each match was found: exact, normalized, leetspeak, evasion or fuzzy	via on every item from find() and analyze().locations
Category, editorial severity, region, and a 0-100 score with a level	classify() from indonesian-badwords/metadata
Inserted separators, stretched letters and lookalike characters (opt-in, bounded)	boundedEvasion() from indonesian-badwords/evasion
Typos and swapped letters (opt-in, bounded)	boundedFuzzy() from indonesian-badwords/fuzzy
Exceptions at the narrowest scope: a phrase, or a range in one call	allowPhrases option, ignoreRanges call option
Censoring with ***, mask(), grawlix() or your own callback	indonesian-badwords/censor
Exact UTF-16 offsets into your original text, for highlights	find(), analyze()
Precision, recall and false positives on your own labelled examples, and a before/after comparison of results	evaluate(), compare() from indonesian-badwords/evaluate
Typed errors and finite work limits on untrusted input	maxMatches and the limits of each policy

Methods: flag, find, badwords, censor, filter, analyze. Evasion and fuzzy matching are off unless you pass a policy.

Principles

• Correctness first. A false positive is a bug. The default list leaves out words with everyday meanings, such as anjing and babi; broader matching is something you opt into.
• No guessing from context. The library says what matched and how. Your app decides what to do about it. A score rates the matched words, never the intent of a message.
• Control at the narrowest scope. Change one word, one phrase, one category or one call, not the whole checker.
• Reliable on untrusted input. Results are deterministic, work is bounded, and a limit throws a typed error instead of quietly returning "clean".
• Fast. Chat-sized checks take microseconds, and optional features never grow the default import.

Recipes

Censor exact and leetspeak matches, queue fuzzy ones for review

Every occurrence carries via, so one pass can treat sure matches and guesses differently.

import { Badwords } from 'indonesian-badwords';
import { boundedFuzzy } from 'indonesian-badwords/fuzzy';

const checker = new Badwords({ dictionary: ['goblok'], fuzzy: boundedFuzzy() });
const review: string[] = [];

const output = checker.censor('goblok g0bl0k goblox', (context) => {
  const kinds = context.occurrences().map(({ via }) => via.kind);
  if (kinds.every((kind) => kind === 'fuzzy')) {
    review.push(context.match);
    return context.match;
  }
  return '***';
});

output; // => '*** *** goblox'
review; // => ['goblox']

Classify a message and decide what to do

classify() adds category, severity and region to each match and summarizes the message. level is none, unrated, mild, moderate, strong or extreme. A level of unrated (with a score of null) means words matched but have no severity, such as custom words. It is never clean.

import { Badwords } from 'indonesian-badwords';
import { classify } from 'indonesian-badwords/metadata';

const checker = new Badwords();

function decide(message: string): 'allow' | 'mask' | 'review' | 'block' {
  const result = classify(checker.analyze(message));
  if (result.level === 'none') return 'allow';
  // Matched but unrated is never clean: a person looks at it.
  if (result.level === 'unrated' || result.summary.unrated + result.summary.unclassified > 0) return 'review';
  return (result.score ?? 0) >= 60 ? 'block' : 'mask';
}

decide('terima kasih'); // => 'allow'
decide('dasar g0bl0k!'); // => 'mask'
decide('dasar gebleg'); // => 'review'

How the score and level are defined

Choose your vocabulary

import { Badwords } from 'indonesian-badwords';
import { selectDictionary } from 'indonesian-badwords/metadata';

const checker = new Badwords({
  dictionary: selectDictionary({ categories: ['insult'], minSeverity: 2 }),
});

checker.flag('dasar goblok'); // => true

Selectors are profiles, categories, regions and minSeverity (1 to 4). Regional vocabulary includes Javanese, such as jancuk, and Sundanese, such as kehed. Severity describes a dictionary entry, not a whole message. Dictionary guide.

Catch disguised spellings and typos

g0bl0k and full-width text already match with no setup. Evasion and fuzzy policies catch more, and also match ordinary words: with maxEdits: 1, bangsa, kampung and kontrol are each one edit from a listed word. Test on text from your app, and send fuzzy matches to review (first recipe) instead of censoring them.

import { Badwords } from 'indonesian-badwords';
import { boundedEvasion, latinLookalikes } from 'indonesian-badwords/evasion';
import { boundedFuzzy } from 'indonesian-badwords/fuzzy';

const tolerant = new Badwords({
  fuzzy: boundedFuzzy({ model: 'osa', maxEdits: 1 }),
  evasion: boundedEvasion({
    separators: { characters: [' ', '.', '_', '-'], maxGap: 1, maxTotal: 8 },
    repetitions: { maxExtraPerLetter: 2, maxTotal: 8 },
    confusables: { mappings: latinLookalikes, maxTotal: 2 },
  }),
});

tolerant.censor('dasar g-o-b-l-o-x!'); // => 'dasar ***!'

Evasion options · Fuzzy matching

Decide how it looks, and what it leaves alone

import { Badwords } from 'indonesian-badwords';
import { mask } from 'indonesian-badwords/censor';

const plain = new Badwords();
plain.censor('dasar g0bl0k!', mask({ keepFirst: 1, keepLast: 1 })); // => 'dasar g****k!'
plain.censor('dasar g0bl0k!', ({ words }) => `[${words[0]}]`); // => 'dasar [goblok]!'

const checker = new Badwords({ allowPhrases: ['Baju Bodo'] });
checker.censor('Baju Bodo goblok bodo'); // => 'Baju Bodo *** ***'
checker.censor('goblok\nBaju Bodo goblok', '***', { ignoreRanges: [{ index: 0, end: 7 }] }); // => 'goblok\nBaju Bodo ***'

An allowed phrase exempts only itself, never the rest of the message. Censoring · Exceptions

Highlight the original text

index and end are UTF-16 offsets into the text you passed in, and match equals text.slice(index, end), even after normalization.

import { Badwords } from 'indonesian-badwords';

const checker = new Badwords();

checker.find('👩🏽‍💻 g0bl0k!').map(({ word, index, end, match }) => ({ word, index, end, match }));
// [{ word: 'goblok', index: 8, end: 14, match: 'g0bl0k' }]

Speed

Measured with the default checker on Node.js 26.10.0 (AMD Ryzen 7 9700X, Linux; warmed synthetic medians, setup excluded).

1.05 µs Clean ASCII chat check 280 characters

~14 KB Default browser bundle minified + gzip

0 Runtime dependencies

Workload	Operation	Median per call
Clean ASCII paragraph, 4,000 characters	flag	5.17 µs
Clean ASCII paragraph, 4,000 characters	censor	5.13 µs
Accents, emoji, and one literal match, 4,000 UTF-16 units	flag	29.65 µs

Optional policies have their own costs. Run npm run benchmark from the repository to measure your runtime. Methodology and full results.

Documentation

Documentation index · API reference · Dictionaries and metadata · Chat moderation guide · Troubleshooting · llms.txt

Contributing

Missing a word or getting a false positive? Open an issue with the example, the result you expected, and a similar benign case. For dictionary additions, include sources showing the abusive usage. Use the dictionary report form and the contribution guide.

Semoga bermanfaat 😉