A tiny multilingual spell checker with correction suggestions. Dictionaries are bundled, so npm i fixnow gives you everything — with zero runtime dependencies, in both ESM and CommonJS.

Features

• 📦 Zero Dependencies — Keeps your node_modules clean and lightweight.
• 🌍 Built-in Dictionaries — Includes Arabic, German, English, Spanish, French, Portuguese, Russian, and Vietnamese.
• ⚡ Slim Builds — Import only the language you need (e.g. import { check } from "fixnow/es") to optimize bundle size.
• 🛡️ Smart Tokenization — Automatically ignores code spans, URLs, emails, and identifiers to prevent false positives.
• 🧩 Universal — Works seamlessly in both ESM and CommonJS projects.

Architecture

flowchart LR
    Input["Input Text"] --> Tokenizer
    Tokenizer --> |"Skips Code / URLs"| Check{"Dictionary"}
    Check -->|"Valid Word"| Pass["No Issues"]
    Check -->|"Typo Detected"| Engine("Correction Engine")
    Engine --> Issues["SpellIssue[]"]

    style Input fill:#18181b,stroke:#ef4444,stroke-width:2px,color:#fff
    style Issues fill:#18181b,stroke:#ef4444,stroke-width:2px,color:#fff

Install

npm i fixnow

Languages

| Code | Language | Dictionary license | | ---- | ---------- | ------------------ | | ar | Arabic | LGPL-3.0 | | de | German | LGPL-3.0 | | en | English | MIT | | es | Spanish | LGPL-3.0 | | fr | French | MIT | | pt | Portuguese | GPL-3.0-or-later | | ru | Russian | GPL-3.0-or-later | | vi | Vietnamese | MIT |

Usage

import { checkText, suggest, createChecker } from "fixnow";

// English
const enIssues = await checkText("This sentance has a typo", {
  language: "en",
  suggestions: true,
});
// -> [{ offset: 5, length: 8, word: 'sentance', suggestions: [...] }]

// Spanish — opt in to accent leniency if you don't want "codigo" flagged.
const esIssues = await checkText("Esto es un herror", {
  language: "es",
  suggestions: true,
  acceptAccentOmissions: true,
});
// -> [{ offset: 11, length: 6, word: 'herror', suggestions: [...] }]

// One-off correction suggestions
await suggest("bonjoor", { language: "fr" }); // -> ['bonjour', ...]

// A checker bound to one language
const de = createChecker("de");
await de.isCorrect("Haus"); // -> true

CommonJS works too:

const { checkText } = require("fixnow");

API

• checkText(text, options) → Promise<SpellIssue[]>
• isCorrect(word, language, options?) → Promise<boolean>
• suggest(word, { language, max? }) → Promise<string[]>
• createChecker(language) → bound { check, suggest, isCorrect, warmup }
• warmup(language?) — preload dictionaries (skip first-call decode cost)
• tokenize(text, protectedSegments?), DEFAULT_PROTECTED_PATTERN
• SUPPORTED_LANGUAGES, LANGUAGES, isSupportedLanguage

CheckOptions: language (required), caseSensitive (false), acceptAccentOmissions (false; Spanish only), suggestions, maxSuggestions (5), minWordLength (3), ignoreWords, flagWords, isProtectedWord, protectedSegments.

Tokenization

checkText skips anything inside a "protected segment" (code spans, URLs, emails, paths, CLI flags, hex colors, ACRONYMS, file names and dotted identifiers). Override the patterns with protectedSegments:

import { checkText, DEFAULT_PROTECTED_PATTERN } from "fixnow";

// Use only your own pattern
await checkText(text, { language: "en", protectedSegments: /\{\{[^}]+\}\}/g });

// Compose with the default
await checkText(text, {
  language: "en",
  protectedSegments: [DEFAULT_PROTECTED_PATTERN, /\{\{[^}]+\}\}/g],
});

// Disable protection entirely
await checkText(text, { language: "en", protectedSegments: false });

The same option is exposed on tokenize(text, protectedSegments).

Slim Builds

If you only need one language, import it via the language subpath. Your bundler only copies the dictionary you actually use:

import { check, suggest } from "fixnow/es";

const issues = await check("Esto es un herror", { suggestions: true });
await suggest("bonjoor", 3); // bound suggest is (word, max?)

The slim entries (fixnow/ar, fixnow/de, fixnow/en, fixnow/es, fixnow/fr, fixnow/pt, fixnow/ru, fixnow/vi) re-export a checker pre-bound to that language.

Bundling

fixnow reads its dictionaries from disk at runtime — they ship as files under node_modules/fixnow/dictionaries/, not as inlined bytes in the JS. So any bundler must treat fixnow as external, leaving it to load from node_modules at runtime. This is required for VS Code extensions and any CJS bundle: inlining fixnow into a CJS output strips the path anchor it uses to find its dictionaries, and it will throw a clear "mark 'fixnow' as external" error instead of resolving them.

// esbuild
await esbuild.build({
  entryPoints: ["src/extension.ts"],
  bundle: true,
  format: "cjs",
  platform: "node",
  external: ["fixnow"],
});

The matching option for other bundlers:

• Vite — build.rollupOptions.external: ['fixnow']
• Rollup — external: ['fixnow']
• webpack — externals: { fixnow: 'commonjs fixnow' }

Migrating from 1.x

2.0.0 cleans up three rough edges from the extraction-from-F1 release. Each is a breaking change:

• language is now required. There is no default language anymore.

  // before
  await checkText("hola"); // implicitly Spanish
  // after
  await checkText("hola", { language: "es" });

• strict is split into caseSensitive and acceptAccentOmissions. The new default is strict (the old strict: true). If you relied on strict: false to tolerate Spanish accent omissions, opt in explicitly:

  // before
  await checkText("codigo", { language: "es" }); // accepted
  // after
  await checkText("codigo", { language: "es", acceptAccentOmissions: true });

  The legacy strict key still works in 2.x with a console.warn; it is removed in 3.0.0.
• F1-specific markers are gone from the default tokenizer. [Image #1], [Skills #…], /skills #N, and /skill no longer auto-skip. If you need them, pass them via protectedSegments:

  const F1_MARKERS =
    /\[(?:Image|Code|Text) #\d+[^\]\n]*\]|\[Skills? #[^\]\n]+\]|\/skills #\d+|\/skill\b/g;
  await checkText(text, {
    language: "en",
    protectedSegments: [DEFAULT_PROTECTED_PATTERN, F1_MARKERS],
  });

License

MIT