English Grammar Checker (Free Spelling & Grammar Check)
Paste any English text and instantly see common mistakes explained in plain English. No length limit, and nothing ever leaves your device.
Loading the tool...
How to use it
- Paste or type your English text into the box. There is no length limit.
- Common mistakes (articles, agreement, much/many, since/for, make/do, and more) and spelling issues appear right below, each with a short explanation of why it's flagged and a suggested fix.
- If you want a more natural rewrite, click "Polish with AI". The first click downloads a small AI model (the size is shown before you commit), and after that each sentence gets a before/after comparison you can read through.
What kinds of mistakes does it catch
This checker was built around the mistakes that English learners, and Korean speakers in particular, run into most often. Layer 1 is a set of JavaScript rules that run instantly and never contact a server; everything happens inside your browser. It looks for things like:
- Articles a/an: using "a" before a vowel sound or "an" before a consonant sound by mistake.
- Subject-verb agreement: missing -(e)s on the verb after a third-person singular subject like he/she/it.
- There is / there are: mismatches between the verb and the number of the noun that follows.
- Much vs. many, a few vs. a little: confusing countable and uncountable nouns.
- Since vs. for: confusing a specific point in time with a duration.
- Make vs. do collocations: fixed expressions like "make a mistake" or "do homework" used with the wrong verb.
- Double negatives: sentences like "don't... nothing" that use two negatives where English uses one.
- its/it's, your/you're, then/than, affect/effect: classic sound-alike word pairs that get swapped.
- Capitalization, spacing, punctuation, repeated words: sentence-start capitals, spacing around commas and periods, and accidental word doubling.
- Commonly misspelled words: a curated list of words English learners frequently misspell, such as receive, separate, and definitely.
Because Layer 1 relies on hand-written rules and fixed word lists rather than deep language understanding, it can miss things that require real context, such as subject-verb agreement with a general noun subject rather than a pronoun, nuanced preposition choice, or phrasing that is grammatically correct but sounds awkward. This tool is not a substitute for a professional editor or a full-featured paid grammar checker, and it does not catch every possible error. For anything important, we recommend having a person review your writing as well.
Honest limits compared to Grammarly and similar tools
Paid, full-featured grammar checkers such as Grammarly use large language models and years of training data to catch subtle style issues, tone problems, and a very wide range of grammar rules, including ones that depend on deep contextual understanding of a whole paragraph. This free tool takes a lighter, more transparent approach: Layer 1 only checks for a focused set of well-known mistake patterns and a curated spelling list, so it is fast, free, and fully private, but it is not as comprehensive. Layer 2 (the optional AI polishing step) adds a small open-source language model that can rewrite a sentence more naturally, but a 0.5-billion-parameter model is much smaller than the proprietary models behind commercial tools, so its suggestions, while often helpful, can occasionally be imperfect or miss subtler issues too. We'd rather tell you this clearly than overstate what a free, privacy-first, in-browser tool can do.
AI sentence polishing (optional Layer 2)
If the rule-based check isn't enough, click "Polish with AI". This feature runs an open-source small language model, Qwen2.5-0.5B-Instruct, directly in your browser to rewrite a sentence more naturally and show you a before/after comparison. Because the model files are large (roughly 300-550MB), they are not downloaded automatically; you control the download by clicking the button, and the size is shown beforehand. On computers without WebGPU support, the tool automatically falls back to CPU-based WASM execution. On very old devices or browsers this feature may still fail to load, and in that case the app fails softly: Layer 1's rule-based results stay fully usable and an error is recorded internally for debugging, but nothing ever breaks the page.
Open-source credits
- Qwen2.5-0.5B-Instruct (Alibaba Cloud's Qwen team, Apache License 2.0) — the small language model used for the optional "Polish with AI" rewriting step.
- Transformers.js (Hugging Face, Apache License 2.0) — the library used to load and run the AI model in the browser.
- ONNX Runtime Web (Microsoft, MIT License) — the inference engine that runs the model via WASM or WebGPU.
- The Layer 1 grammar rules and the commonly-misspelled-word list were written and curated by Dagochim.
Privacy
Your text is never uploaded anywhere. The rule-based check (Layer 1) runs entirely as JavaScript in your browser with zero network requests. The optional AI step (Layer 2) downloads model files once and then runs inference locally in your browser as well; your sentences are never sent to any server during that process either.
Frequently asked questions
How is this different from Grammarly?
This checker is free, requires no sign-up, and never sends your text to a server for the rule-based layer; everything runs in your browser. It does not catch every grammar issue the way a paid service like Grammarly does. Instead it focuses on specific patterns that English learners (especially Korean speakers) commonly get wrong, like articles, subject-verb agreement and prepositions, plus a curated list of commonly misspelled words. For more natural rewriting, an optional small AI model is available.
Is my text sent to a server?
The rule-based (Layer 1) check never sends anything anywhere; it runs entirely in JavaScript in your browser. If you click "Polish with AI", a small AI model is downloaded and also runs locally in your browser. In both cases your text stays on your device.
Why do I have to click a button for AI polishing?
The AI model files are fairly large (roughly 300-550MB), so downloading them automatically would use a lot of data and time. The rule-based check always works instantly without any download, while AI polishing is opt-in via a button so you control when that download happens.
Will AI polishing work on an older computer?
If your browser or GPU doesn't support WebGPU acceleration, the tool automatically falls back to CPU-based WASM execution. On very old browsers or low-memory devices this may still fail, but the rule-based check in Layer 1 always keeps working regardless.
Tools you might also like
Related guide
Learn more in our 20 English Grammar Mistakes Korean Speakers Commonly Make guide.