Spelling & grammar checker

Spelling, grammar and punctuation fixed in one click. Free, no upload. 1 GB max · Up to 1 GB · Processed locally, never sent to a server.

This page corrects a text, and what sets it apart from the other tools in this section has to come first: in the normal case, your text leaves your browser. When you press the button, the content of the box is sent as a POST request to api.languagetool.org, the public server of the open-source checker LanguageTool, which returns the list of errors it detected and their suggestions.

Those suggestions are not put to you for approval: the page applies the first proposal for each error automatically, rebuilds the corrected text and displays the list of replacements underneath, each with the rule category that triggered it. What comes back is a version that has already been modified, and that has to be re-read.

If the call fails — no connection, blocked domain, server error, or a reply taking more than nine seconds — a fallback corrector built into the page takes over and handles common misspellings and typography. The label shown above the list says which of the two engines did the work: it is your only indicator of whether the text left the browser or not.

How to use it

Paste your text The online check covers the first 19,000 characters, roughly three thousand words; the rest is handed back as it was, with no warning and no cut mark.
Click Run The button switches to a waiting label while the request is out. After nine seconds without a reply the request is abandoned and the local corrector takes over.
Read the engine label It either credits LanguageTool, meaning the text was submitted to the external service, or states that the correction was done inside the page, with far thinner coverage.
Go through the list of replacements Each line shows the original fragment, its correction and the rule category. Forty lines at most are displayed, with the exact total in the heading above them.
Copy, then re-read The copy button places the corrected text on the clipboard. The input box is left untouched, so the two versions can be compared side by side.

What leaves your browser, and where it goes

The request goes to api.languagetool.org and carries exactly two fields: your text, truncated to 19,000 characters, and a language code. The page adds no file name, no account identifier and no reference to the site. Like any web request, though, it exposes your IP address and the usual browser headers to the receiving server.

That server does not belong to NexavoTools. It is the public endpoint of the LanguageTool project, run by its publisher and subject to its own privacy policy. What it logs, and for how long, therefore does not depend on this site, which can make no commitment in its place.

The consequence is direct for a confidential document. A contract clause under negotiation, minutes covered by professional privilege, a medical file or a thesis chapter under embargo have no business being pasted here as they stand: anonymise the names and the amounts, or give up the online check. The four other tools in this section — the word counter, the case converter, the Lorem Ipsum generator and the text comparison — issue no network request at all. This one is the only one of the set that does.

Forcing the local corrector

The interface offers no switch between the two engines: the built-in corrector only comes into play when the call fails. Cutting the connection before pressing the button, or blocking the api.languagetool.org domain, is what triggers the fallback.

That fallback rests on a fixed dictionary of 76 entries, 73 of which actually change something. Replacements apply to whole words, ignore case, and preserve an initial capital. A word written entirely in capitals and at least two letters long is left intact, which keeps acronyms from being turned into ordinary words. The list itself targets common French misspellings, forms such as “acceuil”, “developement” or “parmis”, so it has nothing to find in an English text.

A typography pass comes on top: double spaces collapsed, a stray space removed before a comma, a missing space added after punctuation, a non-breaking space inserted before the semicolon, the colon, the question mark and the closing guillemet, three dots converted into a single ellipsis character, and a capital restored at the start of a sentence. Those spacing rules follow French convention, and the space inserted is the ordinary non-breaking space U+00A0, not the thin space that careful typography reserves for high punctuation. Offline, in other words, this page tidies spacing and does very little beyond that.

An automatic correction has to be re-read

No validation step is offered: the first suggestion is applied, whatever it happens to be. On a proper noun, a brand, a term of art or a quotation, the replacement can degrade the text; since your original version stays intact in the top box, compare the two before publishing anything. The fallback corrector calls for the same care, a fixed list of substitutions having no way of telling a mistake from a deliberate choice, and applying the same rule to a technical term as to a typo.

What this page does not do

LanguageTool itself covers dozens of languages, English among them, but this page does not detect the language of your text: the request carries one language code, fixed in the page, and the server applies the rules of that language and no other. A text written in another language therefore comes back with absurd corrections rather than an error message. Beyond that, the corrector judges neither structure, nor style, nor repetition, and it opens no file: a .docx or a PDF has to be turned into text before being pasted.

Frequently asked questions

Where exactly does my text go when I run the check?

To api.languagetool.org, as a POST request, with two fields: the text truncated to 19,000 characters and a language code. That server belongs to the LanguageTool project, not to NexavoTools, and the project’s own privacy policy applies to it.

Can I check a sensitive text without it leaving the page?

Only by making the call fail, since no setting allows it: cut the connection or block the domain before pressing the button. The local corrector then takes over, limited to a fixed list of substitutions and to typography.

What happens to the part of the text beyond 19,000 characters?

It is handed back word for word, unchecked. Nothing on screen points this out, which is why long documents are better split up by hand before being pasted.

Does it work on languages other than English?

LanguageTool covers dozens of languages, but this page sends one fixed language code with every request and never detects the language of your text. Anything written in a different language is measured against the wrong rule set and comes back with meaningless replacements.

Why was a correct sentence modified?

Because the first suggestion for each detected error is applied without arbitration. Proper nouns, jargon and deliberate turns of phrase are the main casualties. Compare with the original, which stays in the top box, and undo whatever went wrong.

Does the fallback corrector catch grammar mistakes?

No, it runs no grammatical analysis: it applies a fixed list of known misspellings and a set of typography rules. A wrong verb agreement goes straight through. It also skips any word written entirely in capitals of at least two letters, on the assumption that it is an acronym.

Similar tools

Discover other tools

JPG → PDF — Combine images into a single PDF.
Redirect checker — See where a URL ends up and whether it redirects.
Rephrase text — Shorter, more formal or simpler version.
PDF → Word — Convert PDF text into an editable .docx document.