LanguageTool checks grammar, punctuation, style and typography across more than 30 languages from a browser extension, Office plugin, desktop app or HTTP API. Its rule engine is open source under LGPL-2.1 and can be self-hosted, while AI-based rules stay cloud-only.
Used this tool? Rate it
Used this tool? Rate it
LanguageTool is a multilingual grammar checker that reviews grammar, spelling, punctuation, casing, style and typography, and that can rewrite individual sentences on request. LanguageTool describes itself as "more than a grammar checker" and, in its own words, as an AI-based grammar checker that you can use by pasting text or typing directly into the editor — no account required for the basic web experience.
The product occupies an unusual position in a crowded category. Most writing assistants are closed commercial services; LanguageTool is a commercial service built on top of a genuinely open codebase, which makes it one of the few products that can honestly be called an open-source grammar checker. The project started in 2003 as an open-source effort, and the company still lists transparency and openness as primary values, pointing to the fact that the core technology is published as open-source software so that anyone can review it, contribute to it, or run their own server. That dual nature is the single most important thing to understand before deciding whether this tool fits you, because the free open-source engine and the paid cloud service are not the same product with different quotas — they have materially different capabilities, and the gap is documented by the vendor itself.
The second thing worth knowing up front is corporate. European tech publication Tech.eu reported in April 2023 that the German-based LanguageTool had been acquired by US-headquartered Learneo, with financial details undisclosed, and that Learneo then comprised seven business lines including CliffsNotes, Course Hero, LitCharts, QuillBot, Scribbr and Symbolab. Accordingly, the Terms and Conditions, effective 21 August 2026, state that LanguageTool is owned and operated by Learneo, Inc., a Delaware corporation with registration number 4243650 and a San Francisco mailing address. At the same time, the Chrome Web Store still lists LanguageTooler GmbH in Potsdam, Germany as the extension's publisher, complete with a D-U-N-S number. Both facts are current and neither cancels the other: the engineering and extension-publishing entity remains German, while the legal owner and data controller is American. If your evaluation hinges on "this is a German, GDPR-native alternative to US writing tools", that framing needs adjusting — the reality is more layered, and the details matter for procurement.
For scale, the About page publishes six vendor-reported figures: 2003 as the founding year, 30+ supported languages, 2M+ active browser add-on installations, 143 contributors to the technology, 150+ companies using the product, and 20M+ texts improved every day. None of these numbers carry an independent audit, so treat them as the company's own accounting rather than verified market data. The one number that can be cross-checked from outside is installation reach: the Chrome extension holds a 4.7 rating from 12.6K ratings with more than 2,000,000 users, and was last updated on 12 August 2026 at version 11.2.3 — which independently corroborates the browser add-on figure, at least on Chrome alone.
The homepage groups the product into six advertised capabilities: correcting spelling, checking grammar, fixing punctuation, confirming casing, improving style, and an AI paraphrasing tool that rewrites sentences to be more formal, fluent, simple or concise. That top-level summary is accurate but coarse. The plan comparison table is far more useful, because it enumerates the individual detection classes and marks which ones are gated behind payment.
Concretely, the comparison table splits checks into five groups — grammar and punctuation, style, semantics, format and typography, and spelling — with the semantics group covering incorrect dates, lack of clarity, word confusion, double negation and even the wrong recipient name in an e-mail. That last item is worth pausing on: detecting that the salutation of your message names a person who does not match the recipient is a genuinely specific, high-value check that generic spell checkers do not attempt.
The format and typography group is similarly concrete: inconsistent use of numbers and letters, incorrect time and date formats, inconsistent spacing, and incorrect currency formats. The spelling group includes ISBN and IBAN validation — the tool will tell you that an IBAN beginning with "GB" should be 22 characters and yours has 21. These are checks aimed at business correspondence and published documents, not at essay drafts.
It helps to picture the product as two layers. The lower layer is a rule engine that has been accumulating since 2003, expressed largely as XML rules per language. The upper layer is a set of AI-based capabilities added later: the paraphrasing tool, and what the vendor calls AI-based rules.
The distinction is not academic, because the two layers ship differently. The official server documentation is blunt about it: running it yourself "will give you a basic LanguageTool server without AI-based rules", because the AI-based rules are only available in the cloud, and the same page warns that the route is only meant for advanced users comfortable with the command line. So if you self-host to keep text off third-party servers, you keep the rule engine and lose the AI layer. There is no configuration that gets you both.
Premium adds more than 20,000 additional checks, but only for English, German, French, Spanish, Dutch, Polish and Portuguese. Read that list carefully — it is seven languages out of a claimed thirty-plus. If you write mainly in Italian, Swedish, Greek or Ukrainian, the Premium upgrade buys you the larger character limit and unlimited rewriting, but not the twenty thousand extra checks, because those were never built for your language.
Paraphrasing is a first-class product line rather than a bolt-on: it has its own page, its own navigation entry, and it is the sharpest dividing line between the free and paid tiers. The free tier allows three AI sentence rewrites per day, which is enough to evaluate the feature and not enough to use it in a workflow.
For teams that want the engine rather than the interface, the proofreading API exposes the same engine that powers Premium, caps each request at 60,000 characters, allows specific grammar or style rules to be disabled, and sells access in daily call tiers of 100, 250, 500, 1,000 and 10,000. The ability to disable individual rules matters more than it sounds: house styles frequently conflict with a general-purpose checker's defaults, and being able to switch off a specific rule is the difference between an integration people keep and one they mute entirely. The personal dictionary is also reachable through the API, so custom terminology survives the move from the web editor to your own pipeline.
The open-source core is published under LGPL-2.1 or later, and the repository — which describes itself as a style and grammar checker for 25+ languages — carries roughly 14.8k stars, 1.6k forks and more than 82,000 commits. Eighty-two thousand commits against a rule-based linguistics engine is a meaningful signal: this is not a codebase that appeared alongside the current AI cycle. It also means the licensing question has a clear answer. LGPL-2.1 permits commercial use and linking, with obligations that attach if you modify the library itself.
The scenario the feature set fits best is a professional writing in a language that is not their first, across e-mail and web forms. The semantics checks — wrong recipient name, incorrect dates, double negation — target exactly the errors that damage credibility in business messages, and the formatting checks catch currency and date-format slips that betray a document assembled from multiple sources.
The 150,000-character ceiling on a single text field puts full reports, dissertation chapters and book sections inside a single check. This is the concrete reason long-form writers upgrade: the free ceiling of 2,000 characters forces you to paste a long document in dozens of fragments, and fragmenting a document breaks exactly the cross-sentence consistency checks you wanted.
Most real usage happens through the extension, not the website. The point is that text never moves: you write in Gmail, Google Docs or your CMS, and corrections appear in place.
The team tier's value is not extra checking power but shared vocabulary. A team style guide and a shared dictionary mean that a product name, a spelling convention or a banned phrase is enforced identically for everyone, instead of each writer maintaining private exceptions.
For organisations that cannot send drafts to a third-party cloud, the self-hosted server is the reason LanguageTool appears on shortlists where Grammarly cannot. The trade-off is explicit and documented, and the sober version of this use case is: self-host when the requirement is data residency, not when the requirement is best-in-class suggestions.
The API tier suits teams building writing quality into their own application — a CMS that flags issues before publication, a support console that checks replies before they are sent, or a localisation pipeline that lints translated strings.
Open the site, paste or type into the editor, and corrections appear as you write. No registration is needed for the basic web experience, which makes it unusually cheap to evaluate — you can judge suggestion quality in your actual language on your actual text before creating an account.
English alone ships in Australian, Canadian, GB, New Zealand, South African and US variants; German has Austria, Germany and Swiss; Portuguese covers Brazil, Portugal, Angola and Mozambique. Picking the variant rather than the bare language is what turns a stream of false flags into useful output.
Coverage spans five browser extensions (Chrome, Edge, Firefox, Safari, Opera), e-mail add-ons for Gmail, Apple Mail and Thunderbird, office plugins for Google Docs, Word, Apple Pages and LibreOffice, plus native desktop apps for macOS and Windows. Install the one matching your primary writing surface first.
Every product name, surname and domain-specific term you use will be flagged until you add it to the personal dictionary. Spending ten minutes on this at the start removes the majority of ongoing noise.
Picky Mode raises strictness and surfaces stylistic suggestions the default profile suppresses. It is limited on the free tier. Treat it as a mode you switch on for final polish rather than one you leave on while drafting.
If self-hosting is on the table, test it before paying, because what you get is a different product. Building from source requires Java 17 and Apache Maven, and a full clone downloads more than 500 MiB while needing more than 1,500 MiB on disk, though a shallow clone brings that down to under 60 MiB and 200 MiB.
Check the rule count for your language before committing. The official rules table is public, and it is the single most useful page for setting expectations. If your language sits at the thin end of the table, no subscription changes that.
Do not buy the quarterly plan. The quarterly tier is the trap in this pricing table: at $49.90 per quarter it works out to $16.63 per month, nearly three times the $5.83 effective monthly rate of the annual plan. If you need a short commitment, monthly at $24.90 costs more per month but far less in total for a brief project.
Test with your real text, in your real language, before paying. Reviews for this product diverge sharply by language, so a colleague's experience in German tells you little about Danish or Italian.
Treat rewriting suggestions as drafts, not authority. Users have reported rewrites that changed meaning, so read every accepted rewrite in context.
Use the API's rule-disabling capability early. If you integrate the API, curate the enabled rule set in the first week.
Remember the fourteen-day window. Every Premium package carries a 14-day money-back guarantee, subscriptions can be cancelled at any time and simply end when the current billing period expires, and one account covers all apps and extensions across devices at no extra cost. Do your evaluation inside that window.
Non-native professionals writing in a major European language. The strongest fit. If you work in English, German, French, Spanish, Dutch, Polish or Portuguese, you are inside the set that receives the twenty thousand extra Premium checks.
Multilingual writers who switch languages daily. A single subscription covering all languages and dialects is materially cheaper than several single-language tools.
Privacy-constrained and open-source-committed teams. If self-hosting or licence auditability is a requirement, this is one of very few credible options.
Developers embedding checks in a product. The API tier serves this cleanly.
Long-document writers. The 150,000-character ceiling is the deciding feature.
Who should look elsewhere: writers working primarily in a thinly-supported language; anyone expecting the self-hosted build to match the cloud; and users who want an AI writing partner that drafts and restructures rather than a checker that corrects.
The reach is broad and the browser extension is the centre of gravity. On Chrome specifically, the extension holds 4.7 stars from 12.6K ratings across more than 2,000,000 users, at version 11.2.3 updated 12 August 2026, weighing 14.95 MiB, with an interface localised into twelve languages.
By default the extension sends your text to languagetool.org over an encrypted connection, needs no account, and the listing states that IP addresses are not stored. Desktop apps exist for macOS and Windows. Note one asymmetry: the Windows desktop app does not work against a self-hosted server.
Self-hosting also costs you three smaller things: language detection depends on fastText, which is unavailable on Windows and degrades detection there; the self-hosted server does not support synonyms, so that feature in the add-on stops working; and the Homebrew path is community-maintained rather than officially supported. Docker images exist but are community-contributed rather than official.
At the time of writing, individual Premium was priced at $119.80 for two years ($4.99 per month), $69.90 annually ($5.83 per month, flagged as the bestseller), $49.90 quarterly ($16.63 per month) and $24.90 monthly, with all prices marked as including VAT. A limited-time discount banner was running during this check, so treat these as a snapshot rather than a permanent rate card.
For groups, Teams starts at two seats for $132.80 per year, discounted from $139.80, which works out to $66.40 per user per year, and layers user management, a team style guide and a team dictionary on top of everything in Premium.
The free tier is usable rather than ornamental, but the boundaries are firm: the free tier caps a single text field at 2,000 characters against Premium's 150,000, restricts Picky Mode, and allows only three AI sentence rewrites per day.
API pricing is sold separately in daily-call tiers and is not covered by a Premium subscription. During this review the API page's price figures did not render, so no API prices are quoted here.
Grammarly is the direct comparison for English. It is deeper in English and far shallower in everything else, and it is closed-source with no self-hosting path. On Trustpilot, the same page lists Grammarly at 3.2 from 10K reviews.
QuillBot overlaps specifically on rewriting — and it is now a sibling product: LanguageTool and QuillBot now sit under the same corporate roof and overlap directly on sentence rewriting, which is precisely why the privacy policy has to state their training-data positions separately.
ProWritingAid targets long-form fiction and manuscript-level reporting, and is a better fit for novelists than for business correspondence.
Built-in checkers in Word and Google Docs are free and adequate for typos; the reason to add a dedicated tool is the style, semantics and typography layers.
Self-hosted LanguageTool is, unusually, its own alternative — and the honest comparison is against the cloud version of itself.
The official rules table for version 6.6 shows how uneven "30+ languages" really is: Catalan carries 8,251 XML rules, French 6,984, English 6,074 and German 5,224, while Italian has just 141, Greek 55, Khmer 33 and Swedish 32. A tool with 32 rules for Swedish is not a Swedish grammar checker in any meaningful sense.
There is also a counting discrepancy worth flagging: the marketing site says 30+ languages while the official GitHub repository describes itself as a checker for 25+ languages, so the two numbers should be read as different counting conventions rather than a single verified figure.
Trustpilot tells a very different story from the app stores, with a 2.0 "Poor" score across 53 reviews, 45 percent of them one-star, and a platform note that the company has not replied to negative reviews. Set against 4.7 stars from 12.6K Chrome ratings, the contrast is stark — and the sample sizes differ by more than two orders of magnitude. With only 53 reviews, the Trustpilot score is directional rather than conclusive, and the two populations differ: app-store ratings skew toward free extension users, while Trustpilot skews toward people motivated by a billing dispute. Neither number should be read alone.
What the negative reviews say is more informative than the score. The most repeated complaint concerns non-English coverage: an Italian reviewer reported that a missing apostrophe in "Un avventura bellissima" went unflagged, and a Danish reviewer said the tool failed to find even the simplest grammar errors in Danish text. That maps precisely onto the rules table.
A second cluster of complaints targets the AI layer itself, with one reviewer saying that after AI was added the tool now flags a possible mistake in every sentence, and another reporting suggestions that changed second person to first and altered the meaning of sentences.
A third cluster centres on monetisation and billing, including resentment that browser extensions became subscriber-only and warnings about automatic renewal; for context, the same page lists Grammarly at 3.2 from 10K reviews and QuillBot at 4.8 from 14K.
This deserves restating because it is the most common misunderstanding: the open-source build does not include the AI-based rules.
Beyond the 2023 acquisition coverage, we found no independent in-depth product review of LanguageTool from major technology outlets such as TechCrunch, VentureBeat, The Verge or ZDNet; most search hits were syndicated versions of the acquisition press release. That absence is itself a finding.
The developer API page makes a narrower and more specific promise: all servers are hosted in Germany, the service is described as fully GDPR-compliant, and none of the texts sent to the API are stored. That promise is scoped to the API. Meanwhile the business page says only that the company works hard to comply with GDPR and points to a data processing agreement, without publishing an ISO 27001 or SOC 2 certification. If your procurement process requires an audited certification, the public pages do not currently supply one.
The clearest and most favourable commitment is on training data: Learneo states that it does not use inputs from any LanguageTool user, on either the Individual or Team plan, to train its language algorithms or models, and that this is the default setting requiring no opt-out. Written as a default rather than a setting, it is stronger than most competing policies.
The controller question is where expectations often need correcting: the privacy policy names Learneo, Inc. as the data controller, routes EEA and UK requests through an authorised representative in Dublin, and relies on the EU-U.S. and Swiss-U.S. Data Privacy Frameworks for cross-border transfers. Learneo also discloses offices in the United States, India, the Netherlands, Germany and Canada.
At the extension level, the store's data disclosure lists personally identifiable information, personal communications, location, user activity and website content as data the extension handles, alongside a declaration that the data is not sold to third parties. "Personal communications" and "website content" are unavoidable for a tool that reads what you type, but they should be evaluated consciously. The privacy policy also discloses the use of third-party session-replay tooling.
Yes, with real limits. The web editor and extensions work without an account for basic checking. The free tier caps a single text field at 2,000 characters against Premium's 150,000, restricts Picky Mode, and allows only three AI sentence rewrites per day.
At the time of writing, individual Premium was priced at $119.80 for two years ($4.99 per month), $69.90 annually ($5.83 per month, flagged as the bestseller), $49.90 quarterly ($16.63 per month) and $24.90 monthly, with all prices marked as including VAT. A discount promotion was active during this check, so verify current rates on the pricing page.
This is the most important question to get right. The official server documentation is blunt about it: running it yourself "will give you a basic LanguageTool server without AI-based rules", because the AI-based rules are only available in the cloud. You keep the rule engine and lose the AI layer.
No, according to the policy. Learneo states that it does not use inputs from any LanguageTool user, on either the Individual or Team plan, to train its language algorithms or models, and that this is the default setting requiring no opt-out.
Partly. The Terms and Conditions state that LanguageTool is owned and operated by Learneo, Inc., a Delaware corporation, while the Chrome Web Store still lists LanguageTooler GmbH in Potsdam, Germany as the extension's publisher. The engineering entity is German; the legal owner and data controller is American.
Fewer than the headline suggests. The official rules table for version 6.6 shows Catalan at 8,251 XML rules, French 6,984, English 6,074 and German 5,224, while Italian has just 141, Greek 55, Khmer 33 and Swedish 32. Check your language in that table before subscribing.
Yes, and this is where most day-to-day usage actually happens rather than on the website. There are dedicated office plugins covering the major word processors and cloud editors, mail add-ons for the main desktop and webmail clients, and five separate browser builds. The practical advice is to install only the surface you write on most: each additional integration adds another update to maintain without adding checking power, since all of them call the same engine.
Yes. The proofreading API exposes the same engine that powers Premium, caps each request at 60,000 characters, allows specific grammar or style rules to be disabled, and sells access in daily call tiers of 100, 250, 500, 1,000 and 10,000. API access is billed separately from a Premium subscription.
Two things dominate. Non-English coverage: an Italian reviewer reported that a missing apostrophe in "Un avventura bellissima" went unflagged, and a Danish reviewer said the tool failed to find even the simplest grammar errors in Danish text. And the AI layer: one reviewer said that after AI was added the tool now flags a possible mistake in every sentence.
Yes, within a defined window. Every Premium package carries a 14-day money-back guarantee, subscriptions can be cancelled at any time and simply end when the current billing period expires, and one account covers all apps and extensions across devices at no extra cost. Several Trustpilot reviewers reported disputes over automatic renewal, so diary the renewal date.
