Select what you wrote
Highlight a sentence in an email, a form, a post or a chat. A small VerbaClara button appears. Right-click works too — or let Auto mode check when you pause.
VerbaClara 0.9.0 Chrome · Edge · Brave For people who write in a second language
Select a sentence in most places you write. VerbaClara fixes it, explains the rule in the language you think in — offline for common mistakes — and turns the mistakes you keep making into practice. Harper runs on your computer by default.
Free with Harper, Ollama, or your own key No account No VerbaClara servers
New message
Draft savedHi team, I am writing to follow up on the report. I have went1 through the numbers and I think we should discuss about2 the timeline.
Live demo — this is the real panel, rebuilt for the page. Change the language, switch the tone, or apply the fix. Nothing is sent to VerbaClara. Model mode shown. In the extension, the first explanation works offline in seven languages; the chips here are a demo control.
01 How it works
The loop is small on purpose. It fits between two thoughts — not in another tab, not in a twelve-step grammar course.
Highlight a sentence in an email, a form, a post or a chat. A small VerbaClara button appears. Right-click works too — or let Auto mode check when you pause.
Keep it simple with Fix, or shift the register with Formal, Casual or Shorter. Meanings stay yours.
The rule arrives in the language you chose, in plain words. One tap files the mistake for practice.
Works in normal text fields and many rich editors. Google Docs is the honest exception — see coverage.
02 The part everyone skips
Most tools stop at the green check. VerbaClara is built for the moment after it.
You don’t need this sentence fixed. You need the rule that fixes the next one.
VerbaClara keeps the thing you actually wrote: the wrong phrase, the fix, the rule. It names the pattern, explains it in your language, and hands it to the Practice Lab. The next time it shows up in your inbox, you recognise it.
chrome.storage.local.Honest bit: the first explanation is built offline for common grammar categories. Deeper, context-specific explanations need a model. Harper, the default, gives you on-device grammar and the offline first explanation.
Hi team, I am writing to follow up on the report. I have went through the numbers and I think we should discuss about the timeline.
En el presente perfecto, después de have va el participio: gone, no went.
Discuss heißt „besprechen“. About ist überflüssig.
Two corrections. Two rules. Both are saved for practice.
03 Practice Lab
Not a generic grammar course. The Lab clusters the crumbs from your own writing, then quizzes the pattern until it moves from New to Mastered. No history yet? Four foundational clusters get you started.
Detected weakness clusters
Built from crumbs. Never full emails.
We will meet ___ Monday morning.
Pick one. The Lab records mastery after you answer.
04 Choose the brain
One setting decides where your sentence goes. Start fully on your computer. Add a model only for tone, translation and lessons.
Runs on this computer
Harper is a grammar engine compiled to WebAssembly inside the extension. It checks English spelling and grammar, applies its own suggestions, and files the fixes it applies into one of nine learning categories. Fix mode only.
Runs on your machine
Run llama3.1 or qwen2.5:7b in Ollama, point VerbaClara at localhost, and get tone rewrites, deeper translated explanations and Lab lessons without a cloud key.
OLLAMA_ORIGINS="*" so the extension can reach it.Your key, your call
Paste your own OpenAI key. The model list fills from the key itself, so you never type a model name. For fix requests, emails, phone numbers, URLs and long digit strings are masked before the call.
Your key, your call
Bring a Google AI Studio key and pick from the models it unlocks. Same fix-request masking, same local history, same promise: no VerbaClara account in the middle.
05 Privacy
The free tier has no backend and no account. For fixes, your sentence goes to Harper on your computer, to a model on your localhost, or to the provider whose key you pasted. Practice Lab sends only short crumbs to the lesson model you choose. That is the whole map.
Hi Priya, send the signed copy to [EMAIL] or call [PHONE]. The brief is at [URL] — ref [NUMBER].
Hi Priya, send the signed copy to priya@northwind.co or call +385 91 555 0199. The brief is at https://northwind.co/brief/2291 — ref 448210.
For fix and rewrite requests, emails, phone numbers, URLs and long digit strings become placeholders before the call, then are restored. Practice Lab uses a separate lesson model and sends only short crumbs. Names and context are not masked.
There is no VerbaClara account, no sync and no analytics SDK. The extension ships with network access only for localhost and the AI provider domains in its manifest.
History stores the wrong phrase, the correction, the rule label and the category — capped at 500 entries in chrome.storage.local. Each side is limited to 80 characters; full messages and documents are never saved.
Local-only mode blocks every non-localhost endpoint, including a cloud lesson model. Harper and a local Ollama model already keep fix text on your machine; the switch closes the rest.
Read the full privacy policy No cookies on this page. No analytics. No VerbaClara servers.
06 Honest coverage
Some editors accept inserted text. Some intercept it. A few block it entirely. Where the Fix UI runs, VerbaClara offers Copy when Replace fails — and tells you before you find out mid-email.
The clean path: the corrected sentence goes straight back into the field where you wrote it.
These editors often intercept programmatic edits. VerbaClara hands you the corrected text and a one-key copy instead of pretending.
Docs blocks selection tools; cross-origin iframes never receive the injected UI. On Docs you get a clear message and Copy when the editor permits it. In an iframe, use the browser’s own copy/paste — the extension cannot reach it.
Copy is the fallback. We’d rather hand you the sentence — and admit when a page blocks that too — than claim a guarantee the page can’t keep.
07 Price
The whole coaching loop is free with Ollama or your own key. Hosted AI is planned for people who want zero setup — never as a gate on lessons.
The core works offline. Add Ollama or your own key for deeper explanations, tone rewrites and Lab lessons.
No card. No trial. No “free plan” asterisk.
For people who want zero setup: we run the model. The learning features stay free.
Price announced when it ships. No lifetime deal, no launch discount.
08 Install
The Chrome Web Store listing is next, not live. Until it is, the unpacked build below is the real extension — same code, version 0.9.0.
Unpacked install · Chrome, Edge, Brave
chrome://extensions and turn on Developer mode.The Store build will come from this same source. When the listing goes live, this page will switch to the one-click link.
09 FAQ
The honest ones first. If something is missing, the privacy policy and the build are both linked at the bottom.
No. Grammarly corrects. VerbaClara corrects and then teaches: the rule in your language, and practice built from your own mistakes. It also isn’t a subscription: the core is free, and local Ollama or your own key unlocks deeper explanations without a VerbaClara server.
Not to start. Harper runs inside the extension on your computer — no key, account or card. The first explanation of common grammar mistakes is built offline in your language. For deeper explanations, tone rewrites and Practice Lab lessons, choose a local Ollama model or paste your own OpenAI or Google key.
Start with Harper for English grammar and the offline explanations. A model adds context-specific explanations in Spanish, German, Japanese or another language. Locally, llama3.1 is a good default and qwen2.5:7b is stronger for multilingual work. The build also supports the OpenAI-compatible servers listed in the engines section.
With Harper or Ollama, a fix or rewrite stays on your computer. With a cloud key, the selected sentence goes to the endpoint you configured after emails, phone numbers, URLs and long numbers are masked.
Practice Lab is separate: it uses the lesson model you choose — local, OpenAI or Google — and sends only short wrong → correct crumbs, never the original email. Masking covers fix and rewrite requests; lesson crumbs are short but not PII-masked. Names and context are not masked on either path, so use Ollama or Local-only if that matters.
Gmail and Slack often intercept in-place edits, so VerbaClara uses the Copy fallback there. Google Docs blocks selection tools almost entirely, and cross-origin iframes never get the injected UI. The coverage section is the honest map; Copy works wherever the editor permits it.
Harper is English-only. The offline explanation pack covers common categories in seven languages; a model can check other target languages and explain deeper context. The interface is English. Set “I write in” and “Explain my mistakes in” separately in Settings.
It is stored in chrome.storage.local on your device and sent only to the endpoint you configured. No VerbaClara server ever sees it. For the strongest guarantee, use Ollama or turn on Local-only mode, which blocks every non-localhost endpoint.
Submission is the next step; right now you can install the unpacked build from the zip above. It is the same code. Hosted Pro ships only after the listing has real reviews — not before.
It is on the roadmap after Chrome. VerbaClara is Manifest V3 today, so Firefox is a port — not a promise with a date attached.