I built PROOFER - Privacy first Chrome extension that proofreads your texts using Gemma 4

This is a submission for the Gemma 4 Challenge: Build with Gemma 4

What I Built

I built PROOFER, a privacy first browser extension that helps me proofread text right where I am already typing, whether that is an email draft, a Discord message, a social post, or any other editable text field on the web.

Instead of copying text into a separate chatbot tab, PROOFER lets me select the text in place and trigger proofreading from a floating action button, a context menu entry, or a keyboard shortcut. The extension then sends only the selected text to the model, streams back the response, and shows a suggestion popup with a visual diff so I can quickly compare the original and the revised version.

The extension is built around a bring-your-own-model setup. Users configure their own endpoint in the popup, choose a model, and can connect PROOFER to LM Studio, Ollama, or any OpenAI-compatible API. This makes the tool flexible, but more importantly, it keeps the privacy model in the user's hands. If they run Gemma 4 locally, their writing never has to leave their machine.

PROOFER currently supports four editing styles:

• Proofread for typos, grammar, and punctuation fixes
• Rephrase for better clarity and flow
• Professional for a more polished and formal tone
• Shorten for turning longer text into a tighter version

There are also a few practical quality-of-life features built in, such as streaming responses, connection testing inside the popup, and the ability to disable the extension on specific domains or by regex patterns.

A solution to a trivial problem we all face

This project came from a very small but very real annoyance in day to day working.

I often catch myself writing fast on platforms like Discord, X, email clients, and various web forms, then realizing a message sounds off, has a typo, or could be phrased better. The usual fix is awkward: copy the text, open a separate LLM, paste it there, ask for a correction, copy the result back, and then return to the original tab. It is a tiny interruption, but when it happens repeatedly, it breaks flow.

PROOFER is my attempt to remove that friction. The tool brings proofreading directly into the page I am already using. I can select a sentence, ask for a rewrite in one click, review the suggested result, and keep moving. It solves a trivial problem, but it solves it in the exact place where the friction happens.

That is also why privacy mattered so much in this build. A lot of the text we type into browsers is personal, work-related, or simply not something we want to send to a random remote service. By letting users run Gemma 4 through their own local setup, PROOFER keeps the experience lightweight and much more trustworthy.

And, while I was building and testing PROOFER, I found myself using it almost immediately while chatting with my friends on Discord. This made it clear that it was definitely going to help me in future.

Demo

Code

Explore the project's codebase here: GitHub Repository

How I Used Gemma 4

Gemma 4 is the core intelligence behind PROOFER. I designed the extension so the selected text is wrapped in a strict proofreading prompt, paired with the chosen editing style, and then sent to the configured model endpoint. The model returns only the transformed text, which makes it ideal for an inline editing workflow where the result needs to drop directly back into the user's writing flow.

For this project, I chose Gemma 4 E2B. (For powering AI proofreading)

E2B felt like the right fit because PROOFER is meant to be an edge friendly, privacy focused utility. The extension is most useful when it can run against a local model through tools like LM Studio or Ollama, and Gemma 4 E2B is well suited for exactly that kind of setup. It is lightweight enough to be practical on local or self-hosted environments, while still being strong enough for quick proofreading, tone adjustment, and concise rewriting.

That balance matters a lot here. PROOFER is not trying to generate long essays or act like a general purpose assistant. It needs a model that can respond quickly, preserve intent, follow style instructions reliably, and stay efficient enough for real-time use. Gemma 4 E2B gives me that balance, which is why it became the heart of the project.

In short, Gemma 4 E2B made it possible to build a tool that feels useful in everyday writing, while still staying aligned with the project's main goal: fast, local-first AI assistance for proofreading.

Try PROOFER in your browser

If you want to try PROOFER locally, you can clone the repository and load it as an unpacked browser extension.

git clone https://github.com/SamirMishra27/proofer_proofreading_web_extension.git
cd proofer_proofreading_web_extension
npm install
npm run build

After that:

1. Open Chrome or Microsoft Edge.
2. Go to the extensions page.
3. Turn on Developer mode.
4. Click Load unpacked.
5. Select the project folder.

Once the extension is loaded:

1. Click the PROOFER extension icon.
2. Choose your provider: LM Studio, Ollama, or any OpenAI-compatible endpoint.
3. Enter the base URL for your local or self-hosted model server.
4. Add an API key if your endpoint requires one.
5. Enter the Gemma 4 model name you want to use.
6. Use Test Connection to confirm everything is working.

To use it on a page:

1. Open any website with an editable text field.
2. Type something and select the text you want to improve.
3. Trigger PROOFER using the floating button, the right-click context menu, or the keyboard shortcut.
4. Pick a style like Proofread, Rephrase, Professional, or Shorten.
5. Review the streamed suggestion and use it as your improved version.

For the best privacy focused setup, run Gemma 4 locally through LM Studio or Ollama so your text stays on your own machine while you write.