Highlight any text, press a shortcut or right-click, and the improved version lands back in place.
Grammar, tone, translation, formatting, a drafted reply. Powered by your Gemini key or a local LM Studio model.

---

Two apps, one idea

	What it is	Where it works
Browser extension	Chrome and Edge, Manifest V3, built with WXT	Any website: inputs, textareas, rich editors
Desktop app	Electron for Windows with a tray icon	Any application, through global hotkeys or a Compose window

They share the same actions, the same prompts and the same "bring your own AI" model, and neither depends on the other. Install one or both.

Why people like it

• It edits in place. The result replaces your selection right inside Gmail, Slack, Jira, Notion, a form, anything. No copy and paste round trip.
• Your key, your data. There is no service in the middle. Text goes straight from your browser to Google Gemini or to LM Studio on your own machine. Settings, prompts and history never leave the device.
• Zero footprint until you ask. No content script on every page. The extension injects itself into the active tab only when you trigger an action, inside an isolated shadow root.
• Budget guard rails. Optional per-minute and per-day limits on requests and tokens, reserved before each call, so retries and parallel requests cannot blow past your quota.
• Fallback provider. Gemini can fall back to LM Studio, and LM Studio to Gemini or another local model, when the primary fails or hits its budget.
• Editable prompts. Every action's system prompt is yours to change in the Prompts tab.
• Review mode. Alt+Shift+R opens a panel with your text and an editable result before anything is applied.

What it can do

Action	Result
Fix spelling and grammar	Corrected text, same language, same meaning
Add polish	More professional, more eloquent
Change tone	Formal, Casual, Empathetic, or with emojis
Official email	Rewritten as a business email
Shorter / longer	Condensed or expanded
Translate	Polish, Ukrainian, English (US)
Format	Markdown or a Slack message
Fix keyboard layout	Ukrainian and English layout mix-ups, done locally with no AI call
Reply Assist	Drafts a reply that matches the tone of the conversation
Custom prompt	Uses the selection as a one-off prompt

Quick start

Browser extension

cd "extension for browsers"
npm install
npm run build

1. Open chrome://extensions/, turn on Developer mode, click Load unpacked and pick extension for browsers/.output/chrome-mv3.
2. Open the popup and paste a Gemini API key from Google AI Studio, or switch the provider to LM Studio and enter the server URL (default http://localhost:1234).
3. Confirm the shortcuts at chrome://extensions/shortcuts.

Select text on any page, then right-click → AI Text Tools, or press:

Shortcut	Action
Alt+Shift+F	Fix spelling and grammar now, using the auto-select setting
Alt+Shift+G	Fix only the manually selected text now
Alt+Shift+A	Reply Assist with optional conversation context
Alt+Shift+R	Review panel: edit the result before applying
Alt+Shift+Esc	Cancel the running request; your text stays untouched

Desktop app

cd desktop-app
npm install
npm start          # development
npm run build      # NSIS installer under dist/

Paste into the Compose tab or run selected text through a global shortcut. Each action can have its own hotkey, the layout fixer defaults to Shift+Alt+L, and the app can start hidden in the tray with Windows. Details in desktop-app/README.md.

How it works

select text ──► background worker ──► Gemini API  or  LM Studio (localhost)
                      │                              │
                      ◄──────── result ──────────────┘
                      ▼
      content script replaces the selection, fires input/change events,
      shows a toast, saves the entry to local history (last 50)

Requests carry a 30-second timeout. The content script handles <input>, <textarea> and contenteditable editors and fires the events frameworks like React and Vue listen for, so the page sees the change as if you typed it.

Model limits and budgets

Settings → Model limits has an Enable custom budget limits switch. New setups work without it; existing saved budgets stay enabled after an upgrade.

Gemini's input and output maximums load automatically with your API key. Google no longer publishes a universal free-tier table, so copy your project's active rates from Google AI Studio into the primary and fallback model profiles, or set smaller personal caps.

When enabled, the extension enforces requests per minute and per day, input tokens per minute, an optional daily input-token cap, and per-request input and output caps. Reservations are saved before generation, so concurrent requests, retries, worker restarts and failed attempts cannot bypass them. Daily budgets reset at midnight Pacific time. Local models use a conservative text-size estimate; set their input cap below the loaded context window.

Usage covers this extension in this browser only. Google quotas are shared across the project, so calls from other apps are not visible here.

Why Alt+Shift+Esc is different

Chrome reserves Esc and will not bind it through chrome.commands, so the cancel shortcut is handled by the content script. It fires only while the page has focus, not from the address bar or another window, and it does not appear on the shortcuts page.

Repository layout

extension for browsers/
  wxt.config.ts       WXT-generated MV3 manifest and permissions
  entrypoints/        Background service worker
  public/             Popup, content script, icons, shared prompts and provider logic
  store/              Chrome Web Store listing, privacy policy, permission rationale
  test/               Mock AI server, test page and Playwright end-to-end run
  zip-extension.sh    Builds the upload package
desktop-app/
  main/ renderer/ shared/   Electron main process, UI, shared constants
landing-page/         Public site and hosted privacy policy for GitHub Pages
.github/copilot-instructions.md   Architecture and data flow in depth

Publishing to the Chrome Web Store

cd "extension for browsers"
npm run zip

The version comes from manifest.json; bump it and package.json before every upload, since the store rejects a package that is not newer than the published one. Listing copy, the privacy policy and the per-permission justifications live in extension for browsers/store/, starting with STORE_LISTING.md.

The privacy policy must be reachable at a public URL. Enable GitHub Pages for this repository (Settings → Pages → deploy from main, root) and the copy in landing-page/ is served at https://sanekxarcs.github.io/fix-errors-ex/landing-page/privacy.html.

Testing

cd "extension for browsers"
npm test            # text replacement logic
npm run test:env    # local mock AI server and a page with every editor type
npm run test:e2e    # builds the extension and drives it with Playwright

Privacy

• Settings, prompts and history live on your device only: chrome.storage.local in the extension, electron-store in the desktop app.
• Text you act on is sent only to the provider you configured: Google Gemini, or your own LM Studio server.
• Nothing is collected, sold or shared. There are no servers in this project.

Full text: extension for browsers/store/PRIVACY_POLICY.md.

Contributing

The Gemini model list and the default prompts are duplicated in extension for browsers/constants.js and desktop-app/shared/constants.js. Update both when you change one. .github/copilot-instructions.md documents the architecture and data flow in more detail than this file.