Noteshack

A fast, private word processor and PDF editor that runs entirely in your browser. No account, no server, no tracking. Your documents stay on your device. It saves as you type, works offline, opens and saves Word documents, and lets you edit and sign PDFs.

Features

Writing

• Headings, quotes, code blocks, bulleted, numbered and check lists
• Bold, italic, underline, strikethrough, superscript, subscript, inline code
• Any font size and font for selected text, a text color and highlight palette
• Alignment, indent, links, horizontal lines, page breaks, special characters
• Tables: insert from a size grid, add or delete rows and columns, header row, Tab between cells
• Images: insert, paste or drag in; resize by dragging or by preset; center or wrap text; alt text
• Markdown shortcuts while you type: # heading, - bullet, 1. numbers, [ ] checklist, > quote, ``` code, --- line

Documents

• Unlimited documents, with a searchable sidebar
• A welcome on opening that sums up what Noteshack does (tick "Don't show this again" to stop it)
• Autosave plus version history: browse earlier versions and restore any of them
• Open .docx, .md, .html and .txt, by menu or by dropping files on the window (PDFs and images open in the PDF editor, below)
• Save (Ctrl+S) writes over the document's file; the first time, it is Save as. Save as… (Ctrl+Shift+S) picks the name, folder and kind: Word (.docx), web page, Markdown or plain text. Export… makes a PDF (through Print). Everything is also kept in Noteshack as you work ("Saved in Noteshack").
  • In Chrome and Edge on a computer, Save as uses the system's own Save dialog (pick the kind of file in its type list), and Save (Ctrl+S) then writes straight back to that file, like a desktop program. A file opened through File → Open, or dropped on the window, is saved back to the same way. This lasts while the tab is open: after a reload or a new visit, the first Save shows the dialog again, in the same folder with the same name. Saving over a Word file or web page made elsewhere asks first (once), since Noteshack opens their everyday parts, not all of them.
  • Firefox, Safari and phones do not let websites write to a file on disk, so there each save is a new copy in Downloads; Save as asks which kind of file.
• Page setup per document: font, size, line spacing, Letter or A4, margins
• The app in your language: English, Spanish, French, German, Portuguese or Italian, chosen in View → App language (or the browser's language, at first). Menus, messages, dates and numbers follow it; each document keeps a language of its own. The translations were written with the help of AI and have not yet been checked by native speakers. Corrections are welcome through Tools → Report a problem.
• Report a problem (Tools): say what went wrong, in the app, with no account; the report is filed on Noteshack's GitHub page, where it is public.
• Each document has a language, shown in the status bar: worked out from its text on your computer, or chosen from 55. It sets spell checking and which way text runs (Arabic, Hebrew, Persian and Urdu run right to left), and saved Word files, web pages and PDFs record it.
• Spell checking in the document's language: English, Spanish, French, German, Portuguese and Italian are checked by Noteshack itself, with the Hunspell dictionaries LibreOffice uses. Misspelled words are underlined; right-click one for corrections, to add it to your dictionary, or to ignore it in that document (Shift + right-click gives the browser's own menu). Other languages use the browser's spell check.
• Translation, on your computer (Tools → Translate…): makes a translated copy of a document; with a translator on your computer, the text is not sent anywhere. Noteshack's own translator is Firefox's translation engine with Mozilla's models, between English, Spanish, French, German, Portuguese and Italian (two of the others go through English); it keeps bold, links and other formatting, and downloads each language's model (about 25 MB) the first time, then keeps it. Where the browser has a translator of its own (Chrome), its languages are offered too, as plain text. Online, only when you choose it, Noteshack's own translation service on Cloudflare translates between 53 languages (Meta's M2M-100 model, run by Workers AI), as plain text: the text is sent there, translated and sent back, and not kept, so it is not for anything confidential. Names and terms you list are left as they are, and so are pictures and code. Each paragraph of the copy is highlighted until checked: the panel shows its original, which you can change and translate again. A PDF (or a scan, read first) can be translated keeping its layout: a copy in which each paragraph is replaced where it stands, in its matched font, set a little smaller where the translation runs longer so it stays in its space, with the original left as it was; or into a document of its words, to check paragraph by paragraph and save as Word or text.
• Back up every document to one file and restore it anywhere

PDFs

• Open any PDF (menu, or drop it on the window). It goes in your documents list like everything else.
• Edit text: click any line of the PDF's own text and retype it. Only the words you change are replaced; the rest of the line is left exactly as it was.
• The original font: the app works out which font the words were set in (from the font's name in the PDF, or, on a scan, by comparing the letters with 31 common document fonts) and downloads it, so new words match. Paid fonts like Arial, Times New Roman and Calibri are matched with free fonts drawn to the same shapes and widths (Arimo, Tinos, Carlito).
• Text in any script: Arabic, Hebrew, Hindi and the other Indian scripts, Thai, Chinese, Japanese, Korean, Cyrillic, Greek and Vietnamese can be typed into a PDF, or translated into one, and are saved shaped (letters joined, conjuncts formed, marks placed) and in the right order, right to left where they run that way, in a Noto font. The font for a script is downloaded the first time it is used, then kept, and only the letters used go into the saved PDF, so a page of Chinese adds kilobytes, not the font's 10 MB. Such text can be searched and copied from in the saved file. Offline, before a script's font has been downloaded, its letters are saved as “?” and the app says so.
• Add text, highlight, white-out, draw, boxes, images, and your signature (drawn once, remembered if you like)
• Fill in PDF forms: text boxes, checkboxes, dropdowns and radio buttons
• Images too: open a JPG, PNG, WebP, TIFF or BMP (a scan saved as an image, or a photo of a page) and it opens in the same editor, with every tool here. A photo taken sideways is turned upright. Save gives back the same kind of file at the same size in pixels, with your changes drawn in: a JPG comes back as a JPG, and a multi-page TIFF as a multi-page TIFF. (A JPG, PNG or WebP that has gained pages comes back as a zip with one image per page.) Export… makes it a PDF, a Word document, text or another kind of image.
• Scanned pages become editable text (OCR): with Edit text, click a scanned page and the app reads that page (a few seconds; the line you clicked then opens for editing), takes the scanned words off it, and draws every line again as real text in the matched font, size, weight (bold and italic words included) and color, where it was on the page. Edits are then ordinary text edits that match the lines around them, and the downloaded PDF can be searched and copied from. This works the same way on pages whose letters were saved as drawn shapes.
  • A page scanned sideways or upside down is turned the right way up first, and one that went through the scanner crooked is straightened, so lines read cleanly and the new text sits level.
  • Headings set in small caps stay in small caps.

What PDF editing cannot do, so nothing surprises you:

• Text you change or white-out is erased from the file when you save, not just covered, and the rest of the line stays exactly where it was. That includes scanned pages and other pictures: the covered pixels are painted over inside the image itself. Only drawings (lines and shapes) under a white-out stay in the file, hidden. The app tells you when that happens.
• A color scan you have erased part of is saved again as JPEG, so that page's file size can roughly double. Black-and-white scans stay small.
• OCR can misread faint, tiny or handwritten text. Check what it read (the Original button) before relying on it.
• Redrawing a scanned page needs its fonts. Offline, the page stays as scanned and Edit text changes single words instead; choose Pages → Make this page's text editable when you are back online.
• Font matching needs an internet connection the first time a font is used. The 31 common fonts come from Noteshack itself; a less common font a PDF names comes from the free Fontsource library through jsDelivr (nothing from your document is sent). Offline, or for a font it cannot find, the closest standard font is used.
• A scan's font is matched to the closest of the common fonts the app knows. An unusual or custom typeface will be matched to something similar, not identical.
• Password-protected PDFs cannot be changed.

Tools

• Find and replace with highlighting, match case, whole words and regex
• Word count details, reading time, selection count, and a word goal with a progress bar
• Read aloud with word highlighting, a choice of voice and speed
• Spell check toggle, focus mode, zoom, light and dark themes
• Full keyboard shortcuts (press Ctrl+/ to see them all)
• Works on phones and tablets, works offline, installable as an app

Where your writing is kept

Documents are stored in your browser's IndexedDB on this device. They are never uploaded. Clearing your browser's site data deletes them, and they don't follow you to other browsers or computers. Use File → Back up all documents to keep a copy, or to move them.

Upgrading from the first version? Your old draft is moved into the documents list automatically the first time you open the new version.

Privacy

There is no account, no ads and no cookies, and documents are never uploaded. Three things are fetched from the internet, and two are sent only when you ask:

• Uncommon font files. The 31 common fonts the app matches are served by Noteshack itself. When a PDF names a font outside them, that font file is downloaded from the free Fontsource library through jsDelivr (cdn.jsdelivr.net). Like any website, jsDelivr sees the visitor's IP address; nothing from the document is sent.

• Linked pictures. A picture in a document that links to a web address (for example, pasted from a web page) is loaded from that address.

• Translation models and script fonts, from Noteshack's own file storage (files.noteshack.org), kept in the browser after the first time: the model for two languages, the first time you translate between them, and the font for a script (Arabic, Chinese, Hindi and others), the first time a PDF needs it. Only the file is downloaded; the text is translated, and fonts are cut down to the letters used, on the device.

• The online translator, only when you choose it (Translate → Online): the text being translated is sent to Noteshack's translation service on Cloudflare (api.noteshack.org), which translates it with Workers AI and sends it back. Nothing is kept or logged.

• A problem report, when you send one (Tools → Report a problem): what you write, the app's version, your browser and the app's language go through the same service to Noteshack's GitHub page, where the report is public. Nothing from your documents is sent.

Everything else, text recognition included, runs on the device.

Disclaimer

Noteshack is provided as is, without warranty of any kind. Text recognition and font matching can be wrong: check a PDF before sending, submitting or signing it, and keep backups of anything important.

Users are responsible for the documents they change. Do not alter a document in a way that misleads anyone, such as a form or record someone else filled in or signed.

License

Noteshack's own code is released under the MIT License. The libraries in vendor/ keep their own licenses, listed below.

Credits and licenses

Noteshack uses these open-source projects, kept unmodified in vendor/ with their license texts:

• PDF.js (Mozilla), Apache License 2.0
• pdf-lib and @pdf-lib/fontkit, MIT License
• Tesseract.js, Tesseract OCR and its English, Spanish, French, German, Portuguese and Italian data, Apache License 2.0
• opentype.js, MIT License
• UTIF.js (Photopea), MIT License, with pako, MIT License
• Hunspell (as hunspell-wasm), MPL 1.1, with Hunspell dictionaries from wooorm/dictionaries, each under its own license (see vendor/dictionaries)
• franc, MIT License
• Bergamot translator (Firefox's translation engine), Mozilla Public License 2.0, with Mozilla's translation models, Mozilla Public License 2.0 (served from files.noteshack.org, not stored in this repository; tools/translation-models.py puts them there)
• M2M-100 (Meta), MIT License, run by Cloudflare Workers AI for the online translator (worker/)
• HarfBuzz (as harfbuzzjs, with its subsetter), MIT License (Old MIT)
• bidi-js, MIT License

Fonts come from Fontsource when needed and are not stored in this repository; each is under its own free license (SIL Open Font License or Apache License 2.0). Fonts made from a scan's letters are changed copies of those fonts, so they are renamed (Noteshack Scan 1, 2…) and keep their license notices, as the Open Font License asks.

Trademarks

Adobe, Acrobat and ClearScan are trademarks of Adobe. Microsoft, Word and Calibri are trademarks of Microsoft. Arial and Times New Roman are trademarks of The Monotype Corporation. They are named only to describe compatibility; Noteshack is not affiliated with or endorsed by them.

Running it

It is a static site with no build step and no dependencies.

• Locally: serve the folder over http, for example python3 -m http.server, and open http://localhost:8000. (Opened straight from disk, the word processor works but PDFs do not: browsers will not load PDF.js's modules from a file:// page.)
• GitHub Pages: Settings → Pages → Deploy from a branch → main, root folder. The live site is https://noteshack.org, set by the CNAME file and the domain's DNS records (four A records and four AAAA records for GitHub Pages, and a www CNAME to bobothehutt.github.io).

Each web address keeps its own saved documents, so after moving the site to a new address, open the old one, use File → Back up all documents, and restore that file at the new address.

Tests

The app needs nothing installed; the tests need Node.js.

npm install
npx playwright install chromium firefox webkit
npm test

They serve the repository as the site and use it in Chrome, Firefox and Safari's engine (WebKit): writing, exporting, editing and downloading PDFs, filling forms, reading scans, storage, printing, accessibility (axe), privacy (which sites are contacted), small screens and offline use. The PDFs they open are made fresh each run from invented text, so no real document is in the repository. Only the ones a change affects are run; GitHub runs those in all three when started by hand (.github/workflows/test.yml).

The app's languages

The code is written in English, with each phrase the app shows wrapped: _('Saved to “{name}”', { name }), or for a number, _n(count, '{n} page', '{n} pages'). Each language's file in locales/ gives the same phrase translated, keyed by the English (a plural under the English plural, one form per plural category of the language). The text in index.html is translated where it stands when the app starts.

A new or changed phrase needs a translation in every file in locales/: the tests list any that is missing, any no longer used, and any whose {…} parts differ from the English. node tools/phrases.mjs lists the phrases the code uses. Two test languages, from the address only, help find what was missed: ?lang=en-XA shows every phrase accented and longer (so text left in plain English, or cut off, stands out), and ?lang=ar-XB does the same right to left.

Files

index.html             page structure, menus and toolbar
js/boot.js             sets the theme before the page first draws
styles.css             all styling: light and dark themes, phone layout, print
js/app.js              the editor and the rest of the app
js/store.js            document storage (IndexedDB)
js/sanitize.js         cleans pasted and opened content
js/markdown.js         Markdown import and export
js/docx.js             Word .docx import and export
js/zip.js              the zip reading and writing .docx needs
js/pdf.js              the PDF editor
js/fonts.js            identifies and downloads the font a PDF was set in
js/scanfont.js         builds a font from a scan's own letters
js/images.js           opens images in the PDF editor and saves them back as images
js/i18n.js             the app's own language: phrases, plurals, the test languages
locales/               the app's phrases in Spanish, French, German, Portuguese and Italian
js/lang.js             the languages, and working out which one a text is in
js/spell.js            spell checking (js/spellworker.js runs Hunspell)
js/translate.js        translation on the device: Noteshack's engine (Bergamot) or the browser's
js/remote.js           downloads from files.noteshack.org: checked, and kept in the browser
js/shaping.js          text in other scripts in saved PDFs: Noto fonts, shaped (HarfBuzz), cut down to the letters used
worker/                Noteshack's online service (api.noteshack.org), a Cloudflare Worker: the online translator, and problem
                       reports filed on GitHub; GitHub deploys it when it changes (tools/worker.py)
vendor/                the libraries (see vendor/README.md), with their licenses; loaded only when needed
tools/                 the upload scripts for the translation models and the Noto fonts, the online translator's deploy script,
                       test-setup.py (the tests' fonts and models, where the network cannot reach them); phrases.mjs lists the phrases the code uses
js/icons.js            toolbar icons
sw.js                  offline support
manifest.webmanifest   app install details
AUDIT.md               audits: the first version, and the app as released
LICENSE                the MIT License, for this project's own code
CLAUDE.md              notes for an AI assistant working on the code: releasing, secrets, open work
tests/                 the tests (see Tests above)

After changing any file, bump CACHE in sw.js so installed copies pick up the update, and the ?v= number on each of the app's files in index.html to match it (Cloudflare and browsers keep the code files for hours; a new number is a new address, so nobody gets new pages with old code; the tests check the two agree). After changing a file under vendor/, bump VENDOR there too: those files are cached apart and are not fetched again otherwise.