Read the web faster. ArcaRead is a Manifest V3 browser extension that applies bionic reading — bolding the first half of each word — to any webpage, plus ships a full Reader view with on-device AI summaries, a dyslexia toolkit, and a PDF viewer.
Free and open source on browsers. Native desktop / mobile apps are a separate (paid) product.
- What you get
- Install
- Usage
- Privacy
- How it works
- Project layout
- Develop locally
- Safari packaging
- Roadmap
- License
- Bionic bolding on any page, with per-site adjustable intensity (25–75%) and minimum word length
- Smart article detection — the effect is scoped to the actual article body so nav, sidebar, footer, and ads stay plain
- Link-density filter — skips blocks that are mostly hyperlinks (news feeds, tag clouds)
- Per-site memory — sliders tuned for Medium remember Medium settings; Wikipedia gets its own
- Auto-skip on app-like pages — optional; ArcaRead stays quiet on Gmail, dashboards, and anything without a detectable article
- Focus mode — dim every paragraph except the one you hover
- Dyslexia toolkit — line-height, letter-spacing, word-spacing sliders + optional Lexend reading font
- Reader view — clean distilled article tab with TTS playback, phrase-chunking speed-read mode, and five contrast tints (Paper / Sepia / Solarized / Dim / Night)
- PDF reader — rendered through PDF.js with bionic applied, same toolbar
- On-device AI summary — Chrome's built-in Gemini Nano generates a TL;DR for any article. Zero API cost, zero data leaving the device
- Reading analytics — daily / all-time minutes read + WPM estimate, stored locally only
- Keyboard shortcut —
Cmd/Ctrl + Shift + Ftoggles the current site
- Clone this repo or download as a zip.
- Open
chrome://extensions(edge://extensions,brave://extensions, etc.). - Toggle Developer mode.
- Click Load unpacked and select the repo folder.
- Pin the extension from the puzzle-piece menu.
A Chrome Web Store listing is on the roadmap; until then, unpacked install is the route.
MV3 landed in Firefox 121+. ArcaRead targets the Chromium API surface; Firefox works with minor caveats (service worker semantics differ slightly). Same about:debugging → "Load Temporary Add-on" flow. A web-ext build and AMO submission is planned.
See Safari packaging below — one xcrun invocation produces an Xcode project.
- Click the toolbar icon to open the popup. Everything you need is there.
- The preview at the top shows the current effect live as you drag the sliders.
- Open this page in Reader distills the article into a dedicated reader tab.
- In Reader: hit ▶ for TTS, Chunks for a speed-read (RSVP) overlay, Summarize for a Nano-generated TL;DR, and the tint dropdown for palette.
- Keyboard shortcut:
Cmd+Shift+F(orCtrl+Shift+F) toggles ArcaRead for the current site.
ArcaRead does not ship any data off your device. Ever. Specifically:
- Content scripts. Yes, ArcaRead runs a script on every page you visit — that's how bionic bolding works. It reads text nodes to bold their prefixes. It does not transmit any of that text.
- Settings. Sliders and per-site overrides are stored in
chrome.storage.syncso they follow you across signed-in Chrome profiles, via Google's standard sync. ArcaRead does not see or receive any of that data. - Analytics. The reading-time and WPM stats live in
chrome.storage.localon this machine only — never synced, never sent anywhere. - AI summaries. The Summarize button uses Chrome's on-device Gemini Nano. Article text is passed to Nano locally; nothing leaves your computer. If Nano isn't available (Safari, Firefox, older Chrome), the button shows a clear "not available" message — there is no network fallback.
- PDF reader. When you opt into "Open PDFs in reader", ArcaRead fetches the PDF bytes from its original URL and renders them locally via PDF.js. No third party sees the fetch.
- Network. The extension makes no requests to any server operated by the author. No analytics SDK, no crash reporter, no telemetry.
If you're still uncertain: the full source is right here. content.js is the only file that touches page content. grep "fetch\|XMLHttpRequest\|sendBeacon" shows every network call — you'll find only the PDFs you explicitly asked to open.
core.js is a ~60-line dependency-free module exposing FocusCore.transform(text, settings) which returns { segments, modified } where each segment is { text, bold }. It handles Unicode letters, punctuation, and the minimum-word-length / intensity settings. It is UMD-exported (CommonJS + global), so the exact same code runs in the content script, in Node (tests), and later in a JavaScriptCore bridge for iOS / macOS.
content.js walks text nodes under either a detected article root (if smart mode is on and an article exists) or the full document.body. Each text node is replaced with a <span class="focusread-processed" data-focusread-original="..."> containing the transformed HTML. The data-focusread-original attribute preserves the source text so toggling off cleanly restores the DOM.
Skip rules:
- Tags in the skip list (
SCRIPT,STYLE,CODE,PRE,INPUT,TEXTAREA,NAV,ASIDE,FOOTER, SVG, canvas, etc.) - ARIA roles (
navigation,complementary,banner,search,contentinfo,menu,menubar,tablist) contenteditableelements- Font size below the minimum or above the maximum threshold
- Blocks where >60% of text is inside
<a>tags (news feeds, tag clouds)
A MutationObserver on the chosen root handles SPA navigation and lazy-loaded content. Open shadow roots are traversed explicitly at initial load and when new shadow-hosting elements are added.
Two tiers, merged at read time:
- Global defaults (stored in
chrome.storage.sync) — base values that apply to every site. - Per-site overrides (
siteSettings[hostname]) — any subset of keys overrides the global for that hostname.
Changing a slider in the popup always writes to the current site's overrides, never to the global. A "Reset" button wipes the overrides for this site. A small indicator dot lights up when overrides exist.
When the user clicks Open this page in Reader, the content script clones the detected article root, strips unsafe tags (<script>, <iframe>, <object>, <embed>), absolutizes relative URLs, and hands the sanitized HTML to the popup. The popup stashes it in chrome.storage.session and opens reader.html?id=<key>. The reader page reads the stashed payload, renders it with FocusCore.transform applied, and exposes TTS, chunks, tints, and summarize.
vendor/pdf.min.js + pdf.worker.min.js are the stock Mozilla PDF.js build. When Open PDFs in reader is enabled, the background service worker intercepts navigations to *.pdf URLs and redirects them to viewer.html?file=<url>. The viewer fetches the PDF, extracts text with PDF.js, groups items into paragraphs by vertical gap, and renders each paragraph with bionic applied.
core.js # portable transformation — no DOM, no browser APIs
manifest.json # MV3 manifest
background.js # service worker — defaults, keyboard shortcut,
# PDF intercept, analytics aggregator, welcome trigger
content.js # DOM walker + MutationObserver (consumes core.js)
popup.html, popup.js # toolbar popup — settings UI, Open in Reader button
reader.html, reader.js # clean reader view — TTS, chunks, tints, summarize
viewer.html, viewer.js # PDF reader — wraps PDF.js with bionic rendering
welcome.html # first-run tour
fonts/ # Lexend variable font + SIL OFL licence
vendor/ # PDF.js bundled build
icons/ # app icon 16/32/48/128 + SVG source + design explorations
tests/ # unit tests for core.js (node --test)
.github/workflows/ # CI: tests, syntax, zip artifact
# Run the unit tests
node --test tests/
# Syntax-check all extension JS
for f in core.js content.js background.js popup.js reader.js viewer.js; do
node --check "$f"
done
# Validate the manifest
python3 -m json.tool < manifest.json > /dev/nullChanges to extension files are picked up by clicking the reload button on the ArcaRead card at chrome://extensions. Changes to content scripts also require reloading any open tabs you want to test on — Chrome does not retroactively inject into open tabs.
CI runs the same three checks on every push and builds a downloadable zip artifact on the Actions tab for handing to testers.
The extension is a standard MV3 WebExtension and packages through Apple's converter. One-time Xcode setup if you haven't already:
sudo xcodebuild -runFirstLaunchThen from the repo parent directory:
xcrun safari-web-extension-converter ./focusread \
--project-location ./focusread/safari \
--app-name "ArcaRead" \
--bundle-identifier "com.damienjerry.focusread" \
--copy-resources --no-open --forceOpen the generated safari/ArcaRead/ArcaRead.xcodeproj in Xcode, sign with your Apple ID team, build, and run. In Safari, enable Develop → Allow unsigned extensions or use your signed build, then flip it on under Settings → Extensions.
- Chrome Web Store / Edge Add-ons / Firefox AMO / Safari App Store listings
- Shadow-DOM mutation observers (catch changes inside existing shadow roots, not just on initial load)
- Native iOS app with Share Sheet integration — paid, separate repo
- Native macOS / Windows / Linux readers — paid, separate repo
- Optional
en/es/jasummariser output (currently English only) - BYOK AI (OpenAI / Anthropic) as a power-user alternative to Gemini Nano
ArcaRead source: MIT — see LICENSE.
Lexend font (bundled under fonts/): SIL Open Font License 1.1 — see fonts/OFL.txt.
PDF.js (bundled under vendor/): Apache 2.0.