hide-em

A personal attention filter for the web. You add names, keywords, or phrases. Matching content disappears from any site you visit, shortly after it appears. Text matching only. Fully local. No backend.

What it is

A Manifest V3 Chromium browser extension (Chrome, Brave, Edge, etc.) that runs a single universal content script on every page you load. The script walks text nodes, tests each against your blocklist, and on a match hides the nearest semantic card-like ancestor — <article>, <li>, ARIA role="article" / role="listitem", or a custom element matching *-RENDERER / *-CARD (which catches YouTube tiles, Reddit cards, etc.).

The rules engine is platform-agnostic. There is no site-specific code anywhere in the codebase. Adding a new site is zero work.

What it isn’t

• Not a moderation tool, recommendation engine, or content classifier — it’s a tool for one user, doing exactly what they configured.
• Not AI-based — exact text matching with Unicode normalization, no semantic models, no remote calls.
• Not a thumbnail blocker — text matching only. If the title doesn’t say “Mr Beast,” the tile isn’t hidden.
• Not a community blocklist platform — your rules are your rules. JSON import/export is supported; sharing isn’t a feature.

See the “Non-goals” section below for the long form.

Status

• Engine: stable, 38 unit tests pass.
• Scanner: stable on YouTube, Reddit, generic article sites. Hides via display: none on the matched card.
• Options page: add/edit/delete rules, JSON import/export, debug toggle.
• Popup: not built yet.
• Icons: not designed yet — Chrome shows a generic puzzle icon.
• Web Store: not submitted. Currently distributed only as an unpacked extension.

Install (unpacked, for personal use)

1. Clone this repo and build:

   npm install
   npm run build
   

2. Open chrome://extensions (or brave://extensions, edge://extensions).
3. Toggle Developer mode on.
4. Click Load unpacked and select the dist/ folder.
5. Click the extension’s options entry (right-click the extension icon → Options) and add your first rule.

The extension takes effect immediately on the next page load. Existing tabs need a refresh.

Usage

Open the options page. Each rule has:

Field	What it does
Type	creator / keyword / phrase / regex
Value	The text to match (or a regex pattern, for regex type)
Aliases	Alternative forms — all matched as if they were the value
Whole word	\b…\b boundary — defaults to true for creator rules
Case sensitive	Defaults to false
Scopes	titles / channels / comments / descriptions — currently informational; the universal scanner tests all text against all scopes

Use Export to download your rules as JSON; Import to round-trip them onto another machine.

Debug overlay

Enable Debug in Settings, then on any page press Alt + Shift + D to toggle a small overlay showing scanned, hidden, and last batch time. Useful for verifying the scanner is doing work.

Safety net (console)

The scanner exposes window.__heDebug on every page:

__heDebug.stats     // live counters and state
__heDebug.kill()    // disable the scanner, disconnect the observer, remove all hides
__heDebug.unkill()  // re-enable

It also self-terminates if any single drain exceeds 1s or cumulative scan time exceeds 60s, surfacing a structured console.warn with the relevant counters. This is a deliberate failure mode — visible flash > blank page.

How it works (short version)

┌────────────────────────────────────────────┐
│  Engine (src/engine/, src/shared/storage)  │
│  Pure logic. Compiles rules to one regex   │
│  per scope. No DOM imports.                │
└────────────────────────────────────────────┘
                    ▲
                    │ "does this text match?"
                    │
┌────────────────────────────────────────────┐
│  Universal scanner                         │
│  (src/content/universal-scanner.ts)        │
│  One content script, all sites.            │
│  TreeWalker-free, per-child decomposition, │
│  idle-callback yielding, kill-switches.    │
└────────────────────────────────────────────┘

Development

npm install
npm run dev         # vite build --watch — rebuilds dist/ on file change
npm run build       # one-shot production build
npm run test        # vitest run — engine + shared utilities only
npm run typecheck   # tsc --noEmit

Reload the extension in chrome://extensions after each build (the reload icon on the card).

Repository layout

src/
  background/             # service worker: storage subscriptions, lifecycle
  engine/                 # pure matcher — NO DOM imports allowed
  content/
    universal-scanner.ts  # the one content script — runs on all sites
    debug-overlay.ts      # optional stats overlay (Alt+Shift+D)
  ui/
    options/              # React app: full blocklist management, import/export
  shared/                 # storage wrapper, shared types
vite.config.ts            # manifest is defined inline here

Architecture rules

1. Universal scanner only. One content script runs on all sites. No site-specific code anywhere.
2. No pre-hiding. Items are visible by default. Hidden only on a positive match.
3. The engine never imports DOM types. It is pure logic, testable in Node with no JSDOM and no Chrome API mocks.
4. Per-file soft cap: 200 lines. Split when it grows past that.

Tests

npm run test

The engine has full coverage (rules, scopes, normalization edge cases including ZWJ, diacritics, fullwidth Unicode). The scanner is not unit-tested — it’s DOM-coupled and JSDOM tests of MutationObserver behavior have low ROI. Verification is the kill-switch safety net + the debug overlay + manual smoke testing.

Non-goals

Explicit non-goals — these will not be added without an explicit request and should never be introduced incrementally:

• No image or thumbnail OCR
• No face detection
• No AI or semantic matching of any kind
• No backend, telemetry, or cloud sync beyond chrome.storage.sync
• No community blocklist sharing
• No filtering of content you yourself authored
• No “auto-suggest blocks” — you add rules, the extension applies them
• No site-specific code paths or selectors

Tech stack

• Manifest V3, Chromium-based browsers
• TypeScript, strict mode
• Vanilla DOM in the content script (no framework)
• React for the options page (off the hot path)
• chrome.storage.sync for rules, chrome.storage.local for hit counters
• Bundler: Vite + @crxjs/vite-plugin
• Tests: Vitest

License

MIT — see LICENSE.

This site is open source. Improve this page.