What does this do that uBlock Origin cannot?

Network and CSS rules struggle when a sponsored card shares the first-party domain, markup and styling with an organic post. This extension sends candidate elements to Jev and collapses the ones the noul primitive scores as promotional, judging phrasing and promotional intent instead of class names or URL lists.

Does anything leave my machine besides the model call?

The project's claim is zero telemetry, with your key held in chrome.storage.local and sent only to https://api.typesafe.ai. The part worth reading before trusting it is the MAIN-world scriptlet: it is injected at document_start into the page's own execution context so it can prune ad payloads before media players initialise, which is a far bigger surface than a content script.

Will it block posts I actually wanted to read?

That is a threshold problem rather than a bug. The dashboard has a sensitivity slider defaulting to 85% plus a site whitelist toggle, and every block is listed in the activity log with its confidence score and a text snippet, so you can see what went wrong. A pre-filter also drops anything 40px or smaller in either dimension before Jev is consulted at all.

Will it break video playback?

The media engine exists to avoid that. It accelerates in-stream ads muted with legitimate frame decoding specifically to prevent buffer starvation and black-screen freezes, and it simulates the full pointerdown through click gesture chain so skip actions register reliably inside modern custom web components.

How do I build and load the extension?

Clone the repository, run npm install, then npm run build to produce the dist/ bundle, and load dist unpacked from chrome://extensions with Developer mode on. It needs Node.js 20 or higher, a Chromium browser and a TypeSafe API key.

What does it block at the network layer?

35 major ad delivery, header-bidding, tracking and publisher SSP networks through DeclarativeNetRequest rules, including googleads, doubleclick, googlesyndication, mediavine, adthrive/raptive, ezoic, pubmatic, rubicon, triplelift, sharethrough, taboola, criteo and openx.

How does it avoid paying to re-score the same content?

Candidate feed cards are batched into one TypeSafe call, and evaluated snippet hashes are cached in chrome.storage.session so they survive Manifest V3 service worker idle shutdowns. The dashboard's cache-efficiency metric reports how much that saved.

CommunityJev Demos & Experiments

Jev Shield — vmendes90

Manifest V3 ad blocker that scores candidate feed elements with Jev's noul primitive to collapse native ads, layered over network rules and a media engine.

See highlights

Open github.com

Stars: 4
Language: TypeScript
License: MIT
Published by: Community
Author: vmendes90
Last verified: Oct 2026

vmendes90/jev-shield

User repository on GitHub

View on GitHub

Privacy-first Chrome extension that semantically blocks native ads, sponsored feed cards, and video ads using TypeSafe Jev

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

This is one of several Jev listings for the same project. The canonical entry is Jev Shield — caiovicentino.

About this project

🎬 Watch the launch video: brag-output/brag.mp4. Traditional ad blockers (like uBlock Origin) rely primarily on static URL filterlists and CSS selector lists. While effective for traditional banner networks, they struggle with native ads and sponsored feed cards (on platforms like X/Twitter, Reddit, LinkedIn, and modern digital feeds) where promotional posts share the exact same first-party domain, markup structure, and styling as organic content.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Moderation

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

35 ad networks are blocked at the network layer with DeclarativeNetRequest, alongside banner, iframe and sticky-widget removers.
Candidate feed cards go to Jev in one batched call, with evaluated snippet hashes cached in session storage across service worker shutdowns.
Elements 40px or smaller in either dimension are dropped before the heuristic and AI layers, so tracking pixels never cost a request.
A three-tab popup reports blocked counts, scanned totals, cache efficiency and Jev API calls, with a sensitivity slider and site whitelist.

Watch out

Every judgment is a noul probability compared against a sensitivity threshold that defaults to 85%, and moving the slider changes the cut without re-scoring, so a bad setting collapses real posts as easily as ads. The bounding-box pre-filter discards anything 40px or smaller in either dimension, which will also drop small genuine controls. Failures back off two minutes on a 401 or 429 rather than retrying, and the MAIN-world scriptlet injected at document_start into the page's own execution context is a much larger surface than an ordinary content script. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

AskJev

Claude-driven browser automation where plain-English chat drives Brave or Chrome, Jev makes on-page decisions, and a guard freezes irreversible clicks.