Question 1

Should I actually use this instead of uBlock Origin?

Accepted Answer

No, and the README says so itself: this is a fun side project, not a real ad blocker. It costs tokens per page, it will miss ads, it will occasionally eat something that is not an ad, and it does nothing about tracking, malware, or video ads. It exists to show a small decision model judging DOM elements; if you want ads blocked, use uBlock Origin.

Question 2

What exactly leaves my browser on a page load?

Accepted Answer

One request per batch to api.typesafe.ai carries the page hostname and title plus, per candidate, tag, classes, label, up to 220 characters of visible text, link hostnames, image alt text, and attribute names. Numbers are turned into words first because Jev is weak on them. There is no backend, no page is ever sent whole, and the key you paste lives in chrome.storage.sync.

Question 3

Will it remove something that is not an ad?

Accepted Answer

Possibly, and the evidence is narrow: on a German news fixture it reported 11 of 11 hand-written candidates judged correct, 8 of 8 ads removed, and 0 false removals, at roughly 0.7 to 1.9 s per batch. Everything at or above the threshold is removed with element.remove(), and the popup lets you drop that threshold from 0.70 to as low as 0.30, which trades precision for coverage.

Question 4

Does it catch ads with no label and no ad-network link?

Accepted Answer

No. Jev reads literally, and only ad-shaped candidates reach the model at all: iframe and ins tags, id and class tokens such as ad, sponsored, or banner, data-ad attributes, links to known ad networks, and short labels including Sponsored, Advertisement, Anzeige, and Werbung. An ad with none of those never enters the candidate list, and ads inside cross-origin iframes are removed as a whole iframe without the script ever seeing their content.

Question 5

What do I need to install and enable the extension?

Accepted Answer

You load it unpacked: open chrome://extensions, turn on Developer mode, and pick the repository folder, with no build step. Click the icon, paste a TypeSafe API key and hit Test. You can also exercise the real content script against a German news fixture through npm run relay and npm run serve before installing anything.

Question 6

How fast does it judge a page, and how big are the batches?

Accepted Answer

Batches are debounced by 600 ms and capped at 30 candidates per request, and the README's fixture reported roughly 0.7 to 1.9 seconds per batch. A MutationObserver plus a scroll handler catches lazy-loaded ads, and Rescan page forgets the already-judged set and scans again. Those timings come from that one fixture, not a general guarantee.

Question 7

Can I see why an element was flagged without removing it?

Accepted Answer

Yes. A Highlight only mode outlines the element in red and shows the probability instead of deleting it, which the README calls the debug mode. Everything at or above the threshold is otherwise removed with element.remove(), and the pop animation can be turned off so removal is instant rather than a pulsing outline.

TypeSafe Fun AdBlocker

About this project

Highlights

Quickstart

Questions about this entry

Reactions & coverage

From the Jev guides

From the Jev AI model community