Does this send my whole feed to a third party?

Author name and displayed post text go to the TypeSafe API for classification; nothing else leaves the browser and the project runs no server or analytics. Keys use synced storage and verdicts clear with the session or settings change.

Why is a post I care about hidden or vice versa?

Plain code maps probabilities to verdicts with your noise, interest, and blocked-topic thresholds. Near-threshold keeps show an orange barely-kept line, and buildQuestions plus decide are the tuning points.

Will it break infinite scroll or slow browsing?

One call per post evaluates all questions in parallel with per-session caching, and the feed nudges load-more without moving your reading position. Sponsored detection runs locally without an API call.

What happens when LinkedIn changes markup?

Selectors live grouped at the top of the content file. If nothing is detected after landing on the feed, a diagnostic block in the page console describes the structure, and a red badge marks failed API calls.

What are the five questions Jev answers for each post?

engagement_bait and self_promo as nouls, matches_interests as a noul, top_interest as a choice over your listed interests, and blocked_topic as a choice. They go out in one request with a state of { user_interests, post }, and every question is evaluated in parallel.

What are the default thresholds and how do I tune them?

The noise threshold is 70%, hiding a post when P(bait) or P(self-promo) is above it. The interest threshold is 35%, hiding when P(matches interests) is below it. Blocked topics hide when Jev is at least 60% confident. A kept post close to the threshold turns orange, marked kept, barely, to make tuning easy.

Does every post get sent, or are some skipped locally?

Sponsored posts are detected locally with no API call, and posts shorter than the minimum length, 40 characters by default, are never evaluated. Apart from that, the author name and text of each post displayed in your feed are sent to TypeSafe for classification.

CommunityJev Demos & Experiments

Lkclean

Manifest V3 Chrome extension in TypeScript that sends each LinkedIn post to Jev and collapses bait, self-promo, off-topic, and sponsored posts into a one-line bar with reasons and scores.

See highlights

Open github.com

Stars: 14
Language: TypeScript
License: MIT
Published by: Community
Author: stefw
Last verified: Oct 2026

stefw/lkclean

User repository on GitHub

View on GitHub

Chrome extension that cleans up your LinkedIn feed: hides engagement bait, self-promo and off-topic posts using Jev, TypeSafe AI's typed classification model — and explains every decision.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A Chrome extension (Manifest V3, TypeScript) that cleans up your LinkedIn feed. Every post is sent to Jev, the typed classification model from TypeSafe AI, which returns probabilities instead of prose. Engagement bait, self-promotion, off-topic and sponsored posts are collapsed into a one-line bar, and every decision is explained right on the post. Hides the noise – engagement bait, humble-brags, sponsored posts, topics you block, and posts unrelated to the interests you list.

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

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

One request per post evaluates engagement_bait, self_promo, interest match, and blocked topics in parallel.
Hidden posts show reason and scores with Show toggle; kept posts show discreet Kept line with near-threshold warning.
Floating sticker reports analysed counts, in-flight calls, tokens, cost, latency, sparkline, and verdicts.
Verdicts cached per session; API key stays in the service worker and never reaches page scripts.

Watch out

You need your own TypeSafe key and posts under 40 characters are never evaluated. Empty interests filter only noise; thresholds and blocked topics control the rest. Selectors may need updates when LinkedIn changes markup. 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

laya-server

Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.