Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

VideoAdGuard-Jev

Chinese Bilibili extension fork that uses Jev Noul and Choice on subtitles to find sponsored segments and auto-skip them when confident.

ProjectbetaGPL-2.0-or-later
View on GitHub ↗

Overview

VideoAdGuard Jev is a derivative of the VideoAdGuard browser extension that adds a TypeSafe Jev provider for detecting embedded ads in Bilibili videos from their subtitles. Documentation is in Chinese. It uses independent Noul probabilities for ad, purchase prompt, normal content and product matching, and a Choice to pick the product name from the pinned comment or link title rather than generating text.

Analysis is two-stage: overlapping subtitle windows are screened, then candidate regions are refined subtitle by subtitle. Subtitle indexing, time mapping, interval merging and safety thresholds are all done in local code, and auto-skip happens only when confidence is high and segment count and total duration pass checks. Jev's cache is isolated from other models. Videos without subtitles need optional Groq transcription. It is GPL-2.0-or-later with tests.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this subtitle window an advertisement?yes · no
Yes / noDoes this subtitle contain a purchase prompt?yes · no
ChoiceWhich product candidate (from the pinned comment or link title) is being promoted?—

How it works

  1. Collect Bilibili subtitles (or Groq transcription).
  2. Screen overlapping windows with Noul questions.
  3. Refine candidate regions subtitle by subtitle; choose the product.
  4. Merge intervals locally and auto-skip only when safety checks pass.

Strengths

  • Conservative auto-skip with structural checks.
  • No free-text generation for product names.
  • Tests for gateway, windows and interval merging.

Caveats

  • Subtitles and titles are sent to the model provider.
  • Model errors can still cause wrong skips; manual mode recommended for important cases.
  • Chinese documentation.

Getting started

npm ci && npm test -- --run && npm run build:chrome, then load builds/chrome/ in chrome://extensions

At a glance

Type
Project
Status
beta
Language
TypeScript
Runtime
Chrome/Firefox MV3 extension
Has code
yes
Has tests
yes
License
GPL-2.0-or-later

Use cases and integrations

How curated lists describe it

  • This project integrates Jev to provide structured decisions for its workflow. See the repository for implementation details.

    logicrw/awesome-jev-projects

Frequently asked questions

What does Jev decide?

Whether subtitle windows are ads or purchase prompts, and which product is promoted.

When does it auto-skip?

Only when confidence is high and segment count and duration pass safety checks.

Is it the official VideoAdGuard?

No, it is a derivative fork.

Related projects

  • sloppy-jevs-extension

    Open-source Chrome extension that blurs AI-generated prose and ads on any page using separate Jev Noul questions with adjustable cutoffs.

    Browser extensionbetaJavaScript
  • bluenoise

    Privacy-first X filtering extension with local keyword and account rules, plus an optional Jev second pass on replies no rule matched.

    Pluginproduction
  • osso

    Chrome extension that asks Jev, sentence by sentence, whether text is what the reader came for and strikes out the filler.

    Browser extensionproductionJavaScript
  • Sharp

    Browser extension that filters your X timeline by plain-language rules, with a decision model like Jev scoring every post by default.

    Browser extensionproductionTypeScript
  • docent_chrome_plugin

    Chrome extension page guide whose small typed decisions (request type, element to click, match or not) use Jev or local Laya.

    ProjectbetaJavaScript
  • Feed Lens

    Chrome extension that labels Weibo, Threads and X posts by topic and expression style using Jev and your own label dictionary.

    Browser extensionbetaTypeScript