What does regex catch that the AI layer does not, and vice versa?

Regex catches API keys, JWTs, emails, IBANs, Luhn-valid cards, phones, and high-entropy tokens instantly. The Jev layer scores short value-like snippets such as names, orgs, account IDs, and balances, blurring anything at or above your threshold.

How much of my browsing actually leaves the browser?

Only 2-200 char snippets under 12 words plus about 160 chars of context, never whole pages. Calls go through the service worker so the page never sees the key, and verdicts cache in chrome.storage.local for zero-token repeats.

How do I tune blurring when it hides too much or too little?

Set the 0.4-0.95 sensitivity threshold with optional per-site overrides, toggle hover-to-reveal, master enable, per-site disable, rescan, and the AI classifier itself. Decisive verdicts reuse across contexts; borderline ones stay context-keyed.

What is the install path from checkout to working popup?

Run npm install and npm run build, load .output/chrome-mv3 unpacked with Developer mode, then paste the key in the popup. Try demo/index.html over HTTP and use npm run dev, compile, test, and gen-icons for development.

Does it keep scanning as the page changes?

Yes. A MutationObserver keeps scanning as the page changes, and input values containing keys are blurred too, so new content is covered without a manual rescan. Blurring wraps only the exact matching substring, leaving the rest of the page readable.

Where is my API key stored, and do verdicts leave the browser?

The key is stored in chrome.storage.local only, and classification verdicts are cached persistently there too, never leaving the browser, so repeat visits cost zero tokens. All API calls go through the extension's service worker, so the page never sees the key.

How does it decide to reuse a cached verdict?

Decisive verdicts are reused across contexts, while borderline ones are keyed to their exact context, and both are cached persistently in chrome.storage.local. That is what lets a repeat page cost nothing while a genuinely ambiguous snippet is re-judged in its own surroundings.

CommunityJev Tools & CLIs

Mrsecret

Chrome MV3 extension blurring secrets and PII in-page with instant regex plus a Jev classifier for ambiguous snippets, keeping the rest of the page readable.

See highlights

Open github.com

Stars: 1
Language: TypeScript
Published by: Community
Author: vkpdeveloper
Last verified: Oct 2026

vkpdeveloper/mrsecret

User repository on GitHub

View on GitHub

Mr. Secret — blurs secrets & PII on any page using TypeSafe AI Jev

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A Chrome MV3 extension that blurs secrets and PII on any web page — useful when screen sharing, streaming, or recording demos. Two detection layers run in the page: Regex layer (deterministic, instant): API keys (sk-…, AKIA…, ghp…, xoxb-…, AIza…, glpat-…, Stripe, HF, private-key blocks), JWTs, emails, IBANs, Luhn-valid card numbers, phone numbers, and generic high-entropy tokens (Shannon entropy > 3.5).

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

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Deterministic regex for keys, JWTs, emails, IBANs, cards, phones, and entropy over 3.5.
Jev probabilities for ambiguous short snippets with adjustable threshold and per-site overrides.
Exact-substring blur spans with hover reveal and MutationObserver plus input coverage.
Service-worker API calls, local-only key storage, and persistent verdict cache.

Watch out

Sends short snippets plus context to TypeSafe; keys stay in chrome.storage.local and verdicts cache locally. 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

Jevest

GitHub Action that reviews pull requests with a read-only finder agent plus a refuting verifier, using Jev only to triage, route, label, and gate merging.