typesafe-adblock and jevfill: Form-Filling and Ad-Blocking With Jev

Most ad blockers work from static rule lists. Most form fillers work from stored field-name mappings. Two small Chrome extensions built on Jev try a different approach for each: instead of rules, ask a fast typed-decision model to classify each candidate element on the page. typesafe-adblock asks "is this DOM element an ad?" for candidates it finds; jevfill asks "which line from my notes best matches this form field?" for the fields on a page.

Both are early and both are candid about it. Neither project claims to replace the tools it resembles — the adblock README calls itself "a fun experiment," and jevfill's own dataset listing notes its claims are "pending human review, with no independent runtime or performance verification." Here's what each one does, and what to weigh before pointing either at a real page.

Key Takeaways

• Same mechanism, opposite goal. Both extensions batch element text and metadata into typed Jev questions and act on a threshold — one to hide/remove elements, one to fill them.
• typesafe-adblock is explicit about its own failure modes. It can miss ads or remove non-ads, and doesn't touch tracking, malware, or video ads.
• jevfill never sends password or payment fields to Jev. That exclusion is hardcoded, not a setting.
• Both require your own TypeSafe API key and send element context to Jev's hosted API — nothing here runs fully offline.
• Both are small: 72 and 18 stars respectively as of September 2026 — early enough that reliability claims should be treated as unverified until you test them yourself.

typesafe-adblock: classifying elements instead of matching rules

typesafe-adblock (MIT, 72 stars as of September 2026) is an experimental Chrome extension that batches candidate DOM elements — their text, labels, and link information — into Jev questions, and applies a confidence threshold to decide whether to highlight or remove each one. Where a traditional blocker like uBlock Origin matches a URL or a DOM pattern against a maintained filter list, this asks a model to judge each element on its own content.

The author's own framing is worth repeating in full: it's called "a fun experiment," and the README states plainly that it can miss real ads and remove elements that aren't ads. It doesn't attempt tracking prevention, malware blocking, or video-ad handling — the scope is limited to detecting and hiding ad-like page elements by content. It requires the user's own TypeSafe API key, meaning element text and link data leave the browser for classification on every batch. There's no published accuracy figure in the source material we reviewed, and none should be assumed. Aiskill.market entry: /skills/typesafe-adblock-realzachi.

For a different content-classification use of Jev — sentence-level scoring rather than page-element classification — see creative Jev's jevmeter, which applies the same "batch candidates into typed questions" pattern to video transcripts instead of ads.

jevfill: matching form fields to your own notes

jevfill (MIT, 18 stars as of September 2026) is a Manifest V3 Chrome extension built in TypeScript with Vite. The idea: paste your personal details once as free-text notes, then click "Autofill page" and let Jev match each detected form field to the best-fitting line from those notes. Rather than requiring you to map "shipping_address_1" to a specific value ahead of time, the extension sends field context and your notes to Jev, which returns a match.

The extension exposes a confidence threshold you can tune — lower fills more fields (with more risk of a wrong match), higher is more conservative. Notes are stored locally in browser storage, not synced anywhere, and the extension explicitly never sends password or payment fields to Jev or autofills them; that's a hardcoded exclusion, not a toggle. The project ships with test/sample-form.html for trying it before pointing it at a real site. As with typesafe-adblock, this is early software: 18 stars, 5 commits at the point we reviewed it, and no independent accuracy testing beyond the author's own description. Aiskill.market entry: /skills/jevfill-imohitmayank.

What both projects get right about disclosure

Neither extension pretends to be more mature than it is, and that's the more useful signal than either one's feature list. typesafe-adblock names its own failure modes instead of implying it's a uBlock replacement. jevfill hardcodes the one exclusion (password/payment fields) that actually matters for user trust, rather than leaving it as an option someone could misconfigure. Both require you to bring your own API key, which means you're opting into sending page content to a third-party service on every use — worth weighing against a rule-based blocker or a conventional autofill manager that never leaves your device.

Neither project publishes a comparison against the incumbent tool in its category (uBlock Origin for ads, browser-native or password-manager autofill for forms), so there's no benchmark to cite here. That's consistent with most of this series: small, new Jev projects tend to demonstrate the mechanism rather than prove it beats what you already use.

What to do next

1. Test typesafe-adblock on a page with known ads first, and check both what it catches and what it wrongly removes, before trusting it on a page you rely on.
2. Try jevfill on the bundled test/sample-form.html before your first real form, and start with a higher confidence threshold to limit wrong matches.
3. Confirm your API key usage and spend limits in the TypeSafe console before leaving either extension running continuously — both call Jev on every relevant page load.
4. Don't paste sensitive notes into jevfill beyond what you're comfortable having matched and (per its exclusion) never having sent for password/payment fields.
5. Compare against your incumbent tool. Neither extension published a head-to-head, so if ad-blocking or autofill accuracy matters to you, run your own side-by-side — see how to evaluate Jev projects honestly for a checklist, or browse the wider set on the Awesome Jev radar.