Organization repository on GitHub
CLI that finds PII in text with TypeSafe Jev: presence, sensitivity, and located spans
- Forks
- 2
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Jev Pii Checker
A CLI that scans text and files for PII in three layers — thirteen typed category questions plus a sensitivity score, regex extraction, and Intl.Segmenter name candidates judged by Jev — with masked output and exit codes.
- Stars
- 3
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- coo-quack
- Last verified
- Oct 2026
About this project
CLI for scanning text and files for PII using TypeSafe's Jev model, regex patterns, and word segmentation. 📖 Documentation — getting started, CLI reference, how the three layers work, categories and sensitivity levels, limitations. Gate — 13 PII category Nouls (person name, email/phone, postal address, date of birth, government ID, financial account, health info, biometric, IP address, SNS handle, employment, race/religion) + 3-level sensitivity score (none/low/high, IBM taxonomy)
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Thirteen category nouls plus a three-level sensitivity score on the IBM taxonomy, with regex and segmentation underneath.
- Values are masked as first2…last1 by default, and --fail-on high turns detected high-sensitivity findings into exit code 2.
- Gate accuracy was 100% on person-name, email-or-phone and phone presence over an IBM corpus, with 22 of 24 sensitivity levels right.
- Corporate noise is filtered explicitly: noreply@ addresses, 0120- phone prefixes and server IPs are excluded from the regex layer.
Watch out
The README leads with a warning worth repeating: the scanned text is sent to TypeSafe's Jev API, detection happens on their servers, and everything you scan leaves your machine. Its own advice is to use it only on data you are allowed to hand to a third party — text you were going to send to a model anyway — and to check TypeSafe's retention terms first; for a guard that never sends anything it points at a separate project. The accuracy figures are narrow: 100% gate accuracy on person-name, email-or-phone and phone presence over an IBM corpus, and 22 of 24 sensitivity levels. The limitations section admits borderline HR and disciplinary records flip between low and high across runs when probabilities sit near 0.5. No Luhn or passport format checking, so synthetic numbers get flagged; organization names are out of scope; and code snippets are treated as plain text, so a variable named user_id reads as PII. The --concurrency flag is documented but not implemented. 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.
Not as a local filter. The README's warning is that the scanned text is sent to TypeSafe's Jev API, detection happens on their servers, and everything you scan leaves your machine. Its recommendation is to use it only on data you are allowed to hand to a third party, such as text you were going to send to an LLM anyway, and to check TypeSafe's data-retention terms before scanning anything else. For a guard that never sends anything, it points to a different project.
It can flip between runs when the probabilities sit near 0.5 — the README gives a 0.47/0.53 boundary between low and high — and it names confidential performance reviews and disciplinary records as the cases where this bites. The suggested response is to run it multiple times for high-confidence assessments rather than trusting a single pass.
It can. JSON, SQL and regex patterns are treated as plain text, so variables named after PII attributes — person_name, user_id, phone_number — get over-reported. In the other direction it cannot identify organization or brand names at all, and public figures such as Natsume Soseki are flagged as person names, with names always treated as PII unless a human reviews them.
Exit codes are the integration surface: 0 for nothing found, 1 for an error, and 2 when the detected severity meets the --fail-on threshold, so --fail-on high turns a scan into a gate. It does not validate checksums — Luhn for credit cards and passport or ID formats are explicitly not verified, so synthetic and invalid numbers are reported as findings. Values are masked first2…last1 unless you pass --show-values for debugging.
Either bun install -g @coo-quack/jev-pii-checker, or bunx @coo-quack/jev-pii-checker with no install at all; Node.js 20+ or Bun covers the runtime. From source it is bun install then bun run build. It is MIT licensed, and TYPESAFE_API_KEY is the only credential it reads.
On the IBM corpus the gate reported 100% on person_name, email_or_phone and phone presence, and got 22 of 24 sensitivity levels right, with the seminar/clinic context distinction called out as working. Digit disambiguation was checked separately against credit cards, My Numbers, order numbers and product serials.
--threshold sets the gate probability, default 0.5; --span-threshold sets the Intl.Segmenter name-candidate bar, default 0.8; --max-chars sets the chunk size in bytes, default 4000. --no-spans skips name extraction and runs regex only, and --concurrency 4 is accepted but the README marks it as not yet implemented.
More Jev resources like this
98GitHub stars
Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
1Panel-dev/laya-serverPython
1Panel-dev/laya-serverPython20GitHub stars
General-purpose System One agent where Jev chooses tools and arguments while Jot executes them and loops until it can reply.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
runta-dev/jotTypeScript
runta-dev/jotTypeScript23GitHub stars
Browser drone-fleet lab where TypeSafe Jev steers each drone as fast reflexes and an optional planner offers one-use strategy advice when confidence drops, without pausing control.
Jev Demos & Experiments#community#jevbest#catalog-import
23GitHub stars
Chrome extension that sends each X post to Jev with six typed questions in one call near the viewport, then renders a behavior chip under the post plus a session cost and latency panel.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
oso95/x-scannerTypeScript
oso95/x-scannerTypeScript14GitHub stars
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.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
stefw/lkcleanTypeScript
stefw/lkcleanTypeScript11GitHub stars
Standalone Flue example that routes support chat through Cloudflare Workers AI, using gpt-4o-mini for chat and Jev for typed intent and urgency routing.
Jev Demos & Experiments#community#awesomejev#catalog-import
Community
matthewp/flue-jev-demoTypeScript
matthewp/flue-jev-demoTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.