Organization repository on GitHub
A Taboo-style parlour game for design systems, refereed by Jev (TypeSafe System One model)
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Component Charades
A guessing game where you describe a UI component to Jev without naming it, and the model re-ranks 135 components and puts live odds on every one of them.
- Stars
- 2
- Language
- TypeScript
- Published by
- Community
- Author
- southleft
- Last verified
- Oct 2026
About this project
It's Taboo for design systems, and the other player is a model that can't talk. You're dealt a UI component. Toast, say. You have to get Jev to guess it without using the word "toast", or "snackbar", or "notification", or any of the other words on the card. As you type, Jev re-ranks all 135 components in the library and puts odds on every single one. When it's sure enough, it buzzes. Fewer words, better score.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- One request every 350ms of typing; the odds board is the raw Choice distribution and the buzz is a threshold on confidence.
- Thresholds live in one questions file: BUZZ_IN at 0.85, LEANING at 0.5, TOSS_UP at 0.25, plus speculative side questions.
- Runs with no key on a keyword-overlap stand-in, and the usage strip says source: fixtures so it cannot be mistaken.
- Round tokens are HMAC-signed and banned words are re-checked server-side; the key is only ever read in route handlers.
Watch out
The accuracy figures come from 40 descriptions the author wrote himself, ranked against 135 hand-written component labels, and the documented misses are near-neighbours rather than errors, with one case where the model's answer looks better than the label. Jev is decisive once a description has a behaviour or a placement, so most rounds resolve within the first six words, which is also why the score is word count. The README's own lesson is that wording moved the results where no threshold did. Running the eval needs a key and costs about a cent. 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.
The app does. verdict.ts turns Jev's numbers into listening, leaning, buzzed, vague or not_a_component, and Jev never decides to buzz. The round is protected by HMAC-signed tokens so the browser cannot forge a win, and banned words are re-checked on the server rather than trusted to the client.
No. Those are 40 descriptions written by the author against 135 hand-written component labels, so it is a useful smoke test rather than a measurement. The misses it lists are genuinely arguable: agree to the terms ranking Button over Checkbox, or turn dark mode on ranking Theme Toggle over Switch, which the README concedes is probably the better answer. A near tie such as Segmented Control 0.52 versus Tabs 0.47 is flagged by the confidence value as the coin flip it is.
Yes. Without TYPESAFE_API_KEY, or with JEV_MODE=fixtures, the app answers from a keyword-overlap stand-in so you can work at zero cost. The usage strip at the bottom reads source: fixtures the whole time, so a fixture run can never be confused with the model. Only npm run eval needs a real key.
Finder mode handles it. You describe a job such as submits a form, and it names the component that does it with a confidence, then separately answers whether the library has anything for it at all. If it does not, you get to name the component that should exist instead of guessing at an existing option.
Below 0.5 confidence Jev is listening, between 0.5 and 0.85 it is leaning, and at 0.85 it commits right or wrong. Your score is how few words that took. The thresholds live in src/lib/jev/questions.ts as BUZZ_IN 0.85, LEANING 0.5 and TOSS_UP 0.25.
Round tokens are HMAC-signed so the browser cannot forge a win, banned words are re-checked server-side, and the API key is only ever read in route handlers and never reaches the browser.
The Vercel project is connected to the repo and pushes to main deploy. Two environment variables are needed: TYPESAFE_API_KEY and ROUND_SECRET, any random string. There is also an OpenNext Cloudflare config if you would rather run it on a Worker with npm run deploy on Node 22+.
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.