User repository on GitHub
A cyberpunk border encounter powered by TypeSafe Jev. Bluff the guard. Inspect the receipts.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Last Exit
Last Exit is a short cyberpunk checkpoint game where you type a cover story, present a manifest and offer a bribe to an inspector who decides whether to let you through. Each reply makes two Jev calls: one turns your statements into typed observations, the other picks his next action.
- Stars
- 0
- Language
- JavaScript
- Published by
- Community
- Author
- 0x963D
- Last verified
- Oct 2026
About this project
One gate. One good lie. There is something alive in your cargo. Convince the inspector there isn't. Play at gate.fade.tools · Findings · How it works. A small cyberpunk encounter built by Dan Doca. Type your own cover story, slide over a manifest, offer a bribe, and live with the inspector's decision. Inspired by the tension of the Nomad border crossing in Cyberpunk 2077; this is an original setting, character and artwork, not a mod or affiliated release.
Quoted from the project's README. Read the full README.
- Languages
- JavaScript
Highlights
- The inspector never sees the hidden cargo or the scenario flag, only the conversation, the papers and the evidence you have uncovered.
- Rules are code rather than model output: scans and searches have deterministic consequences, and claiming to hand over papers does not change inventory.
- The published batch of 100 scripted crossings cost about $0.057 in estimated input tokens, at a median local reply time of 537 ms.
- Authored dialogue lines and deterministic scans keep the fiction fixed, so the only variable across runs is what Jev decides to do.
Watch out
The published numbers come from 50 authored variants each run twice on a pinned Jev build, not 100 independent players, and the current request alias is jev-latest, so outcomes can move under you. Timings are local HTTP round trips covering two sequential model calls but excluding browser paint and typing, which the README says is not a hosted speed promise; the public deployment adds network and database overhead. It is explicit that a believable lie getting through is an intended game outcome and that this is neither a security nor a calibration benchmark. Local play needs a TypeSafe key and a small spend ledger, and scripts/stress.mjs runs paid playthroughs, so review its plan and budget before running it. 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.
No, and the README rules it out explicitly. A believable lie getting through is an intended game outcome rather than a finding, and the project states plainly that it is not a security or calibration benchmark. The batch figures, smugglers escaping 29 of 90 times, honest shipments passing 10 of 10 and undeclared lawn-mower stories passing 0 of 10, describe game difficulty across 50 authored variants run twice each, not model accuracy on adversarial input.
Less than it looks. One Jev call per reply reads your statements into typed observations and a second selects an available action; the inspector acts only on the conversation, the documents and the evidence he has discovered, never on hidden state. Dialogue lines are authored, scans and searches have deterministic consequences, and inventory changes only through the physical document and bribe controls, so saying you handed over the papers does nothing. Seven player turns maximum.
The published evaluation put 100 scripted crossings, 596 Jev requests across 298 turns, at roughly $0.057 of estimated input tokens. Local play draws on a small persistent allowance tracked in a ledger file, and deleting that file resets the accounting rather than the spend. If you deploy it yourself the public budget starts at $1 of estimated model input cost with no automatic refill, and each reply reserves its worst case before inference, refunding the unused part on success.
Replies are sent to TypeSafe for the decisions. Fictional transcripts are kept for a four-hour session, become inaccessible after expiry and are deleted on the next new crossing, and the README asks you to keep personal information out of the game. First-party daily counters record page views, crossings started, saved replies and outcomes with no visitor identifier, dialogue or referrer, rate limiting uses a keyed hash rather than storing your IP, and no third-party tracker or public analytics endpoint is installed.
Node.js 24 is recommended; the local game server uses only Node's standard library. Run npm ci then npm start and open http://127.0.0.1:9442. Connect a TypeSafe key at /setup.html or set TYPESAFE_API_KEY, which setup persists in the ignored .local/typesafe-key so it survives restarts. scripts/stress.mjs runs paid scripted playthroughs, so review its plan and budget before running it.
Code is MIT. The scene is original AI-generated artwork, and the fonts are Rajdhani and DM Sans under their SIL Open Font Licenses in public/fonts. It is an original setting, character and artwork inspired by the Nomad border crossing in Cyberpunk 2077, not a mod or an affiliated release, and not endorsed by TypeSafe or CD PROJEKT RED.
Vercel serves public/ and api/index.mjs while Neon stores sessions, rate counters and spend reservations. Configure DATABASE_URL, TYPESAFE_API_KEY and a random SESSION_SECRET server-side and apply db/schema.sql with scripts/migrate.mjs first; production uses budget public-v1 and previews preview-v1. Each IP gets 12 new crossings and 60 replies per hour through a keyed hash, which the README calls abuse deterrence rather than DDoS defence.
More Jev resources like this
0GitHub stars
Sloppy Jev's is a Manifest V3 Chrome extension that blurs AI-generated prose and ads, classifying extracted page text through jev-latest on the TypeSafe API. Plain JavaScript and CSS, no build step, MIT licensed, loaded unpacked.
Jev Demos & Experiments#community#jevbest#catalog-import
30GitHub stars
Side-by-side game harness where Laya running locally and Jev over API answer the same typed questions for snake, kombat, runner, flappy, tetris, and human-versus-model matches.
Jev Demos & Experiments#community#jevbest#catalog-import
13GitHub stars
Chrome extension that adds a second Twitch chat column showing only messages Jev rates as matching your chosen intent, with every row linked to its probability distribution.
Jev Demos & Experiments#community#jevbest#catalog-import
21GitHub stars
Browser top-down driving sim that sends sensor JSON to Jev every 200ms with four typed driving questions in one call, then applies confidence gating and reflex rules to steer.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
vinilana/live-jevJavaScript
vinilana/live-jevJavaScript12GitHub stars
Chrome extension that labels each X post with Substance, Humor, Chit-chat, Promo or Junk plus AI-written and off-topic flags, collapsing hidden posts to one line.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
bohutang/siftJavaScript
bohutang/siftJavaScript11GitHub stars
Chrome extension that labels each X post with intent plus rage-bait, synthetic, and undisclosed-ad signals from Jev, with threshold-only rendering and local caching.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
manifoldor/xtagsJavaScript
manifoldor/xtagsJavaScriptFrom 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.