User repository on GitHub
Jev (TypeSafe AI) PoC through Game of Thrones
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Got Jev
A Game of Thrones roleplay where a narrator model writes each scene and Jev answers five typed questions about it, with the location, music, artwork and next prompt all following from those answers.
- Stars
- 2
- Language
- TypeScript
- Published by
- Community
- Author
- phureewat29
- Last verified
- Oct 2026
About this project
A Game of Thrones roleplay where you play Jon Snow. Each turn a story model writes the next scene, and then TypeSafe's Jev reads that scene back and answers five questions about it: where Jon now stands, what kind of scene it was, how much danger he is in, what should play under it, and whether the prose stayed inside the fiction.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Jev never writes prose. It answers where Jon stands, the scene beat, the mood, the danger and whether the text stayed in the fiction.
- If the in-fiction check falls below 0.5 the scene is regenerated once, and stored answers can be re-resolved later without asking Jev again.
- Model calls are server-side only, and a turn-number mismatch answers 409 turn_conflict so two open tabs stay honest without a lock.
- Spend is bounded for a public demo: 15 turns per session, 200 characters per action, plus per-minute, per-IP and process-wide caps.
Watch out
Two paid keys are required, one for TypeSafe and one for OpenRouter, and stories live in Redis, so this is not a single-binary demo; Node 20.9+ and pnpm are assumed. Turn 15 is narrated as a closing chapter and an untouched story expires after the configured number of days, so sessions are deliberately short-lived. Your typed action is passed to the narrator as delimited data rather than instructions, with the in-fiction check applied to the result, but the README still says plainly not to put anything private in it. Rate limits are keyed to the caller address, so a new tale mints a new session while only the address stays durable. The world is George R. R. Martin's and the project is described as a non-commercial technical demonstration. 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.
Judging only. A narrator model reached through OpenRouter writes the scene, and Jev then answers five questions about that text: a location id from a closed set, the beat, the mood, a danger score and whether the prose stayed inside the fiction. The README is explicit that Jev never writes anything, it labels the story scenario. The location header, the soundtrack, the artwork behind the page and the next turn's prompt are all computed from those answers.
Your action is passed to the narrator as delimited data, never as instructions, and an in-fiction check runs on the result. When that check falls below 0.5 the README says the scene is regenerated once. If an attempt still lands, the flow says attempts to talk to the narrator get redirected in-world rather than refused, so read that as a prompt-injection behavior to know about rather than a security boundary.
Stories live in Redis, and the quick start runs a redis:alpine container and points REDIS_URL at it, or at a rediss endpoint in a deployment. An untouched story expires after the configured number of days, which defaults to seven, so nothing is kept indefinitely. Three values are required, two API keys and the Redis URL, with optional overrides for both models and a CDN URL for serving the music and artwork.
Four caps guard it: ten turns per minute per caller, sixty per caller per address per day, five hundred process-wide per day, and fifteen turns per session with two hundred characters per action. Sessions are short by design because turn 15 is narrated as a closing chapter. Tabs are handled by sending the turn number the client has seen with each request; a mismatch answers 409 turn_conflict, which keeps two open tabs consistent without taking a lock.
Location, beat, mood, danger, and inFiction. The answers are values: a location id from a closed set, a probability distribution, and a score. The header, soundtrack, backdrop, and the next turn's prompt are all functions of those same five answers, and Jev never writes the story.
It needs Node 20.9 or newer (24 recommended) and pnpm. Start Redis with docker run -d --name story-redis -p 6379:6379 redis:alpine, then pnpm install, copy .env.example to .env.local filling TYPESAFE_API_KEY, OPENROUTER_API_KEY, and REDIS_URL, and run pnpm dev on http://localhost:3000.
MAX_TURNS defaults to 15, with turn 15 narrated as a closing chapter, and actions are capped at 200 characters. Rate limits are 10 turns per minute per caller, 60 per IP per day, and 500 process-wide per day, and an untouched story expires after 7 days.
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-serverPython38GitHub stars
Standalone CLI skill that drives isolated Playwright Chromium with Jev choosing scoped in-page actions, defaulting to Vercel AI Gateway with TypeSafe as an option.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
hqman/jev-browser-skillTypeScript
hqman/jev-browser-skillTypeScript41GitHub stars
Browser-focused TypeScript library for typed decisions that runs open model reproductions on-device via Transformers.js, returning one calibrated distribution per question.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
nico-martin/open-jevTypeScript
nico-martin/open-jevTypeScript34GitHub stars
Interactive playground that teaches Jev by running real typed noul, choice, and score questions and showing the resulting probabilities.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
davila7/jev-explainedTypeScript
davila7/jev-explainedTypeScript47GitHub stars
Zero-dependency Node 22 adapter that puts typed Jev decisions beside any LLM, with swappable backends, confidence gates, MCP tools, an ACP sidecar, and offline-ready recipes.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
gamesonrblx/JevbridgeTypeScript
gamesonrblx/JevbridgeTypeScript22GitHub stars
Local browser-agent extension where FX plans tasks and Jev picks constrained actions from observed elements, with model calls routed through Vercel AI Gateway.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
razaanstha/ulkaTypeScript
razaanstha/ulkaTypeScriptFrom 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.