User repository on GitHub
Observable browser stealth game: Jev makes typed guard judgments while deterministic code owns the world.
- Forks
- 1
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Cookbooks
HEIST//ONE
An observable stealth game where Jev supplies guards' split-second judgments while deterministic code owns the 30 Hz simulation, validates every proposal, and exposes each decision in a Decision Lens.
- Stars
- 8
- Language
- TypeScript
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- AbdelStark
- Last verified
- Sep 2026
About this project
An observable stealth game where Jev supplies guards' split-second judgments while deterministic code remains in control of the world. Watch the 37-second film · Inspect the live-run evidence · Read the technical specification.
Quoted from the project's README. Read the full README.
Used inJev for Gaming
Highlights
- One batched typed request covers four Jev judgments per guard, and code rejects illegal targets, stale context, and policy-incompatible actions.
- Scripted mode is the default, so the museum runs with no account or key; Node.js 22.13+ and pnpm 11 are required.
- A verified live run completed a clean extraction: 12 batched requests, 288 typed judgments, 12 of 12 responses, and 0 fallbacks.
- Median adapter latency was 259.9 ms and p95 344.1 ms; the run used 63,349 input and 12,181 output tokens.
- A Decision Lens shows evidence, distributions, confidence, and applied intent, alongside append-only JSONL traces.
Quickstart
bash
git clone https://github.com/AbdelStark/heist-one.git
cd heist-one
corepack enable
pnpm install --frozen-lockfile
pnpm devWatch out
MIT-licensed; requires Node.js 22.13 or newer and pnpm 11, live play needs a TypeSafe key with a call limit and deadline guard, and it is an independent community experiment not affiliated with TypeSafe AI.
Questions about this entry
Short answers drawn from the linked project or page.
Only advising. Jev proposes whether local evidence suggests a threat, how suspicious a guard should be, which legal tactical intent fits, and which observed entity deserves attention. Deterministic code owns the 30 Hz simulation, visibility, collision, navigation, physics, legal actions, access control, inventory, alarms, win/loss and lockdown. The server rejects illegal targets, stale context revisions and policy-incompatible actions before anything takes effect.
No. Scripted mode is the default, so the museum runs with no account or key. Live decisions need early-access credentials with HEIST_DECISION_MODE=jev and TYPESAFE_API_KEY in .env on the server. A missing key or a failed decision falls back to the scripted policy, and the credential never reaches the browser, logs, traces, fixtures or media.
No, and the README says so. It is evidence of one end-to-end run, explicitly not a benchmark or a claim that Jev beats scripted game AI. That run completed a clean extraction with 12 batched requests, 288 typed judgments, 12 of 12 responses, 0 decision errors, 0 fallbacks, 0 stale decisions, median adapter latency 259.9 ms and p95 344.1 ms. Exact provider cost was not exposed by the response.
That verified run used 63,349 input and 12,181 output tokens for the full jacket, badge, vault, artifact, exit route. Nothing is spent until you leave the briefing, calls are event-driven and bounded by HEIST_LIVE_CALL_LIMIT with a deadline, and the README does not convert those tokens into a currency figure.
One deterministic museum with six guards, three access-controlled doors, disguises, credentials, light control, noise decoys, one artifact and two exits. Move with WASD or arrows, sprint with Shift, crouch with C, interact with E, throw a decoy with Q, and click a guard to pin it in the Decision Lens. There is a clean route but the game does not prescribe it.
The full applied judgment for any guard you select: the evidence it received, its probability distributions, confidence, the proposed intent, the intent actually applied, latency, and whether it fell back. Append-only JSONL traces sit behind it covering requests, responses, material world events, outcomes, usage, fallbacks and deterministic checkpoints, so a run can be audited after the fact rather than taken on trust.
apps/ holds the Fastify authority server with the Jev adapter and trace writer, a React plus Phaser client with the live Decision Lens, and a Remotion video project. packages/game holds the pure simulation, world projection, policy and scripted adapter. The simulation imports a small DecisionEngine interface and never the TypeSafe SDK, which keeps provider access server-side. Node.js 22.13 or newer and pnpm 11, MIT licensed.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Software Bombe and live Enigma demo where Jev ranks cribs and judges plaintext.
ag_212 pointsShow HN: Jev as the Bletchley Park Analyst: A Software Bombe on EnigmaRead the thread on Hacker NewsHacker News: Show HN for public Minecraft PvP where a diffusion LM picks moves from structured state.
2 pointsShow HN: Minecraft PvP against DiffusionGemma-as-Jev, 24 ms per decisionRead the thread on Hacker NewsHacker News: Show HN: Jev finished a Pokémon Red run, with the Hall of Fame stats on the landing page.
pancomplex272 pointsShow HN: Jev Plays Pokémon RedRead the thread on Hacker NewsHacker News: Show HN: a mouse colony where each next move is Jev or a fixed rule.
1 pointShow HN: Jev-mice – mouse colony simulation using Jev and deterministic engineRead the thread on Hacker NewsMore Jev resources like this
A browser Snake game where code computes the legal moves, food distance, and reachable space and Jev picks one move per tick with a single Choice question; a late answer leaves the snake going straight.
Jev Cookbooks#community#typescript#games
Community
sorrycc/typesafe-snakeTypeScript
sorrycc/typesafe-snakeTypeScriptTwo WebAssembly builds of the Chocolate Doom 3.1.1 engine play the same Freedoom map from the same initial state, with a Jev Choice decision picking each tactical macro and a local controller turning it into inputs.
Jev Cookbooks#community#typescript#games
Community
lukaske/jev-doom-agentTypeScript
lukaske/jev-doom-agentTypeScriptAn original Civilization II browser harness for Jev: named Choice vectors cover empire policy, cities, research, diplomacy, units, exploration, and warfare, with live decision probabilities in a Roman-themed spectator display.
Jev Cookbooks#community#python#games
Community
phyous/tsai-civ2Python
phyous/tsai-civ2PythonA side-by-side T-Rex game where local Laya runs on Apple Silicon through MLX and hosted Jev uses the TypeSafe API, showing choices, latency, survival streaks, and crash replays on one course.
Jev Cookbooks#community#python#games
Community
virajbhartiya/laya-vs-jevPython
virajbhartiya/laya-vs-jevPythonA System One harness that plays the original StarCraft shareware mission Strongarm by reading structured game state and issuing mouse and keyboard commands from Jev choices.
Jev Cookbooks#community#python#games
Communityphyous/tsai-scPython
phyous/tsai-scPythonInteractive browser experiments that expose Jev's judgments: a support-message Decision Lab and a driving simulation where structured sensor state becomes typed steer, brake, and overtake decisions.
Jev Cookbooks#community#javascript#demo
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.