What does Jev decide in jev-goal-reflex?

Every movement decision: the README's architecture table lists Jev as the fast reflex layer deciding whether to move, turn, and jump on each tick, while an OpenAI model (gpt-5.4-mini) plans the slow goal once per instruction.

How do the model answers turn into movement?

Code converts them: each decision asks Jev two Score questions (move, turn) and one Noul (jump), with code acting past a 0.33 dead zone, jumping above 0.45, holding unused axes still, and sending no Jev request while no step is active.

What do I need installed before this runs?

Node 20 or later, pnpm, an OpenAI API key and a TypeSafe API key. Create a .env in the project root holding OPENAI_API_KEY and TYPESAFE_API_KEY, run pnpm backend, then pnpm start in a second terminal and open http://localhost:5173 to type an instruction.

What kinds of instructions can I actually type?

Plain-language movement like go to the top right, go to center right then center left then center, or jump and do a 180 in the air. Each step in the plan holds actions that run at the same time, go_to, move, rotate, jump and wait, and Jev decides direction while code decides how much.

Can I see what Jev is being asked and what it answered?

Yes. The side panels show the plan, the exact state and questions sent to Jev, and Jev's answers, and the sliders set top speed and turn rate. docs/jev.md documents the Jev questions and the state Jev receives, while docs/architecture.md covers components, plans, steps, actions and the engine loops.

What license is it under?

MIT. It is a local demo rather than a packaged game, needing both an OpenAI key for planning and a TypeSafe key for Jev, and it also ships pnpm observe for live snapshots, pnpm typecheck, and Biome lint and format commands.

Can I run a second instance for testing without disturbing the first?

Yes. pnpm backend accepts a --port flag, so pnpm backend --port 8766 starts a second backend alongside the default one, and pnpm observe can be pointed at it with --port 8766 to print live snapshots from that instance.

CommunityJev Demos & Experiments

Jev Goal Reflex

Three.js simulation where an OpenAI model turns each plain-language instruction into a plan and Jev makes every reflex decision: two Score questions for move and turn plus one Noul for jump, with code acting only past a 0.33 dead zone and sending no Jev request while no step is active.

See highlights

Open github.com

Stars: 0
Language: TypeScript
License: MIT
Published by: Community
Last verified: Oct 2026

blakeandrewwood/jev-goal-reflex

User repository on GitHub

View on GitHub

Steer a 3D box with plain-language goals: an LLM (System 2) plans each instruction, TypeSafe Jev (System 1) makes every move, turn and jump decision in real time. Three.js + TypeScript.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Steer a box around a 3D world with plain-language movement instructions: where to go, how to move and turn, and when to jump. An LLM turns each instruction into a plan, and Jev makes every movement decision in real time. Each step holds actions that run at the same time: goto, move, rotate, jump and wait. Jev decides direction. Code decides how much. An OpenAI API key and a TypeSafe API key. Create .env in the project root: Start the viewer in a second terminal:

Quoted from the project's README. Read the full README.

Tagscommunity threejs simulation typescript

Works forJev for Realtime control

Used inJev for Gaming

Languages: TypeScript
Platforms: Browser
Capabilities: Open sourceSelf-hostedRequires API key

Highlights

gpt-5.4-mini plans once per instruction; Jev decides movement every tick, while TypeScript control code caps speed and turn rate and applies physics.
Each step holds simultaneous actions - go_to, move, rotate, jump, wait - where Jev decides direction and code decides how much.
Side panels show the plan, the exact state and questions sent to Jev, and Jev's answers; sliders set top speed and turn rate.
Requires Node 20+, pnpm, an OpenAI API key, and a TypeSafe API key; pnpm backend and pnpm start run the backend and viewer.
docs/jev.md documents the Jev questions and the state Jev receives.

Watch out

MIT license; it needs both an OpenAI key for planning and a TypeSafe key for Jev, and it is a local demo rather than a packaged game.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-plays

A Craftax (Crafter) survival harness where deterministic code lists every feasible action with its facts and Jev picks one Choice per step, optionally guided by an LLM-written objective and standing rules; three-seed ablations are included.

Jev Demos & Experiments#community#craftax#games

Community

mansicer/jev-playsPython

FlightBench

Fixed-wing landing simulator where a human can compete against Jev: four Jev Choice decisions map to aileron, elevator, rudder and throttle commands in JSBSim, run in real time. The benchmark also compares Jev, classical controllers and Codex models on landing score and API cost.

Jev Demos & Experiments#community#simulation#aviation

Community

AlperKartkaya/FlightBenchPython

Pacman AI Race

Browser Pac-Man race where a deterministic three-junction simulation filters fatal routes, then Jev makes one typed Choice among the surviving route IDs while the server rejects any answer outside the supplied set. Self-hosted Laya runs the same contract for comparison.

Jev Demos & Experiments#community#gaming#simulation

Community

MaryNfs/pacman-ai-raceJavaScript

Cua

29.3k

The Cua computer-use platform documents a jev-use recipe where TypeSafe Jev picks bounded action IDs and Cua Driver executes clicks and fills in the browser.

Jev Demos & Experiments#community#browser#computer-use

Community

trycua/cuaRust

JevPilot Reflex

Three.js autonomous-driving simulator that splits work between a System 2 slow planner and a Jev-style System 1 reflex loop at 60 Hz, with a safety brake that overrides the planner and applies full braking when time-to-collision drops below threshold.

Jev Demos & Experiments#community#threejs#simulation

Community

manhua-man/jev-pilot-reflexJavaScript

Jev Plays Pokémon Red

127

A completed Pokémon Red run: Jev picked menus, catches, and moves from RAM, and the harness handled pathfinding. The public page records the Hall of Fame stats.

Jev Demos & Experiments#community#games#javascript

Community

christianmat/jev-pokemonTypeScript

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

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

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai