User repository on GitHub
A TypeSafe/Jev agent that plays Super Mario Bros. from structured emulator state.
- Forks
- 62
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Cookbooks
TypeSafe Mario
An experimental controller that has Jev choose NES inputs to play Super Mario Bros. from structured emulator telemetry instead of screenshots, logging every decision with probabilities and latency.
- Stars
- 452
- Language
- Python
- Reading time
- 3 min
- Published by
- Community
- Author
- fhshaik
- Last verified
- Sep 2026
About this project
An experimental controller that lets TypeSafe's Jev model directly choose NES controller inputs for the original Super Mario Bros.
Quoted from the project's README. Read the full README.
Highlights
- Jev sees object-centric JSON: player motion, trajectory, up to three enemies, terrain, and reaction timing.
- Each decision mixes a Choice for the next action, a Noul for jump usefulness, and a Score for danger.
- Decisions are logged to artifacts/run-<timestamp>.jsonl with probabilities, latency, and outcome.
- The dashboard shows the probability distribution, confidence, Jev latency, danger score, and reward.
Watch out
No ROM is included, so you supply a legally obtained copy, plus Python 3.13 or newer and a TypeSafe API key.
Questions about this entry
Short answers drawn from the linked project or page.
Structured telemetry. The model receives no screenshots: emulator telemetry and RAM are parsed into object-centric JSON covering player motion, trajectory, up to three enemies, terrain, measured reaction timing, recent control results and episode progress. The raw tile grid stays in debug logs and the dashboard, and is not part of the model input.
It picks one of seven legal macros (noop, right, right_jump, right_run, right_run_jump, jump, left) as a Choice, plus a Noul for whether a forward jump is useful and a Score for immediate danger. Timing arithmetic stays in code as typed facts, but there is no scripted recovery-action override: Jev owns the choice.
Not by default. play defaults to one decision every eight emulator steps via --frames-per-decision, and every decision is written to artifacts/run-<timestamp>.jsonl with probabilities, confidence, latency, canonical model state and the game outcome. You can also run headless with --display none.
Yes. The repo contains no Nintendo ROM or other copyrighted game data, and sourcing the game and emulator lawfully is on you. The rest is Python 3.13 or newer and a TYPESAFE_API_KEY. Before any of that, typesafe-mario state-demo prints the exact JSON and text that would be sent, without launching the game or calling the API.
Object-centric JSON grouped by meaning, not pixels: player position, velocity, grounded state, jump phase and power-up; trajectory airtime and gap crossing; up to three enemies with projected positions and contact timing; terrain geometry; measured response delay; the last action's result; and episode lives, clock, progress and stalls.
Yes. typesafe-mario state-demo prints the exact JSON and text that would be sent to TypeSafe without launching the game or calling the API. The structured object is canonical, while the compact text view such as the local tile grid is described as being for debugging and the UI only.
The dashboard shows the selected action, the full Choice probability distribution, confidence, Jev latency, jump probability, danger score, reward and the parsed game state. Press R or click Restart for a fresh episode, and the dashboard stays open after death or level completion; --display game shows only the emulator and --display none runs headless.
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 camera-only MuJoCo quadrotor that uses Jev for tactical maneuvers at 2.5 Hz while classical perception and a 50 Hz reflex layer keep safety outside Jev, clearing a five-station course.
Jev Cookbooks#community#python#robotics
Community
RomanSlack/jev-dronePython
RomanSlack/jev-dronePythonA 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
Community
phyous/tsai-scPython
phyous/tsai-scPythonAn autonomous Pac-Man recreation steered by Jev: before each junction the game sends the maze, actors, and legal directions as structured data, and Jev returns one direction with a probability per option.
Jev CookbooksPlayground#community#playground#games
Community
A playable Tetris where you can drop pieces yourself or let the TypeSafe Choice primitive pick every move.
Jev CookbooksPlayground#community#playground#games
Community
A 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 fine-grained robot control study loop for LIBERO tasks: Jev layers an intent, a contact or motion family, and one of 27 control inputs, while reversible MuJoCo previews evaluate each candidate before execution.
Jev Cookbooks#community#python#robotics
Community
Dimweaker/jev-liberoPython
Dimweaker/jev-liberoPythonFrom 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.