Jev.aitools.fyi

Command Palette

Search for a command to run...

Product profile
Python259 starsUpdated 2026-09-17

typesafe-mario

NES RAM becomes JSON; Jev Choice picks legal Mario buttons every few frames with a live probability dashboard.

Overview

typesafe-mario lets Jev choose NES controller actions for Super Mario Bros from structured JSON built out of emulator telemetry and RAM. Faadil Shaik's launch clip shows real-time play without pixels in the prompt. The repo ships CLI tools, a live dashboard, and jsonl artifacts for overlays.

Problem: Vision pipelines for NES agents are heavy; Mario needs fast discrete inputs from reliable state, not screenshot captions.

Built for: Python developers benchmarking Jev on World 1-1 and anyone comparing RAM-first harness design.

First indexed on Jev Directory: 2026-09-16

Creator and team

Name
Faadil Shaik
Handle
@faadilhshaik
“got @typesafeai's new model Jev to play Super Mario Bros. fast inference + structured outputs makes it surprisingly good for real time use cases.”

How Jev is used

Role in the product flow
Per-step controller Choice from structured world model
Primitives
Choice
State in
player, trajectory, hazard, terrain, reaction_timing, and episode fields parsed from RAM (README architecture).
Decision out
One of noop, right, right_jump, right_run, right_run_jump, jump, left with probability distribution logged.
  1. NES emulator advances frames
  2. Parser builds JSON state without screenshots
  3. Jev Choice selects controller action
  4. Harness applies input for configured frames-per-decision (default 8 steps)

The harness keeps cardinality small and legal actions explicit, which is how Jev avoids hallucinated buttons. reaction_timing fields document observation-to-action delay for benchmarking. typesafe-mario state-demo prints payloads without launching the game. Play mode records artifacts/run-*.jsonl for clips and telemetry. You must supply your own lawful ROM and TYPESAFE_API_KEY.

Features and stack

Features

  • Python 3.13 CLI typesafe-mario play and state-demo
  • Live dashboard with probabilities and latency
  • jsonl decision logs for overlays
  • Documented action set and parser fields
  • Hundreds of GitHub stars in the public catalog snapshot

Stack

  • Python 3.13
  • NES emulator harness
  • TypeSafe Jev

Demo

Showcase clip with the same lightbox player as the homepage. Click to play.

Faadil Shaik@faadilhshaik

Super Mario Bros from RAM, not pixels

NES memory becomes JSON, Jev picks run or jump, and World 1-1 advances without a vision model cosplaying as a speedrunner.

gamesemulatordemo

FAQ

Does Jev see the screen?
No. README states the model receives structured JSON from telemetry and RAM, not screenshots.
How often does it decide?
Default play command uses one decision every eight emulator steps; adjust --frames-per-decision for benchmarks.

Related learn guides

Original Jev guidance that pairs with this product pattern.

  • Jev use cases

    The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.

Related products

Hand-picked neighbors with rich profiles or overlapping tags.

Games and simsPython
Demo
Modular emulator agent framework: PyBoy Pokemon Red first, with structured RAM schemas and Choice-driven tactics at System One latency.
gamesemulatoragents
Games and simsTypeScript1
A browser-native Doom agent experiment with structured spatial state, composable AI controls, live decision telemetry, and a Chocolate Doom WebAssembly
gamesbrowser-useagents
Games and sims
DemoFeatured
Shubhankar's FIFA-style soccer sim where every outfield player runs a ~150 ms Jev loop for tackle, pass, shoot, speed, and direction, plus Jev-driven commentary and soundtrack picks.

by @_shubhankar

gamessportsrealtime
By Rishit