User repository on GitHub
- Forks
- 5
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Mario Jev
Python prototype that plays NES Super Mario Bros. by sending decoded RAM geometry to Jev, which answers movement and jump questions that code composes into controller buttons.
- Stars
- 15
- Language
- Python
- Published by
- Community
- Author
- shantanugoel
- Last verified
- Oct 2026
About this project
A uv-managed Python prototype that plays NES Super Mario Bros. (level 1-1 by default). Jev receives structured RAM observations and answers focused questions about movement, starting a jump, and sustaining a jump, plus timing hops under low ceilings. Code composes their answers into controller buttons. The emulator pauses while Jev responds, then advances up to four game frames by default, stopping early on landing. The resizable game window opens at 800×600 by default. No JavaScript is required.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Structured RAM observations covering position, velocity, terrain geometry, and jump corridors.
- Four Jev questions per call for movement, jump start, ceiling hops, and jump sustain.
- Landing-interrupted action holds with recorded actual frames, rewards, and collision estimates.
- Timestamped JSONL decision logs plus keyless replay that verifies positions and stops on divergence.
Watch out
Every Jev decision is a paid API request with retries disabled and errors halting the run, and reported positions are individual runs rather than a measured completion rate, with full-level completion not yet shown. 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.
Take the decisions-per-episode limit times the episode count, since each Jev answer is one paid request with a fixed HTTP timeout, and richer history raises token cost per decision without changing request count.
Yes. The scripted baseline policy and the state dump run with no key and no calls, so start there before spending on a short capped Jev run.
Four frames is the maximum hold with landing interruption asking Jev again immediately, which avoids hiding brief grounded states; six-frame actions would need fewer calls but react more slowly.
No. The scripted baseline and Jev runs report reached positions from individual attempts, and the guide states full-level completion is not yet demonstrated.
Run uv run mario-jev --policy scripted, which needs no API key, or add --headless for no game window and --episodes 3 to bound it. To inspect exactly what Jev will see without calling it, use uv run mario-jev --dump-state --headless.
Timestamped runs/*.jsonl files hold the configuration, input state, controller choice, Jev confidence and probabilities, token usage, API latency, actual frames executed, reward, next position, and episode summaries; they are decision logs, not saved emulator states. Replay with uv run mario-jev --replay runs/<timestamp>.jsonl and optional --speed 2 or --headless, which uses no API calls and stops on divergence.
Jev answers movement (run, walk, brake, or wait), start_jump, ceiling_hop, and sustain_jump in one API call, all as Noul probabilities with a 0.5 threshold. Code selects start-jump only when grounded with A previously released, uses sustain-jump while airborne, and on an enemy approach under low bricks without a competing pipe or pit it uses the focused ceiling_hop answer instead.
More Jev resources like this
46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPython55GitHub stars
Toolkit that makes a model behind an OpenAI-compatible endpoint or running in-process answer typed choice, score, and yes/no questions from next-token probabilities in Jev API shape.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
fidecastro/jevifyPython
fidecastro/jevifyPython164GitHub stars
End-to-end toolkit that synthesizes decision data, trains and calibrates candidate scorers, evaluates fixed splits, and serves Jev-compatible inference for interactive web decisions.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
zwliJay/jev-forgePython
zwliJay/jev-forgePython70GitHub stars
Self-hosted proxy that records typed decisions, distils them into small heads on a frozen encoder, and answers locally when confident.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
bladedevoff/stuntdPython
bladedevoff/stuntdPython42GitHub stars
Open harness that renders allowed JSON answers on a diffusion canvas, denoises for a fixed step budget, then picks the top allowed tokens from final logits with no training.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
JoshuaSP/open-jevPython
JoshuaSP/open-jevPythonFrom 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.