jev-arena-nanojev
Fully local grid tactics game where the open NanoJev model picks each move, attack, heal or dash from real candidate actions.
Overview
Jev Arena by liao96312 is a fully local grid-based tactical game and a reproducible test bed for NanoJev, an open Jev-style model, alongside rule-based agents and search algorithms. Instead of scripted battles, at each step the model faces the real candidate actions and decides among moving, attacking, pushing, shooting, healing, dashing, using an EMP or waiting, with two action points per round. The player must collect gems and survive into the next level.
Difficulty comes from enemy intent, hazardous terrain, limited ammo, skill cooldowns, environmental kills and an accelerating pace, with five enemy types and boss fights every ten levels up to level 100. The Chinese Pygame interface shows model probabilities, reasons, inference time, difficulty, weapons and skills in real time. A default hybrid strategy adds tactical guidance for bosses, while a pure-model mode keeps the raw model choice. The README covers data generation and training on a GTX 1660S, testing, and multilingual documentation. It is MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which candidate action should be taken this step (move, attack, shoot, heal, dash, EMP, wait)? | move · attack · push · shoot · heal · dash · EMP · wait |
How it works
- Game engine enumerates legal candidate actions
- Local NanoJev service scores them
- Hybrid or pure-model policy picks the action
- Engine resolves the turn and displays probabilities
Strengths
- Fully local; no API key or cloud
- Compares model, rules and search on the same game
- Rich environment for testing decision models
Caveats
- Uses NanoJev, not hosted Jev
- Interface primarily Chinese
At a glance
- Type
- Project
- Status
- research
- Language
- Python
- Runtime
- Pygame with local NanoJev service
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev Arena is a fully local grid tactical game arena where NanoJev, rule agents and search algorithms make per-step move, attack, shoot, heal, dash and environment-interaction decisions across multiple levels with a Chinese Pygame interface.
logicrw/awesome-jev-projects完全本地的 NanoJev 网格决策游戏实验场，支持中文 Pygame、多关卡与 GTX 1660S 训练.
hellogumbo/awesome-jev
Frequently asked questions
Which model plays?
NanoJev, an open Jev-style model running locally.
Can I compare against other agents?
Yes, rule agents and search algorithms are included.
Can I train the model?
The README documents data generation and training, tested on a GTX 1660S.
Related projects
Real-time bot battles where you write a fighter's strategy in plain English and a System One model pilots it several times a second.
ProjectbetaTypeScriptBot that plays Red Alert 2 web skirmishes, with code handling mechanics and a decision model making stance, threat and timing calls.
ServicebetaPythonMultiplayer 3D science-fantasy version of Houston's River Oaks District with Jev-powered NPC companion, chauffeur and smart drive.
ProjectbetaTypeScriptBrowser 3D planetary rover sandbox where a TypeSafe controller chooses the next action from complete candidates, versus a baseline.
ProjectbetaTypeScriptKorean RL tycoon game training a company decision model that answers situations with choice, probabilities and confidence.
ProjectprototypePythonDoom runs as a spacecraft payload over real flight software; Jev pilots from the ground every half second while Claude reviews.
ProjectprototypePython