Politiques et juges pour agents de jeu
157 projets · Tous les projets →
Il répond aux questions à choix, booléennes et à scores ordonnés depuis états et candidats, sans décoder de tokens de réponse. Ce modèle local distinct s'inspire de Jev ; les enregistrements comparent Jev et Qwen non ajusté sur les mêmes interfaces de jeu.
Le projet pilote un joueur de Minecraft Java 1.16.5 via Mineflayer, avec un modèle de planification séparé et Jev pour choisir les actions. Configurez le relais des modèles et les chemins locaux, installez les dépendances, lancez le serveur vanilla puis le parcours nether-agent documenté.
La collection fournit des compétences pour le tri de données, la vérification des preuves, l’évaluation par grille et le choix d’actions autorisées. Suivez le guide et choisissez Jev via TypeSafe ou OpenRouter, ou une simulation par l’agent clairement signalée.
TypeSafe Mario est une boucle NES expérimentale utilisant la télémétrie de l’émulateur et des observations issues de la RAM. Avec Python 3.13, une clé TypeSafe et le jeu local, inspectez state-demo avant de lancer play.
JEV-inspired parallel decisions for CUDA LLMs. One context, many decisions. vLLM API, game-agent examples, and reproducible benchmarks.
A Clash Royale bot with no trained policy: Jev (TypeSafe System One) makes every decision from the live game state
2D autonomous car simulation in the browser, driven by TypeSafe's Jev decision model
Real-time drone swarm autonomy simulation using Jev for fast System 1 reflex decisions and collision avoidance, with optional System 2 reasoning for strategic guidance
Aucune description disponible.
Python prototype that plays NES Super Mario Bros. from structured RAM observations, with Jev answering focused movement and jump questions.
Hermes Agent plans, Jev (TypeSafe) picks bounded actions, Mineflayer executes: Minecraft with no screenshots or keypresses from a model. Includes the reproduction of rmalde/minecraft-agent's Ender Dragon run.
Pokemon Red on PyBoy: code owns the route and the arithmetic, Jev picks at branches in about 100 ms, calibration measured instead of assumed
Chess where both players are TypeSafe's Jev model: every move is a typed Choice decision
Observable browser stealth game: Jev makes typed guard judgments while deterministic code owns the world.
Chrome dino game played by Typesafe AI Jev model
An F1 racing game where TypeSafe's Jev picks the racing line and the pedals, and learns each corner's limit between laps
Playing Pokemon Red using TypeSafe Jev
Benchmarks and a playground for TypeSafe's Jev (System One) model: chess, and who-is-the-player-talking-to for speech-to-text game NPCs
Browser driving game that captions dropped road images locally, then uses Jev Choice answers to drive the car while code owns physics and timing.
Jev, TypeSafe's System One model, plays chess against any OpenRouter LLM, Stockfish and you. One-page web app with live moves, Jev's move probabilities, saved games and win rates.
GPT teacher consults for DSH: stateless one-shot codex teachers behind a hard per-task consult budget, a read-only sandbox and a deterministic prefilter.
Chess against Jev, TypeSafe AI's System One model, through OpenRouter. Built with the pico framework.
A Minecraft survival bot that thinks with Jev — TypeSafe's System One AI model
AI life-and-civilization simulation: TypeSafe Jev makes every decision (typed, probabilistic, auditable); LLMs plan — OpenAI-compatible APIs, local models (Ollama, LM Studio), Claude Code, Codex.