Show HN: Teaching a 30ms local model, play browser games with LLM written rules

Category: ai-ml

Tags: llm, browser-games, game-playing, reinforcement-learning, typescript

Score: 7.3/10 (Innovation: 8, Technical: 8, Documentation: 7, Utility: 6)

IronBee Gamer plays browser games automatically by having an LLM write game-specific perception and decision rules, then using a fast local model (Laya) for real-time decisions. It works without game APIs, using canvas or engine state extraction, and includes a training pipeline, Hugging Face library, and a web UI. The project is interesting for combining LLM-guided rule generation with a lightweight decision engine for interactive play.

Target audience: AI researchers, game developers, hobbyists interested in LLM-driven automation

Repository: https://github.com/ironbee-ai/ironbee-gamer · TypeScript · NOASSERTION · 7 stars

View on Hacker News