🧠
Tabular Q-Learning
Classic TD learning with configurable learning rate (α), discount factor (γ), and exploration rate (ε). Supports both full TD updates and simplified terminal-state updates.
💾
Pluggable Storage
Three built-in backends: node:sqlite (zero deps, recommended), better-sqlite3 (high performance), and in-memory (testing). Easy to implement custom backends.
🎮
Custom Environments
Implement the Env interface for any RL problem. Includes RockPaperScissorsEnv with 4 opponent strategies for quick testing.