PolicyArena

A simulation engine for game-theoretic agent research. Pit rule-based strategies, reinforcement learning, and LLM-powered agents against each other — same game, same seed, same metrics.

All built-in games are playable at policyarena.dev. New games added to the repo appear there automatically.

Quick links

• Getting Started — install, run your first simulation
• Architecture — how the engine works under the hood
• Configuration — YAML config reference
• Built-in Games — all available games
• API Reference — auto-generated from source

Install

pip install policy-arena          # core (rule-based + RL)
pip install policy-arena[llm]     # + LLM agents
pip install policy-arena[all]     # everything

Run

import policy_arena as pa

results = pa.run(pa.get_scenario_path("pd_rl_vs_rulebased"))
print(results.model_metrics.tail())

policy-arena run --example pd_rl_vs_rulebased --no-save