I built AI agents that play Pokemon Showdown autonomously using free LLM APIs via tool-calling [P]

I've built a system where models like Llama 3, Qwen, and Gemma play Pokémon Showdown battles autonomously. Instead of simple prompt-response, they analyze the full battle state every turn (type matchups, HP, weather, field conditions, revealed opponent info) and decide whether to attack or switch using structured tool calls. The cool part: I routed everything through LiteLLM and exclusively used models with free API tiers (Groq, Cerebras, OpenRouter, Google AI Studio). So anyone can run this lo