A fully local AI environment where two (or more) LLM agents debate, argue, collaborate, or meltdown in real time. Built with Python + Streamlit + llama-cpp, Agent Arena lets you load quantized GGUF models, assign personalities, configure behavior, and watch unsupervised agent-to-agent. Includes conversation logging, personality creation.