In-Context Reinforcement Learning framework for LLMs — no fine-tuning required.