Ghost in the Minecraft (GITM) integrates Large Language Models with text-based knowledge and memory to create generally capable agents in open-world Minecraft environments. The framework achieves 67.5% success rate on the ObtainDiamond task, outperforming prior approaches like OpenAI VPT by 47.5 percentage points. Unlike traditional RL approaches requiring thousands of GPU days, GITM requires no GPUs and trains on a single CPU node in 2 days. It demonstrates LLMs potential for handling long-horizon complex tasks and adapting to uncertainties in open-world environments.