The Character-LLM project aims to train large language models to mimic historical figures such as Beethoven and Julius Caesar. This trainable intelligent agent differs from conventional prompt-based agents, being specifically trained for role-playing and capable of embodying specific historical characters. Researchers employ "experience reconstruction" techniques to generate detailed and diverse experiential data, enabling the agent to simulate the language style and thought patterns of the characters. Character-LLM is a powerful tool for role-playing enthusiasts and researchers in the field of natural language processing.