Neeko is an innovative framework for efficient multi-character role-playing (MCRP) agents, leveraging dynamic low-rank adapters (LoRA) to adapt seamlessly to diverse characters. Unlike existing methods, Neeko breaks down the role-playing process into three stages: agent pre-training, multiple characters playing, and character incremental learning. It uses distinct LoRA blocks for each character, enhancing adaptability to unique attributes, personalities, and speaking patterns. Neeko outperforms existing methods in MCRP, offering more engaging and versatile user interactions. The framework employs a Mix of Experts (MoE) gate mechanism to select and activate corresponding LoRA blocks for specific characters. For unseen characters, it uses fusion and expansion strategies to generate new LoRA blocks. Neeko's dynamic LoRA and gating modules enable it to handle both seen and unseen characters, with the number of LoRA blocks continuously increasing. The framework is evaluated using metrics for character, knowledge, and dialogue consistency, demonstrating superior performance in MCRP tasks. Experiments on the Character-LLM-Data dataset show that Neeko achieves the best overall performance, outperforming existing methods in stability, knowledge consistency, and role transfer. Neeko's memory usage and training time are comparable to LoRA, making it efficient for large-scale role-playing tasks. The framework addresses the challenge of multi-character role-playing by enabling agents to switch between roles and handle new characters effectively. Neeko's dynamic LoRA and gating mechanisms ensure accurate role-playing and consistent performance across different scenarios. The framework's contributions include formulating the MCRP task, proposing Neeko as an incremental role-playing agent, and conducting extensive experiments to validate its effectiveness. Neeko's approach offers a promising solution for improving the capabilities of role-playing agents in multi-character scenarios.Neeko is an innovative framework for efficient multi-character role-playing (MCRP) agents, leveraging dynamic low-rank adapters (LoRA) to adapt seamlessly to diverse characters. Unlike existing methods, Neeko breaks down the role-playing process into three stages: agent pre-training, multiple characters playing, and character incremental learning. It uses distinct LoRA blocks for each character, enhancing adaptability to unique attributes, personalities, and speaking patterns. Neeko outperforms existing methods in MCRP, offering more engaging and versatile user interactions. The framework employs a Mix of Experts (MoE) gate mechanism to select and activate corresponding LoRA blocks for specific characters. For unseen characters, it uses fusion and expansion strategies to generate new LoRA blocks. Neeko's dynamic LoRA and gating modules enable it to handle both seen and unseen characters, with the number of LoRA blocks continuously increasing. The framework is evaluated using metrics for character, knowledge, and dialogue consistency, demonstrating superior performance in MCRP tasks. Experiments on the Character-LLM-Data dataset show that Neeko achieves the best overall performance, outperforming existing methods in stability, knowledge consistency, and role transfer. Neeko's memory usage and training time are comparable to LoRA, making it efficient for large-scale role-playing tasks. The framework addresses the challenge of multi-character role-playing by enabling agents to switch between roles and handle new characters effectively. Neeko's dynamic LoRA and gating mechanisms ensure accurate role-playing and consistent performance across different scenarios. The framework's contributions include formulating the MCRP task, proposing Neeko as an incremental role-playing agent, and conducting extensive experiments to validate its effectiveness. Neeko's approach offers a promising solution for improving the capabilities of role-playing agents in multi-character scenarios.