In recent years, the rapid development of large language models have revolutionized human-machine interactions. These models have showcased remarkable potential in various tasks, such as open-domain conversations and language generation. However, due to the lack of long-term memory mechanisms, large language models struggle to effectively recall past conversations or remember user personality traits and roles, thus posing a significant challenge in models that require long-term interactions with the users. To address this issue, we propose a novel human-like memory mechanism named MindMemory. Inspired by human long-term memory cognition mechanisms, MindMemory dynamically recalls and updates memories to enhance user interaction. Drawing inspiration from the intricacies of human memory, it divides stored memories into four distinct categories: episodic, semantic, abstract and working. This allows for a more nuanced retrieval process, mimicking the way our minds sift through past experiences. Moreover, MindMemory actively adapts to changes in the user’s personality traits and character information through the lens of mind-based theory, which constructs a more comprehensive portrait of the user during prolonged engagements. By constantly refreshing its memory library, it ensures that interactions remain relevant and personalized over time, enhancing the user’s engagement. The experiment validates that large language models incorporating MindMemory mechanisms perform well in retrieving relevant memories from the past in long-term interaction scenarios. By understanding user roles and personalities, these models can generate higher quality dialogues, enhancing consistency in interactions and user engagement in dialogues.