Temporal Knowledge Graph reasoning frequently involves inferring missing factual components over time. Traditional methods have notable limitations: (1) they often fail to adequately model semantic dependencies within events, and (2) they either overlook temporal relationships or lack the ability to recognize precise temporal information between events. To address these challenges, we introduce the TKGMamba framework. This framework incorporates an event encoder to capture semantic dependencies within events and utilizes Mamba to model temporal dependencies. Additionally, we utilize a time prediction loss function to enhance the model’s time-awareness capabilities. Our experimental results demonstrate that TKGMamba achieves competitive performance relative to Transformer-based methods while requiring fewer trainable parameters and offering greater computational efficiency. These findings suggest that TKGMamba provides a more robust and efficient approach for temporal knowledge graph reasoning.