Large language models (LLMs) show great potential in the next point-of-interest (POI) recommendation. Compared with classical DL-based methods that focus on capturing various kinds of preferences, LLM-based methods are able to further analyze the candidate POIs based on common sense, providing corresponding reasons. However, the existing methods extract only a part of the user’s data as a context input, resulting in inadequate user preference capture and insufficient cooperative signal injection. Therefore, we propose PSLMRec, a novel framework enabling LLMs and plugin models to interact synergistically. Specifically, we use a novel lightweight temporal knowledge graph reasoning model as a plugin model. The plugin model can make adjustments and additions to the input of LLMs. It also guides LLMs to concentrate on the reasoning process related to various fine-grained preferences. Extensive experiments on three real-world datasets demonstrate the efficacy of our proposed PSLMRec.