XRec is a model-agnostic framework that leverages Large Language Models (LLMs) to provide explainable recommendations. It integrates collaborative signals and a lightweight collaborative adaptor to enable LLMs to understand complex user-item interactions and user preferences. The framework uses graph neural networks (GNNs) to encode collaborative relationships and then integrates these signals into LLMs to generate comprehensive explanations. Extensive experiments show that XRec outperforms baseline approaches in explainable recommendation systems, generating meaningful and accurate explanations. The framework is open-sourced and can be used for various recommendation tasks. XRec addresses the challenge of data sparsity and zero-shot scenarios by leveraging collaborative information and maintaining high explainability and stability. The model's robustness is demonstrated across different datasets and scenarios, showing its effectiveness in real-world applications. The framework also incorporates user and item profiles to enhance the quality of explanations and improve the model's generalization capabilities. XRec's ability to generate unique and personalized explanations highlights its potential in enhancing user understanding and trust in recommendation systems. The model's success in zero-shot learning confirms its robust generalization capabilities and its potential to mitigate the cold-start problem. The framework's integration of GNNs and LLMs enables it to capture intricate user dependencies and provide insightful explanations for recommendation outputs. XRec's approach offers a novel solution to the challenge of explainable recommendations, combining the strengths of collaborative filtering and language models to enhance the transparency and effectiveness of recommendation systems.XRec is a model-agnostic framework that leverages Large Language Models (LLMs) to provide explainable recommendations. It integrates collaborative signals and a lightweight collaborative adaptor to enable LLMs to understand complex user-item interactions and user preferences. The framework uses graph neural networks (GNNs) to encode collaborative relationships and then integrates these signals into LLMs to generate comprehensive explanations. Extensive experiments show that XRec outperforms baseline approaches in explainable recommendation systems, generating meaningful and accurate explanations. The framework is open-sourced and can be used for various recommendation tasks. XRec addresses the challenge of data sparsity and zero-shot scenarios by leveraging collaborative information and maintaining high explainability and stability. The model's robustness is demonstrated across different datasets and scenarios, showing its effectiveness in real-world applications. The framework also incorporates user and item profiles to enhance the quality of explanations and improve the model's generalization capabilities. XRec's ability to generate unique and personalized explanations highlights its potential in enhancing user understanding and trust in recommendation systems. The model's success in zero-shot learning confirms its robust generalization capabilities and its potential to mitigate the cold-start problem. The framework's integration of GNNs and LLMs enables it to capture intricate user dependencies and provide insightful explanations for recommendation outputs. XRec's approach offers a novel solution to the challenge of explainable recommendations, combining the strengths of collaborative filtering and language models to enhance the transparency and effectiveness of recommendation systems.