The paper introduces ALCM, an autonomous framework that integrates large language models (LLMs) with data-driven causal discovery algorithms to generate more accurate and interpretable causal graphs. ALCM consists of three components: causal structure learning, causal wrapper, and LLM-driven refiner. The framework autonomously processes data to identify causal relationships, leveraging both statistical methods and LLMs to refine and enhance the causal graph. The causal structure learning component uses traditional algorithms like PC and LiNGAM to generate initial causal graphs. The causal wrapper translates these graphs into prompts for LLMs, which then refine the graphs by adding, removing, or reorienting edges based on contextual knowledge. The LLM-driven refiner improves the accuracy and completeness of the causal graph by incorporating insights from LLMs. The framework is evaluated on seven datasets, demonstrating superior performance compared to conventional causal discovery methods and LLM-based approaches. ALCM outperforms existing methods in precision, recall, F1-score, and accuracy, while showing a significant reduction in Normalized Hamming Distance (NHD), indicating a closer alignment with the ground truth. The hybrid version of ALCM, which combines PC and LiNGAM algorithms with LLMs, achieves even better results, showing the effectiveness of integrating traditional methods with LLMs. The study highlights the potential of combining LLMs with data-driven causal discovery to address challenges in causal inference, such as data sparsity, dynamic environments, and the complexity of causal graphs. ALCM provides a robust solution for generating accurate causal graphs, enhancing the reliability and interpretability of causal reasoning. The framework's ability to uncover hidden variables and causal relationships makes it a valuable tool for causal discovery in various domains. Future work includes integrating ALCM with knowledge graphs and Monte Carlo Tree Search to further enhance its capabilities.The paper introduces ALCM, an autonomous framework that integrates large language models (LLMs) with data-driven causal discovery algorithms to generate more accurate and interpretable causal graphs. ALCM consists of three components: causal structure learning, causal wrapper, and LLM-driven refiner. The framework autonomously processes data to identify causal relationships, leveraging both statistical methods and LLMs to refine and enhance the causal graph. The causal structure learning component uses traditional algorithms like PC and LiNGAM to generate initial causal graphs. The causal wrapper translates these graphs into prompts for LLMs, which then refine the graphs by adding, removing, or reorienting edges based on contextual knowledge. The LLM-driven refiner improves the accuracy and completeness of the causal graph by incorporating insights from LLMs. The framework is evaluated on seven datasets, demonstrating superior performance compared to conventional causal discovery methods and LLM-based approaches. ALCM outperforms existing methods in precision, recall, F1-score, and accuracy, while showing a significant reduction in Normalized Hamming Distance (NHD), indicating a closer alignment with the ground truth. The hybrid version of ALCM, which combines PC and LiNGAM algorithms with LLMs, achieves even better results, showing the effectiveness of integrating traditional methods with LLMs. The study highlights the potential of combining LLMs with data-driven causal discovery to address challenges in causal inference, such as data sparsity, dynamic environments, and the complexity of causal graphs. ALCM provides a robust solution for generating accurate causal graphs, enhancing the reliability and interpretability of causal reasoning. The framework's ability to uncover hidden variables and causal relationships makes it a valuable tool for causal discovery in various domains. Future work includes integrating ALCM with knowledge graphs and Monte Carlo Tree Search to further enhance its capabilities.