This paper introduces a novel multi-agent communication framework to enhance the autonomous problem-solving capabilities of Large Language Models (LLMs). The framework employs multiple LLM agents, each with a distinct persona, engaged in role-playing communication to address diverse problem scenarios. Inspired by the CAMEL model, the framework leverages chain-of-thought prompting to enable collaborative problem-solving. The agents work together to devise solutions for novel problems, with each agent guided by a tailored chain-of-thought prompt. The framework's versatility allows for the incorporation of any persona and chain-of-thought prompt, aligning with the specific problem to be addressed. The framework is available at https://github.com/sumedhrasal/simulation, built on top of CAMEL’s and ChatDev’s framework. In experiments, the framework was tested on arithmetic reasoning tasks using the GSM8K and SVAMP datasets. The multi-agent approach significantly improved accuracy compared to single-agent models. For example, the multi-agent GPT-3 achieved 55% accuracy on GSM8K and 77% on SVAMP, surpassing other models. In commonsense reasoning tasks using the CSQA dataset, the multi-agent approach achieved 83% accuracy, demonstrating its effectiveness in enhancing reasoning capabilities. The framework's approach reduces reliance on human intervention, enabling LLMs to tackle a wide range of tasks independently. It is scalable and adaptable, making it a valuable tool for various domains, including software development and complex decision-making scenarios. The paper highlights the potential of multi-agent communication in overcoming the limitations of individual models, offering a foundation for autonomous problem-solving. However, the framework has limitations, including the need for diverse training data and the ability to incorporate new information. Future work aims to address these limitations and further enhance the framework's capabilities.This paper introduces a novel multi-agent communication framework to enhance the autonomous problem-solving capabilities of Large Language Models (LLMs). The framework employs multiple LLM agents, each with a distinct persona, engaged in role-playing communication to address diverse problem scenarios. Inspired by the CAMEL model, the framework leverages chain-of-thought prompting to enable collaborative problem-solving. The agents work together to devise solutions for novel problems, with each agent guided by a tailored chain-of-thought prompt. The framework's versatility allows for the incorporation of any persona and chain-of-thought prompt, aligning with the specific problem to be addressed. The framework is available at https://github.com/sumedhrasal/simulation, built on top of CAMEL’s and ChatDev’s framework. In experiments, the framework was tested on arithmetic reasoning tasks using the GSM8K and SVAMP datasets. The multi-agent approach significantly improved accuracy compared to single-agent models. For example, the multi-agent GPT-3 achieved 55% accuracy on GSM8K and 77% on SVAMP, surpassing other models. In commonsense reasoning tasks using the CSQA dataset, the multi-agent approach achieved 83% accuracy, demonstrating its effectiveness in enhancing reasoning capabilities. The framework's approach reduces reliance on human intervention, enabling LLMs to tackle a wide range of tasks independently. It is scalable and adaptable, making it a valuable tool for various domains, including software development and complex decision-making scenarios. The paper highlights the potential of multi-agent communication in overcoming the limitations of individual models, offering a foundation for autonomous problem-solving. However, the framework has limitations, including the need for diverse training data and the ability to incorporate new information. Future work aims to address these limitations and further enhance the framework's capabilities.