KNOWAGENT is a novel approach designed to enhance the planning capabilities of large language models (LLMs) by incorporating explicit action knowledge. The method introduces an action knowledge base and a knowledgeable self-learning strategy to constrain the action path during planning, enabling more reasonable trajectory synthesis and improving the planning performance of language agents. The approach involves creating an extensive action knowledge base, converting action knowledge into text, and using a knowledgeable self-learning phase to iteratively improve the model's understanding and application of action knowledge. Experimental results on HotpotQA and ALFWorld demonstrate that KNOWAGENT can achieve comparable or superior performance to existing baselines. The method effectively mitigates planning hallucinations by reducing the frequency of erroneous actions and ensuring that action sequences align with real-world logic. The approach is effective across various models and tasks, showing the benefits of integrating external action knowledge to streamline planning processes and improve performance. The study also highlights the limitations of the approach, including task expandability, multi-agent systems, and automated design of action knowledge bases. The research contributes to the field of knowledge-augmented planning for LLM-based agents, offering a framework to reduce planning errors and enhance the performance of language agents in complex scenarios.KNOWAGENT is a novel approach designed to enhance the planning capabilities of large language models (LLMs) by incorporating explicit action knowledge. The method introduces an action knowledge base and a knowledgeable self-learning strategy to constrain the action path during planning, enabling more reasonable trajectory synthesis and improving the planning performance of language agents. The approach involves creating an extensive action knowledge base, converting action knowledge into text, and using a knowledgeable self-learning phase to iteratively improve the model's understanding and application of action knowledge. Experimental results on HotpotQA and ALFWorld demonstrate that KNOWAGENT can achieve comparable or superior performance to existing baselines. The method effectively mitigates planning hallucinations by reducing the frequency of erroneous actions and ensuring that action sequences align with real-world logic. The approach is effective across various models and tasks, showing the benefits of integrating external action knowledge to streamline planning processes and improve performance. The study also highlights the limitations of the approach, including task expandability, multi-agent systems, and automated design of action knowledge bases. The research contributes to the field of knowledge-augmented planning for LLM-based agents, offering a framework to reduce planning errors and enhance the performance of language agents in complex scenarios.