This paper introduces StreamLLM, a method that uses Large Language Models (LLMs) to generate streamliners for constraint programming. Streamliners narrow the search space to improve the efficiency of solving complex problems but typically require extensive manual design or exhaustive testing. StreamLLM instead leverages LLMs to propose effective streamliners dynamically, incorporating realtime feedback and empirical tests within the MiniZinc modeling language. Evaluated across six diverse constraint satisfaction problems, StreamLLM demonstrates substantial runtime reductions, up to 99% improvement in some cases. This work highlights the potential of combining symbolic reasoning with machine learning techniques to enhance constraint-solving speed and adaptability.