This repository implements LATM (LLMs As Tool Makers), a closed-loop framework where large language models generate their own reusable tools for task solving. The system has two phases: tool-making, where an LLM crafts Python utility functions for given tasks, and tool-using, where an LLM applies those tools for problem-solving. The division between maker and user roles enables cost-effective deployment by using a powerful model as maker and a lightweight model as user. The approach is validated across Big-Bench reasoning tasks.