Monty is an experimental Python interpreter built in Rust, designed to run LLM-generated code quickly and securely. Unlike traditional container-based sandboxes, Monty executes Python code with microsecond startup times. It can run a limited subset of Python, fully block access to the host environment, and allows developers to control which functions are accessible. The performance is competitive with CPython, varying between five times faster and five times slower. The interpreter supports modern Python type hints and can perform type-checking. It offers features like snapshotting the interpreter state, tracking resource usage, and capturing output. However, Monty has limitations: it doesn't support the standard library (with few exceptions) or external libraries, and it cannot define classes or use match statements yet. The project’s main focus is to enable agents to run code generated by AI models efficiently and safely. Monty is positioned as a solution for accelerating AI capabilities by allowing models to write code instead of executing traditional tool calls. It integrates easily with Python, Rust, and JavaScript, making it versatile for different programming environments. The installation process is straightforward, requiring a single command. Examples demonstrate how to use Monty for simple tasks, including handling asynchronous code and managing external function calls iteratively.