Endless Terminals presents a system designed to autonomously generate tasks for training terminal agents using reinforcement learning. This pipeline eliminates the need for human annotation, streamlining the process of creating diverse terminal-use scenarios. It requires Python 3.12 or higher and the uv library, along with several scripts for setup and execution. To get started, users must install Apptainer and its dependencies, download a base container, and launch a local vLLM server. The core of the task generation involves executing a Python script that can create up to 100 tasks in parallel, outputting multiple files for each task, including JSON and Python scripts for testing. The subsequent step involves generating solutions based on these tasks, which are then used to prepare a dataset for training. Training the agents involves setting up Ray and executing a training script with different configuration files tailored for various models, such as Llama and Qwen. The article details how to install SkyRL and run evaluations using the Harbor tool, which allows for parallel processing of model evaluations. The entire setup is aimed at enhancing the efficiency of training terminal agents, making it easier to scale and adapt to different environments without extensive manual intervention.