dsh-notebook

In data science and machine learning, Jupyter Notebooks are widely used, but traditional notebooks are static and not suited for agent-driven interactions. When AI agents need to dynamically manipulate computational environments, they often lack persistent state and controllable interfaces, which limits efficiency. The dsh-notebook plugin addresses this by providing a stateful, agent-controllable Jupyter workspace. It features a persistent kernel that keeps variables alive across cells and offers a suite of agent tools (such as reading, editing, and running cells) for seamless interaction. Core capabilities include runtime introspection, safe edit loops, a VS Code-like UI, session isolation, and kernel management, enabling agents to efficiently drive computational sessions. This plugin is ideal for developers and researchers using DeepSeek Harness, particularly those in data science and machine learning workflows who require AI agents to dynamically operate notebooks, manage state, and execute complex tasks, transforming the notebook from a static document into an intelligent, actionable workspace.