As AI continues to evolve, the role of AI agents in performing human tasks is becoming increasingly prominent. From researching to executing programs, these agents are starting to interact with computers more directly. This shift has been further emphasized by Anthropic’s demonstration of Claude Computer Use, showcasing the potential of AI agents using computers to execute complex tasks.
However, this raises significant challenges, particularly the need for secure and isolated environments for these agents to operate. Enter MarinaBox, an open-source toolkit designed to create isolated desktop or browser sandboxes. It enables AI agents to work on computers in a secure and controlled environment.
In this article, we’ll demonstrate how to use MarinaBox in conjunction with LangGraph, the leading library for building multi-agent AI systems. We’ll provide a detailed walkthrough of setting up MarinaBox as a Computer Tool for LangGraph agents, complete with code samples and implementation details.
MarinaBox library provides Langgraph nodes to start and stop the computer and a Langgraph tool which is used to control the Computer using Claude Computer Use.