agent-infra/sandbox

All-in-One Sandbox for AI Agents that combines Browser, Shell, File, MCP and VSCode Server in a single Docker container.

What it solves

Traditional AI agent sandboxes are often single-purpose, providing only a browser or a shell, which makes it difficult for agents to share files and coordinate functions across different tools. AIO Sandbox provides a unified, secure execution environment where browser, shell, file operations, and development tools all share a single filesystem.

How it works

It packages a browser, shell, VSCode Server, and Jupyter Notebook into a single Docker container. This environment is accessible via APIs or SDKs (Python, TypeScript, Go), allowing AI agents to programmatically control the environment. It also integrates the Model Context Protocol (MCP) to provide standardized tools for web automation, file system operations, and command execution.

Who it’s for

Developers building AI agents that require a secure, multi-tool environment to perform complex tasks like web scraping, code execution, and file manipulation without needing to configure multiple separate services.

Highlights

  • Unified Filesystem: Files downloaded via the browser are immediately available to the shell and Jupyter notebook.
  • Multi-Interface Access: Includes VNC for visual browser interaction, VSCode Server for IDE access, and a WebSocket-based terminal.
  • MCP Integration: Pre-configured MCP servers for browser, file, shell, and document processing (Markitdown).
  • Agent-Ready APIs: Direct integration support for frameworks like LangChain, OpenAI Assistants, and Browser-use.

Related

  • Project
    madarco/agentbox

    A tool for running multiple AI agents in parallel within sandboxed VMs or Docker containers, either locally or in the cloud, with persistent IDE and shell access.

  • Project
    opensandbox-group/OpenSandbox

    A general-purpose sandbox platform for AI applications that provides secure, isolated environments for code execution, agent evaluation, and RL training.

  • Project
    Simpleyyt/ai-manus

    A general-purpose AI agent system that executes tasks in a secure Docker sandbox with capabilities for web browsing, code execution, and file management.

  • Project
    vercel-labs/bash-tool

    A generic bash tool for AI agents that enables them to execute shell commands and manage files within a sandbox environment.

  • Dispatch
    Securing AI Agents with Isolated Docker Sandboxes

    An exploration of agent-sandbox, a tool designed to run AI coding agents within isolated Docker containers to prevent unauthorized system access and ensure execution safety.