Agent-Infra Releases AIO Toolbox: All-in-One Runtime for AI Agents with Browser, Shell, Shared Filesystem, and MCP
In the development of autonomous agents, the technical bottleneck is changing from the model to the implementation area. Although Large-scale Language Models (LLMs) can create code and multi-step programs, providing a single functional environment for that code to run is still a major infrastructure challenge.
The Agent-Infra sandbox, an open source project, addresses this by providing an 'All-in-One' (AIO) operating layer. Unlike traditional containers, which often require manual configuration to integrate tools, AIO Sandbox combines the browser, shell, and file system into a single environment designed for AI agents.
All-in-One Architecture
The main architectural obstacle to agent development is the fragmentation of tools. In general, an agent may need a browser to fetch the data, a Python interpreter to analyze it, and a file system to store the results. Managing these as separate services introduces latency and synchronization difficulties.
Agent-Infra combines these requirements into one container. The sandbox includes:
- Computer Interaction: A Chromium browser managed by the Chrome DevTools Protocol (CDP), with written support for Playwright.
- Code Execution: Pre-configured runtimes for Python and Node.js.
- Common Tools: A bash terminal and file system accessible to all modules.
- Development Interfaces: Integrated VSCode Server and Jupyter Notebook environments for monitoring and debugging.
Integrated File System
An important technical feature of Sandbox is Integrated File System. In a typical agent workflow, the agent might download a file using a browser-based tool. In a batch setup, that file must be moved programmatically to a different location for processing.
I AIO sandbox uses a shared storage layer. This means that a file downloaded through the Chromium browser is immediately visible in the Python interpreter and the Bash shell. This shared environment allows transitions between tasks—such as an agent downloading a CSV from a web portal and quickly running a data cleaning script in Python—without managing external data.
Model Context Protocol (MCP) Integration
Sandbox includes native support for Model Context Protocol (MCP)an open standard that enables communication between AI models and tools. By providing pre-configured MCP servers, Agent-Infra allows developers to expose sandbox capabilities to LLMs through a standardized protocol.
Available MCP servers include:
- Browser: Web browsing and data extraction.
- File: Hardware in the integrated file system.
- Shell: By executing system commands.
- Markitdown: To convert document formats to Markdown to prepare them for LLM use.
Classification and Distribution
Sandbox is designed for 'enterprise-level Docker deployments,' with a focus on scalability and scalability. While it provides a persistent environment for complex tasks—such as maintaining a terminal session with multiple turns—it's designed to be simple enough for high-density use.
Distribution and control:
- Infrastructure: This project includes examples of Kubernetes (K8s) deployments, allowing teams to use native K8s features such as resource limits (CPU and memory) to manage sandbox trajectories.
- Container Classification: By running the agent's operations inside a dedicated container, the sandbox provides a layer of isolation between the code generated by the agent and the host system.
- Access: The environment is controlled by using i API and SDKwhich allows developers to systematically execute commands, execute code, and manage environment.
Technology Comparison: Traditional Docker vs. AIO sandbox
| A feature | The Traditional Docker Method | AIO Sandbox Method (Agent-Infra) |
| Buildings | Usually multiple containers (one browser, one code, one shell). | Combined Container: Browser, Shell, Python, and IDE (VSCode/Jupyter) concurrently. |
| Data Management | It requires volume mounting or “piping” API to move files between containers. | Integrated File System: Files are shared natively. Browser downloads appear faster in shell/Python. |
| Agent integration | It requires custom “glue code” to map LLM actions to the container command. | Native MCP support: Preconfigured Content Model Protocol Servers for a standard proxy. |
| User interface | CLI based; Web-UIs like VSCode or VNC require significant manual setup. | Built-in visuals: Integrated VNC (for Chromium), VSCode server, and Jupyter are ready out of the box. |
| Resource Management | Managed with standard Docker/K8s cgroups and resource limitations. |
It depends on the bottom orchestrator (K8s/Docker) for resources and limitations. |
| Communication | Standard Docker bridge/host networking; manual setup of the proxy is required. | CDP-Based Browser Control: Exclusive cross-browser interaction with the Chrome DevTools Protocol. |
| Persistence | Containers are usually long-lasting or manually reset; State management is a practice. | Key Session Support: It supports continuous terminal and work environment status throughout the work life cycle. |
Scaling the Agent Stack
While the core Sandbox is open source (Apache-2.0), the platform is positioned as a scalable solution for teams building complex agent workflows. By reducing the overhead of 'Agent Ops'—the work required to maintain workstations and handle dependency conflicts—the sandbox allows developers to focus on the agent's logic instead of on the downtime.
As AI agents evolve from simple chatbots to operators that can interact with the web and local files, the workspace becomes an important part of the stack. The Infra agent team is positioning the AIO Sandbox as a standardized, lightweight solution for this transition.
Check it out Repo here. Also, feel free to follow us Twitter and don't forget to join our 120k+ ML SubReddit and Subscribe to Our newspaper. Wait! are you on telegram? now you can join us on telegram too.
Michal Sutter is a data science expert with a Master of Science in Data Science from the University of Padova. With a strong foundation in statistical analysis, machine learning, and data engineering, Michal excels at turning complex data sets into actionable insights.
