An open agent framework where the AI agent gets its own Dockerized Linux desktop, browser, and document tools to carry out real work.
agent-infra/sandbox
An all-in-one Docker sandbox giving AI agents a shared browser, shell, file system, VSCode Server, Jupyter and MCP services.
- Stars
- 6,185
- Forks
- 561
- Open issues
- 73
- Last commit
- last month
What is it?
AIO Sandbox packages a browser, terminal, file system, VSCode Server, Jupyter and MCP services into a single Docker container. Agents and developers can run shell commands, read and write files, control a browser, and execute Python or Node.js code inside the same environment, so files downloaded in the browser are immediately available to shell and file operations.
Access is through HTTP APIs with official Python, TypeScript and Go SDKs. The container exposes a VNC desktop, Chrome DevTools Protocol for browser control, a VSCode web IDE, and pre-configured MCP servers for browser, file, shell and document conversion. An optional API key protects the API, JupyterLab and VNC endpoints.
The project provides Docker Compose and Kubernetes manifests, and it can be combined with agent frameworks such as browser-use and LangChain, or with OpenAI-compatible model APIs for tool calling.
Key features
- 01Browser control through VNC, Chrome DevTools Protocol and MCP tools
- 02Shell terminal and file read/write operations sharing one file system
- 03Embedded VSCode Server and JupyterLab for coding in the browser
- 04Pre-configured MCP servers for browser, file, shell and document conversion
- 05Python and Node.js code execution inside the sandbox
- 06Optional API key authentication for API and service endpoints
Use cases
- Giving an AI agent a safe place to run shell commands and code
- Automating browser tasks that need downloaded files processed afterward
- Hosting a disposable cloud IDE for developers in a container
- Connecting agent frameworks such as browser-use or LangChain to a ready-made tool environment
Quick start
- 1
Run the sandbox container with an API key (recommended)
docker run --security-opt seccomp=unconfined --rm -it -e SANDBOX_API_KEY=your-secret-key -p 127.0.0.1:8080:8080 ghcr.io/agent-infra/sandbox:latest - 2
Open the documentation at localhost:8080
- 3
Install the Python SDK (optional)
pip install agent-sandbox
Things to know
- Requires Docker, and the container runs with seccomp unconfined.
- Setup relies on pulling a large all-in-one container image.
Topics
FAQ
- What is sandbox?
- An all-in-one Docker sandbox giving AI agents a shared browser, shell, file system, VSCode Server, Jupyter and MCP services. AIO Sandbox packages a browser, terminal, file system, VSCode Server, Jupyter and MCP services into a single Docker container. Agents and developers can run shell commands, read and write files, control a browser, and execute Python or Node.js code inside the same environment, so files downloaded in the browser are immediately available to shell and file operations.
- Is sandbox free to use? What license does it have?
- sandbox is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is sandbox still maintained?
- Updated within 6 months. The last commit to sandbox was on Sep 14, 2026.
- What can I use sandbox for?
- Giving an AI agent a safe place to run shell commands and code Automating browser tasks that need downloaded files processed afterward Hosting a disposable cloud IDE for developers in a container Connecting agent frameworks such as browser-use or LangChain to a ready-made tool environment
- How do I install sandbox?
- Run the sandbox container with an API key (recommended): docker run --security-opt seccomp=unconfined --rm -it -e SANDBOX_API_KEY=your-secret-key -p 127.0.0.1:8080:8080 ghcr.io/agent-infra/sandbox:latest Open the documentation at localhost:8080 Install the Python SDK (optional): pip install agent-sandbox
- Where can I get sandbox?
- The source code and installation instructions are on GitHub: https://github.com/agent-infra/sandbox
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A JavaScript library that embeds a GUI agent in any web page, letting users control the interface with natural language.
Open-source infrastructure for running AI-generated code in isolated cloud sandboxes, with JavaScript and Python SDKs.
Self-hosted AI coworker platform where each agent gets its own browser, files and tools, with every action gated by policy and logged.
An Electron-based, local-first browser where AI agents operate inside your real logged-in browser session via MCP or HTTP API.
A headless browser written in Zig from scratch for AI agents and automation, with CDP and WebDriver BiDi servers.