AIO Sandbox: An all-in-one execution environment for AI Agents (github.com)

Agent Infra released AIO Sandbox, a single Docker-deployable execution environment that bundles a VNC/Chromium browser (CDP), shell, shared filesystem, VSCode Server, Jupyter, and MCP (Model Context Protocol) services into one container. You can launch it in ~30 seconds (docker run -p 8080:8080 ghcr.io/agent-infra/sandbox:latest) and interact via UI endpoints (VNC, code-server, /v1/docs) or programmatically through REST/SDKs (Python, TypeScript/JS, Go). The platform exposes APIs like /v1/shell/exec, /v1/file/read, /v1/browser/screenshot and includes preconfigured MCP servers for browser, file, shell and markitdown operations. This matters for AI/ML engineers building autonomous agents and tool-augmented LLM workflows because it removes friction between disparate sandboxes: browser downloads, code execution, and file artifacts share a unified filesystem and runtime, enabling seamless end-to-end automation (e.g., CDP-driven scraping -> Jupyter processing -> shell/file ops). It supports sandboxed Python/Node execution, zero-config tooling, port forwarding and k8s deployment manifests, making rapid prototyping, reproducible testing and secure remote execution easier. With Apache 2.0 licensing and SDKs for common stacks, AIO Sandbox aims to become a convenient local/in-cluster playground for agent development, tool integration and safer evaluation of autonomous behaviors.