Agent-Sandbox Overview¶

Agent-Sandbox is an open-source, Kubernetes-native runtime platform for AI agents. It provides isolated, stateful, multi-tenant sandboxes for code execution, browser/computer tasks, and shell workflows.
The project is designed to be compatible with E2B protocol and SDK workflows while also exposing native REST API, MCP, and a web UI.
Features¶
-
Set up in 5 minutes
Install
Agent-Sandobxwithkubectl applyand get up and running in minutes -
Ai friendly
Supports Skills, Cli and MCP to empower your agents with sandbox capabilities
-
Easy to use
Designed with a simple REST API, UI and minimizing Kubernetes's objects to deploy for easy to use and maintain.
-
Open and flexible
Supports E2B templates and any container image, and can be extended with custom templates.
What it provides¶
- E2B protocol compatibility for sandbox lifecycle APIs and routing.
- Sandbox lifecycle management: create, list, connect, delete.
- Multi-tenant access control with system and regular users.
- Template and pool management for fast sandbox allocation and warmup.
- Observability with sandbox events, metrics, and logs.
- Interactive operations: terminal, file upload/download, and sandbox routing.
- MCP server integration for agent-native automation.
- Web UI for sandbox/template/pool operations and runtime inspection.
Quick start ¶
See Quick Start for a step-by-step guide covering deployment, creating a sandbox via E2B SDK, and using the REST API.