Neura Market
lightbearco/tupper logo

lightbearco/tupper

Free

面向 AI 代理的开源沙箱，在本机安全运行不受信任的 AI 生成代码、Shell 命令与文件读写。基于 Apple Containers，提供 E2B 风格的 TypeScript SDK，内置 MCP 服务器、CLI 与 HTTP API（macOS 用 Apple Containers，Linux/Windows 经 Firecracker/WSL）。

FreeFree tier
Inputs: codeOutputs: text
Type
Open Source
Visit Website GitHub (40 stars)

About lightbearco/tupper

Tupper is an open-source sandbox for AI agents that allows running untrusted, AI-generated code safely on the user's own machine. It provides an isolated environment to execute code, run shell commands, and read/write files without risking the host system. Backed by Apple Containers on macOS (with Firecracker for Linux and WSL for Windows planned), Tupper offers an E2B-style TypeScript SDK, a Model Context Protocol server, CLI, and HTTP API. It is a self-hostable, platform-agnostic alternative to hosted code-execution sandboxes like E2B, Daytona, and Modal.

Key Features

One SDK, every OS – write agent code once; Tupper selects the right sandbox runtime automatically (Apple Containers on macOS, Firecracker on Linux, WSL on Windows).
E2B-style SDK – Sandbox.create(), commands.run(), files.*, plus reconnect and lifecycle control.
Native macOS sandboxing – Apple Containers via the container CLI, no Docker daemon required.
Runs on Node or Bun – lean core with Zod as only runtime dependency; uses Node built-ins only.
Agent-framework ready – drop-in sandbox backends for deepagents and Mastra, plus MCP server, CLI, and HTTP API.
Self-hostable alternative to E2B, Daytona, and Modal.

Pros & Cons

Pros
  • Open-source and self-hostable, giving full control over data and infrastructure
  • Platform-agnostic with a unified SDK across macOS, Linux, and Windows
  • No Docker daemon required on macOS, reducing overhead
  • Compatible with existing E2B SDK patterns, easing migration
  • Integrates with agent frameworks like Mastra and deepagents
  • Lightweight core, runs on Node or Bun without heavy dependencies
Cons
  • Currently only supports macOS 26+ with Apple Containers; Linux and Windows backends are in development
  • Early development stage – APIs may change before 1.0
  • Requires Apple's container CLI on macOS and Node 18+ or Bun 1.1+
  • Limited to code execution and file operations; no GPU or advanced resource controls yet

Best For

Running untrusted AI-generated code in isolationExecuting shell commands without host riskReading and writing files in a sandboxed environmentBuilding AI agents that need a secure code execution backendTesting code snippets in a disposable container