Show HN:寻找 Beta 测试员:在 AgentSphere 沙盒中运行 AI 生成的代码
1 分•作者: AgentSphere•6 个月前
大家好!
我们正在开放 AgentSphere 的抢先体验,这是一个安全的运行时平台,用于在云沙箱中执行 AI 生成的代码。如果您正在使用 LLM(ChatGPT、Claude 等)进行构建,并且希望它们不仅编写代码,还能实际运行代码——那么 AgentSphere 正是为您量身打造的。
--------------------------------------------
什么是 AgentSphere?
AgentSphere 提供了一种简单的方法,让您可以使用我们的 JavaScript SDK、Python SDK 或 API 在云端启动轻量级、隔离的环境。每个沙箱都像一个容器化的运行时环境——可以将其视为“执行即服务”:无需本地设置,无需配置地狱——只需发送一个命令,它就会安全地运行。
--------------------------------------------
AgentSphere 如何与 MCP 客户端配合使用?
AgentSphere 已集成到 MCP 生态系统中,并与以下产品无缝协作:
- Cursor
- Claude Code
- Raycast
- Gemini CLI
- 以及更多 MCP 客户端...
参考文档:[https://www.agentsphere.run/docs/mcp-server](https://www.agentsphere.run/docs/mcp-server)
--------------------------------------------
演示
以下视频片段演示了 AgentSphere 的实际应用。
- Web 游戏马里奥:[https://www.youtube.com/watch?v=5mtqUzAwXoA](https://www.youtube.com/watch?v=5mtqUzAwXoA)
- 计算器:[https://www.youtube.com/watch?v=ZXTPOddl_Jo](https://www.youtube.com/watch?v=ZXTPOddl_Jo)
--------------------------------------------
加入 Beta 测试
如果您想:
- 构建能够实际执行代码和处理数据的 LLM 代理
- 使用 Cursor、Claude Code、Raycast、Gemini CLI 或类似的 MCP 客户端,并希望输出“直接运行”
- 尝试自主代理工作流程
- 在安全、隔离的环境中测试 AI 生成的代码
请试用 AgentSphere。
想使用 AgentSphere 进行构建吗?立即注册,只需点击几下即可获取您的 API 密钥!
[https://www.agentsphere.run/](https://www.agentsphere.run/)
我们正在寻找开发人员、构建者和 AI 工具制造商来测试它,并告诉我们哪些是好的,哪些是不好的。我们非常期待您的反馈。
有问题吗?请在下方留言。谢谢!
查看原文
Hey everyone<p>We’re opening up early access to AgentSphere, a secure runtime platform for executing AI-generated code inside cloud sandboxes. If you’re building with LLMs (ChatGPT, Claude, etc.), and want them to not just write code but actually run it — this is for you.<p>--------------------------------------------<p>What is AgentSphere?<p>AgentSphere provides an easy way for you to launch lightweight, isolated environments in the cloud using our JavaScript SDK, Python SDK or APIs. Each sandbox behaves like a containerized runtime — Think of it as Execution-as-a-Service: no local setup, no config hell — just send a command and it runs securely.<p>--------------------------------------------<p>How AgentSphere Works with MCP Clients?<p>AgentSphere is integrated into the MCP ecosystem and works seamlessly with:<p>- Cursor<p>- Claude Code<p>- Raycast<p>- Gemini CLI<p>- and more MCP clients...<p>Reference Docs: <a href="https://www.agentsphere.run/docs/mcp-server" rel="nofollow">https://www.agentsphere.run/docs/mcp-server</a><p>--------------------------------------------<p>Demo<p>The following video clip demonstrates what AgentSphere can help in action.<p>- Web Game Mario: <a href="https://www.youtube.com/watch?v=5mtqUzAwXoA" rel="nofollow">https://www.youtube.com/watch?v=5mtqUzAwXoA</a><p>- Calculator: <a href="https://www.youtube.com/watch?v=ZXTPOddl_Jo" rel="nofollow">https://www.youtube.com/watch?v=ZXTPOddl_Jo</a><p>--------------------------------------------<p>Join the Beta<p>If you want to:<p>- Build LLM agents that actually execute code and handle data<p>- Use Cursor, Claude Code, Raycast, Gemini CLI, or similar MCP clients and want the output to “just run”<p>- Experiment with autonomous agent workflows<p>- Test AI-generated code in a safe, isolated environment<p>Just give AgentSphere a try.<p>Want to build with AgentSphere? Register now and grab your API key in just a few clicks!<p><a href="https://www.agentsphere.run/" rel="nofollow">https://www.agentsphere.run/</a><p>We're looking for developers, builders, and AI toolmakers to test it out, and tell us what is great and what doesn’t. We'd love your feedback.<p>Got questions? Drop them below. Thank you!