问HN：人工智能代理能否从零开始构建全栈应用程序？
在过去的几个月里，我们一直在思考一个挥之不去的问题：<p>如果人工智能能够真正构建应用程序——不仅仅是生成代码片段，而是从头开始搭建一个完整的工作产品，会怎样？<p>因此，我们开发了saas-builder，这是一个开源的命令行工具，允许你描述你的应用想法，并使用Next.js和Supabase生成一个全栈的基础架构。<p>你只需运行：saas-builder init<p>…然后五个AI代理就会接管：
> UI代理 → 构建响应式前端
> 认证代理 → 连接Supabase认证
> 数据库代理 → 创建模式、迁移和类型
> 终端代理 → 安装依赖和文件夹
> 浏览器代理 → 在本地启动应用<p>你得到的是一个干净、类型安全、准备部署的应用。现在仍处于早期阶段——更多的是一个副项目。但它已经为我们节省了数小时的样板设置时间。<p>更重要的是，它让我们窥见了代理开发可能实际感受如何。<p>我们将其以MIT许可证发布，可以通过pip安装，真诚地希望听到你的想法：<p>GitHub: https://github.com/jatingarg619/saas-builder<p>感谢你的阅读。我们随时欢迎提问、讨论想法或听取你的反馈。
查看原文
Over the past few months, we’ve been tinkering with a question that wouldn’t leave us alone:<p>What if AI could actually build apps — not just generate code snippets, but scaffold an entire working product from scratch?<p>So we built saas-builder, an open-source CLI that lets you describe your app idea and generates a full-stack scaffold using Next.js and Supabase.<p>You run: saas-builder init<p>…and five AI agents take over:
> UI Agent → builds a responsive frontend
> Auth Agent → wires up Supabase Auth
> DB Agent → creates schema, migrations, and types
> Terminal Agent → installs dependencies & folders
> Browser Agent → launches the app locally<p>What you get is a clean, typed, ready-to-deploy app. It’s still early days — very much a side project. But it’s already saved us hours of boilerplate setup.<p>More importantly, it gave us a glimpse of what agentic development might actually feel like.<p>We’re releasing it under MIT, installable via pip, and would genuinely love your thoughts:<p>GitHub: https://github.com/jatingarg619/saas-builder<p>Appreciate you reading. We’re around to answer questions, chat about ideas or hear whatever feedback you’ve got.