Ask HN：你们是如何用 Web 应用实现多智能体工作流的？
1 分•作者: ativzzz•19 天前
我想尝试一下多智能体工作流程——有时候我能让一个智能体的工作流程非常顺畅，它可以独立完成我知道需要完成但需要大量代码的工作。它会运行一段时间，然后我需要在浏览器中验证它是否真的有效，并从那里迭代/调试。或者我有 3 种不同的方法想尝试，我可以让 AI 只做一种，看看它是否与前端配合良好，如果不行就快速回滚。
我希望能够在它运行时同时处理另一个智能体，否则我只能坐在那里等待。
问题是我们的环境实际上不允许同时运行我们应用程序的多个实例——我们的前端很重，需要大量的内存，所以即使我们想办法运行多个后端，内存也会成为一个问题。
似乎很多多智能体工作流程都使用 CLI 工具——这很有道理。有人在 Web 端取得成功了吗？也许也包括一些浏览器自动化？
查看原文
I'd like to try out multi agent workflows - sometimes I get a good flow where I can get one agent to be pretty independent for work where I know what needs to be done but needs a lot of code. It spins for a while, then I need to verify in the browser that it actually works and iterate/debug from there. Or I have 3 different approaches I want to try, and I can have the AI just do 1 and see if it works well with the front end and quickly roll back if not.<p>I'd like to be able to work on another agent while it's spinning, otherwise I just sit there and wait.<p>The issue is our env doesn't really allow multiple instances of our app to run simultaneously - our front end is heavy and takes a ton of memory so even if we figure out how to run multiple backends, RAM would be an issue.<p>It seems a lot of multi agent workflows use CLI tools - which makes sense. Anyone find success with web? Maybe some browser automation too?