请问HN：你们是如何在网页应用中实现多代理工作流的？
我想尝试多代理工作流程——有时候我能找到一个良好的流程，让一个代理在我知道需要做什么但需要大量代码的工作中相对独立地运作。它运行一段时间后，我需要在浏览器中验证它是否真的有效，然后从那里进行迭代和调试。或者我有三种不同的方法想要尝试，我可以让人工智能只执行其中一种，看看它是否与前端良好配合，如果不行可以迅速回滚。
我希望能够在一个代理运行时同时处理另一个代理，否则我就只能坐在那里等待。
问题是我们的环境并不允许我们的应用程序同时运行多个实例——我们的前端很重，占用大量内存，因此即使我们找到了运行多个后端的方法，内存也会成为一个问题。
似乎很多多代理工作流程使用命令行工具——这很有道理。有没有人成功地在网页上实现过？也许还有一些浏览器自动化的方式？
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?