在黑客新闻上没有人能回答这个问题。
大家好。我有一个简单的问题，但我很难找到答案，这里没有人能给出合理的解答。请大家试试看。乍一看似乎很明显，但仔细想想你会发现自己其实并不知道。
我们的初创公司想要构建并使用一个智能代理。我想为这些代理赋予技能、访问我们的API等。我不想重新发明轮子，也不想重新构建代理循环。
理想情况下，我们希望拥有一个托管版本的Claude Code，具备公司特定的技能、API访问层等。就像Claude Code/Cursor可以访问我们的代码库和技能一样，这正是我们想要的，然后开始以有限的方式向我们的内部团队开放，随后再向客户开放。例如，向我们现有的产品添加特定功能，比如“用AI总结这个”，但这个API调用不是发送到模型提供商，而是发送到“我们的”代理。
这样说有道理吗？
请问有人能给我指个方向，告诉我最佳/最简单的解决方案是什么吗？我觉得这应该很简单明了，但我在寻找明确的前进路径时遇到了困难。谢谢大家！- Mike
查看原文
Hey everyone. I have a simple question and have had a very tough time finding the answer, and no one here can give a sensible answer. Just give it a shot. At first glance it seems obvious, but then you'll realize you have no idea.<p>Our startup wants to build and use an agent. I want to give the agents skills, access to our APIs, etc. I do not want to reinvent any wheels, I do not want to reinvent an agent loop.<p>Ideally, we'd effectively have a hosted version of Claude Code with company specific skills, an API access layer, etc. In the way that Claude Code / Cursor has access to our codebase + skills, thats effectively what we want and then start exposing it in a limited fashion to our internal team and then later to our customers. And, for example, add specific features to our existing product such as "summarize this with AI" but instead of that api call going to a model provider, it would go to 'our' agent.<p>Does this make sense?<p>Could someone please point me in the right direction on what the best / simplest solution would be? I feel like it should be easy and obvious but I'm struggling to find a clear path forward. Thank you all! - Mike