在 Warp 中充分利用一次请求

Jeff Lloyd

Jeff Lloyd

August 22, 2025

2025年8月22日

Whether you’re a long-time Warp fan, or you’re just getting started with prompt-driven development, you may be asking: what exactly is a request in Warp? Why do some of my AI interactions cost one request, while others cost a dozen? How do I know what plan (Pro/Turbo/Lightspeed) is right for me? In this blog post, I’ll pull back the curtain on how requests work, so that you can maximize whatever plan fits you the best!

无论你是 Warp 的老粉，还是刚刚接触提示驱动开发，你可能都在问：在 Warp 里，到底什么算一次请求？为什么有的 AI 交互只花 1 个请求，有的却花十几个？我该如何判断 Pro/Turbo/Lightspeed 哪个套餐最适合我？在这篇博客里，我将揭开请求机制的面纱，帮你把最适合你的套餐用到极致！

By the way, my name’s Jeff – I’m one of the Tech Leads here at Warp. If you love thinking about pricing as much as I do, or if you want to build the future of agentic development, we’re hiring!

顺便说一句，我叫 Jeff——我是 Warp 的技术负责人之一。如果你像我一样热爱思考定价问题，或者想一起打造 agentic development 的未来，我们正在招聘！

What is a request?

什么是 request？

A request is the smallest unit of interaction with one of Warp’s AI agents. For example, a simple prompt like “Create a new branch called jeff/my-cool-feature” might use one request. In fact, in my testing, it totally does! You can check the number of requests used since your last user input to the agent via the informational tooltip at the end of the conversation.

请求是与 Warp 的 AI agent 交互的最小单位。例如，一句简单的提示“创建一个名为 jeff/my-cool-feature 的新分支”就可能消耗一次请求。事实上，在我的测试中确实如此！你可以通过对话末尾的信息提示查看自上次用户输入以来 agent 已使用的请求次数。

It’s important to note that you never use requests for running shell commands on your own – only when talking to agents!

请注意，你永远不会因为自己在终端运行 shell 命令而消耗请求——只有在与 agent 对话时才会！

What goes into a request?

什么会算入一个 request？

In general, there are a few critical inputs to how many requests a query will consume:

一般来说，决定一个查询会消耗多少 requests 有几个关键输入：

1. What model are you using? More premium models = more requests

   你使用的是什么模型？越高级的模型 = 越多的 requests

2. How much context are you attaching to your request? More context = more requests

   你在请求中附上了多少上下文？上下文越多 = 请求越多

3. How deeply does the agent need to scan your codebase, filesys...