背景介绍

Built to make you extraordinarily productive, Cursor is the best way to code with AI.

Cursor 是 VS Code 的一个分支。这使我们能够专注于与 AI 进行编码的最佳方式，同时提供熟悉的文本编辑体验。

cursor0

Cursor 背后的公司于 2023 年在旧金山成立，主要开发利用 LLM 从基层建立的 IDE。创始团队目前 2 位已获得 OpenAI 的投资：

Aman Sanger，2022 年毕业于麻省理工学院数学与计算机科学专业，Abelian AI 联合创始人
Michael Truell，2022 年毕业于麻省理工学院数学与计算机科学专业。

可以通过作者 Aman Sanger 在Twitter 上的一个视频来评测它的功能，这个视频一共执行了五条指令：

Build the `SearchResult component showing file icons, names, and paths
Connect this component to redux
How do I add a keyboard shortcut in electron?
Where in the code are shortcuts and redux reducers to open file search
Make cmd+p with label File Search open file search

常用功能

Cursor Tab (选项卡)

Tab is a specialized Cursor model for autocompletion. The more you use it, the better it becomes as you inject intent by accepting Tab or rejecting Escape suggestions. With Tab, you can:

Modify multiple lines at once
Add import statements when missing
Jump within and across files for coordinated edits
Get suggestions based on recent changes, linter errors and accepted edits

Cursor Tab 是本地的自动补全功能。它是一个更强大的 Copilot，能够建议整个差异，并具有特别好的记忆。由自定义模型驱动，Cursor Tab 可以：

在光标周围建议编辑，而不仅仅是插入额外的代码。
同时修改多行。
根据您最近的更改和 linter 错误提出建议。

免费用户可以免费获得 2000 个建议。专业版和商业版计划可以获得无限建议。

用户界面

当 Cursor 仅添加额外文本时，补全将以灰色文本显示。如果建议修改现有代码，它将作为差异弹出窗口出现在当前行的右侧。

cursor1

cursor2

可以通过按 Tab 接受建议，或通过按 Esc 拒绝建议。要逐字部分接受建议，请按 Ctrl/⌘ →。要拒绝建议，只需继续输入，或使用 Escape 取消/隐藏建议。

每次按键或光标移动时，Cursor 将尝试根据您最近的更改提出建议。然而，Cursor 并不总是会显示建议；有时模型预测没有需要更改的内容。

Cursor 可以对当前行上方一行到下方两行进行更改。

切换

要打开或关闭此功能，请将鼠标悬停在应用程序右下角状态栏上的“Cursor Tab”图标上。

在写评论时 Tab 会妨碍我，我该怎么办？

可以通过进入 Cursor Settings > Tab Completion 并取消选中“在评论中触发”来禁用评论中的 Cursor Tab。

Codebase Indexing (代码库索引)

为了更好、更准确地使用 @codebase 或 Ctrl/⌘ Enter 获取代码库答案，用户可以索引用户的代码库。在后台，Cursor 为用户代码库中的每个文件计算嵌入，并将使用这些嵌入来提高代码库答案的准确性。

用户的代码库索引将自动与用户最新的代码库更改同步。

用户代码库索引的状态在 Cursor 设置 > 功能 > 代码库索引 下。

cursor5

高级设置

默认情况下，Cursor 将索引用户代码库中的所有文件。

用户还可以展开显示设置部分以访问更多高级选项。在这里，用户可以决定是否要为新仓库启用自动索引，并配置 Cursor 在仓库索引期间将忽略的文件，除了用户的 .gitignore 设置。

如果用户的项目中有任何大型内容文件，AI 确实不需要读取，忽略这些文件可能会提高答案的准确性。

基本用法

在 Cursor 的 AI 输入框中，例如在 Cmd K、聊天或终端 Cmd K，用户可以通过输入 @ 符号来使用 @ 符号。弹出菜单将出现，列出建议，并会根据用户的输入自动过滤，仅显示最相关的建议。

用户可以使用上下箭头键在建议列表中导航。用户可以按 Enter 选择一个建议。如果建议是一个类别，例如文件，建议将被过滤，仅显示该类别中最相关的项目。

cursor6

用户可以使用上下箭头键在选定的 Cmd K @ 符号列表中导航，按 Enter 展开/折叠选定的上下文项。对于文件引用，用户可以使用 Ctrl/⌘ M 切换文件读取策略。有关文件读取策略的更多信息，请点击这里。

Rules

Rules provide system-level instructions to Agent and Inline Edit. They are persistent context, preferences, or workflows for your projects.

Cursor supports four types of rules:

Project Rules. Stored in .cursor/rules, version-controlled and scoped to your codebase.
User Rules. Global to your Cursor environment. Defined in settings and always applied.
Team Rules. Team-wide rules managed from the dashboard. Available on Team and Enterprise plans.
AGENTS.md. Agent instructions in markdown format. Simple alternative to .cursor/rules.

How rules work?

Large language models don’t retain memory between completions. Rules provide persistent, reusable context at the prompt level.

When applied, rule contents are included at the start of the model context. This gives the AI consistent guidance for generating code, interpreting edits, or helping with workflows.

Best practices

Good rules are focused, actionable, and scoped.

Keep rules under 500 lines
Split large rules into multiple, composable rules
Provide concrete examples or referenced files
Avoid vague guidance. Write rules like clear internal docs
Reuse rules when repeating prompts in chat

配置规则

Cursor 的 .cursorrules 功能为 AI 助手提供了一个定制化的 Prompt。

通过在项目的根目录放置 .cursorrules 文件，我们可以在这个文件里提供更多用与 Cursor 编辑的上下问，比如可以：

指定项目的技术栈
设定开发规范
引导 AI 的问题解决思路
创建自定义指令

这样的预设能显著提高 Curosr 生成代码的准确性和相关性，使其更好地符合项目需求。

参考其他人的规则

这些仓库都维护了各种开发语言的规则。可以选择适合自己的直接使用，或者参考修改一份。

自动生成规则

可以在这个网站描述自己的需求然后生成规则 https://cursorrules.agnt.one/

More: https://docs.cursor.com/context/rules-for-ai

Cursor Agent

Agent is Cursor’s assistant that can complete complex coding tasks independently, run terminal commands, and edit code. Access in sidepane with Cmd+I.

Learn more about how agents work and help your build faster.

Browser (浏览器)

Agent can control a web browser to test applications, audit accessibility, convert designs into code, and more. With full access to console logs and network traffic, Agent can debug issues and automate comprehensive testing workflows.

Tools (工具)

A list of all tools available to modes within the Agent, which you can enable or disable when building your own custom modes.

To understand how tool calling works under the hood, see our tool calling fundamentals.

Tips

安装

Cursor 是一款集成了 AI 技术的代码编辑器，支持多种操作系统，包括 Windows、macOS 和 Linux。

注册与登录

首次使用 Cursor 时，需要注册一个新账号或使用已有账号进行登录。注册过程简单，只需填写必要的个人信息。登录后，可以创建新项目或打开已有项目。

使用功能

Cursor 提供了强大的代码生成和优化功能。用户可以通过输入需求或描述来生成代码，并与代码进行对话以获取反馈和建议。

导入扩展、主题、设置和快捷键

可以通过一键将已有的 VS Code 配置导入到 Cursor。导航到 Cursor Settings > General > Account。

为什么不使用扩展？

作为一个独立应用程序，Cursor 对编辑器的 UI 有更多控制，从而实现更大的 AI 集成。我们的一些功能，如 Cursor Tab 和 CMD-K，作为现有编码环境的插件是不可行的。

Cursor 特定的设置面板

可以通过点击右上角的齿轮按钮，按 Ctrl/⌘ + Shift + J，或使用 Ctrl/⌘ + Shift + P 并输入 Cursor Settings 来打开 Cursor 特定的设置面板。

为什么 Cursor 中的活动栏是水平的？

活动栏默认是水平的，以节省聊天空间。如果更喜欢正常的垂直活动栏，可以前往 VS Code 设置，将 workbench.activityBar.orientation 设置为 vertical，然后重启 Cursor。

有关定价信息

请访问 Cursor Pricing。Cursor 提供多个订阅层级以满足用户的需求。

爱好者
- 14 天专业试用（250 次快速 premium 模型使用）
- 50 次慢速 premium 模型使用
- 2000 次完成使用
专业版
- 每月 500 次快速 premium 模型使用
- 无限次慢速 premium 模型使用
- 无限次完成完成
- 每月 10 次 o1+mini 使用
企业版
- 使用信息与专业版相同
- 全组织强制隐私模式
- 集中团队计费
- 带有使用统计的管理员仪表板
- SAML/OIDC 单点登录

Premium 模型

GPT-4、GPT-4o 和 Claude 3.5 Sonnet 都被视为 premium 模型。每次请求 Claude 3.5 Haiku 被计为 1/3 次 premium 模型请求。

专业试用

所有新用户均可获得 14 天的专业试用，允许访问所有专业功能和 250 次快速 premium 模型请求。在 14 天期满后，未升级的用户将恢复到爱好者计划。

快速和慢速请求

默认情况下，Cursor 服务器会尽量为所有用户提供快速 premium 模型请求。然而，在高峰期，快速 premium 信用用尽的用户将被转移到慢速池，这基本上是等待快速 premium 请求可用的用户队列。

这个队列是公平的，Cursor 将尽一切可能保持队列尽可能短。然而，如果需要更多快速 premium 信用且不想等待，可以在设置页面上添加更多请求。

检查您的使用情况

可以在 Cursor 设置页面上检查使用情况。也可以在 Cursor 应用程序内访问此页面，路径为 Cursor Settings > General > Account，并按“管理订阅”以供专业用户使用，或“管理”以供企业用户使用。

Cursor 使用情况每月重置，基于用户的订阅开始日期。

对话技巧

多用 Git 保存分阶段提交可用的代码，虽然 Cursor 生成代码的能力很强，但是有时生成大量的代码被 Accept All 后容易让整个项目运行失败，建议分阶段提交代码，这样遇到问题可以回滚。
一个 Cursor 窗口打开一整个项目，Cursor 默认的上下文目录就是这个目录，如果这个目录的范围太大或者太小都不利于代码生成。
Cursor 有时会只给出修改的部分代码，其它则忽略掉，这样在 Accept All 后会破坏已有的代码。可以在全局 rules 里写明要生成完整的代码，也可以在对话的时候强调。

Privacy Policy

TLDR

If you enable “Privacy Mode” in Cursor’s settings: zero data retention will be enabled, and none of your code will ever be stored or trained on by us or any third-party.
If you choose to keep “Privacy Mode” off, we collect telemetry and usage data. This may include prompts, editor actions, code snippets, and edits made to this code. We use this data to evaluate and improve our AI features.
With “Privacy Mode” off, if you use autocomplete, Fireworks (our inference provider) may also collect prompts to improve inference speed.

Other notes

Even if you use your API key, your requests will still go through our backend! That’s where we do our final prompt building.
If you choose to index your codebase, Cursor will upload your codebase in small chunks to our server to compute embeddings, but all plaintext code ceases to exist after the life of the request. The embeddings and metadata about your codebase (hashes, file names) may be stored in our database, but none of your code is.
We temporarily cache file contents on our servers to reduce latency and network usage. The files are encrypted using unique client-generated keys, and these encryption keys only exist on our servers for the duration of a request. All cached file contents are temporary, never permanently stored, and never used as training data when privacy mode is enabled.

Changelog

https://www.cursor.com/changelog

Automated and improved rules (0.49.x - April 15, 2025)

You can now generate rules directly from a conversation using the /Generate Cursor Rules command. This is useful when you want to capture the existing context of a conversation to reuse later.

For Auto Attached rules with path patterns defined, Agent will now automatically apply the right rules when reading or writing files

We’ve also fixed a long-standing issue where Always attached rules now persist across longer conversations. Agent can now also edit rules reliably.

cursor6

cursor1

cursor4

More: https://docs.cursor.com/context/rules

More accessible history (0.49.x - April 15, 2025)

Chat history has moved into the command palette. You can access it from the “Show history button” in Chat as well as through the Show Chat History command

cursor3

Making reviews easier (0.49.x - April 15, 2025)

Reviewing agent generated code is now easier with a built-in diff view at the end of each conversation. You’ll find the Review changes button at the bottom of chat after a message from the agent.

Images in MCP (0.49.x - April 15, 2025)

You can now pass images as part of the context in MCP servers. This helps when screenshots, UI mocks, or diagrams add essential context to a question or prompt.

Improved agent terminal control (0.49.x - April 15, 2025)

We’ve added more control for you over terminals started by the agent. Commands can now be edited before they run, or skipped entirely. We’ve also renamed “Pop-out” to “Move to background” to better reflect what it does.

Global ignore files (0.49.x - April 15, 2025)

You can now define global ignore patterns that apply across all projects via your user-level settings. This keeps noisy or sensitive files like build outputs or secrets out of prompts, without needing per-project configuration.

Ignore Files - Control which files Cursor’s AI features and indexing can access using .cursorignore and .cursorindexingignore

Cursor reads and indexes your project’s codebase to power its features. You can control which directories and files Cursor can access by adding a .cursorignore file to your root directory.

Cursor makes its best effort to block access to files listed in .cursorignore from:

Codebase indexing
Code accessible by Tab, Chat, and ⌘K
Code accessible via @ symbol references

Note: Tool calls initiated by Cursor’s Chat feature to services like Terminal and MCP servers are not currently able to block access to code governed by .cursorignore

Global Ignore Files - You can now define ignore patterns that apply across all projects via your user-level settings. This keeps noisy or sensitive files like build outputs or secrets out of prompts, without needing per-project configuration.

Why Ignore Files? - There are two common reasons to configure Cursor to ignore portions of your codebase:

Security

While your codebase is not permanently stored on Cursor’s servers or the LLMs that power its features, you may still want to restrict access to certain files for security reasons, such as files containing API keys, database credentials, and other secrets.

Cursor makes its best effort to block access to ignored files, but due to unpredictable LLM behavior, we cannot guarantee these files will never be exposed.

Performance

If you work in a monorepo or very large codebase where significant portions are irrelevant to the code you’re developing, you might consider configuring Cursor to ignore these parts of the application.

By excluding irrelevant parts of the codebase, Cursor will index large codebases faster and find files with more speed and accuracy when searching for context.

Cursor is designed to support large codebases and is skilled at assessing file relevancy, but the ignore feature is helpful when using a codebase that is especially large or includes files immaterial to your development.

Configuring .cursorignore

To implement Cursor’s ignore feature, add a .cursorignore file to the root of your codebase’s directory and list the directories and files to be ignored.

The .cursorignore file uses pattern matching syntax identical to that used in .gitignore files.

Basic Pattern Examples

# Ignore specific file `config.json`
config.json

# Ignore `dist` directory and all files inside
dist/

# Ignore all files with a `.log` extension
*.log

Advanced Pattern Examples

# Ignore entire codebase
*

# Do not ignore `app` directory
!app/

# Ignores logs directories in any directory
**/logs

New models (0.49.x - April 15, 2025)

We’ve recently added many more models you can use. Try out Gemini 2.5 Pro, Gemini 2.5 Flash, Grok 3, Grok 3 Mini, GPT-4.1, o3 and o4-mini from model settings.

Models - Comprehensive guide to Cursor’s models: features, pricing, context windows, and hosting details for Chat and CMD+K