What is Cursor AI?
By Mari Luukkainen · 10 min read · Last updated
On this page
- What does Cursor do?
- Who is Cursor for?
- How does a first session in Cursor go?
- What I built with Cursor in March 2025
- How do you give Cursor project rules?
- Prices and limits (checked 4 November 2026)
- Who owns Cursor?
- Where does Cursor break?
- What happens to your code if you stop using Cursor?
- What are the alternatives to Cursor?
- Frequently asked questions
Cursor AI is a code editor with a built-in coding agent. It is based on the VS Code codebase, so VS Code users get the same layout and shortcuts. You describe a task, and the agent reads your project, edits files and runs terminal commands. Since August 2026 Cursor has been owned by SpaceX.
I started with Cursor in March 2025. I was a VC at the time, ChatGPT pointed me at it, and I built my first Next.js app on a two hour flight from Helsinki to Berlin (post). I still use Cursor today, next to Claude Code. Plans and prices on this page were checked on 4 November 2026.
Ask AI about this pageChatGPTClaudePerplexityGoogle AI ModeGrok
What does Cursor do?
Cursor is a desktop app for macOS, Windows and Linux. You open a project folder in it, and you get a file tree, an editor and an agent panel. Cursor's own docs call it "a coding agent for building ambitious software" (docs). In practice it is four tools in one window:
| Part | What it does | Source |
|---|---|---|
| Agent | Opens in the side panel with Cmd+I. It searches the codebase and the web, edits files, runs terminal commands and can control a browser to check its own work | Agent docs |
| Plan Mode | Shift+Tab in the agent input. The agent researches the code, asks questions and writes a plan you approve before it builds | Plan Mode docs |
| Tab | Autocomplete that suggests the next edit as you type, including edits in other files. Tab accepts, Escape rejects | Tab help |
| Cloud agents and CLI | Cloud agents run tasks in cloud machines and produce pull requests. The CLI runs the same agent in a terminal with the command agent | Cloud agents, CLI |
The agent also makes checkpoints. Before significant changes it saves a snapshot of the files it touches, and you can restore one from the chat timeline. Checkpoints are stored locally and separately from Git, and Cursor's docs say to use them only for undoing agent changes (agent docs).
Who is Cursor for?
- People who want to see the code: every change shows up as a diff in the file you have open. You learn what a component, a route or an environment variable is by watching them appear.
- VS Code users: settings, keybindings, themes and extensions import in one click from Cursor Settings, General, Account, VS Code Import (migration help).
- Developers who switch between small edits and big tasks: Tab handles the small typing, the agent handles the multi-file work, and both live in the same window.
Cursor is a weaker fit if you have never opened a terminal and do not want to. It runs on your computer, so you install Node.js, Git and whatever your project needs yourself, or ask the agent to do it. If you want a hosted app with no setup at all, a browser builder such as Lovable, Replit or [Base44](/base44-ai-app-builder) starts faster.
How does a first session in Cursor go?
This follows Cursor's own quickstart (docs), with the prompts I would give a beginner.
- Download Cursor from cursor.com/downloads. It needs macOS 12 or later, or Windows 10 or later. Linux has apt and dnf packages and an AppImage.
- Open the app and sign in. If you used VS Code before, import your setup now.
- Create an empty folder, for example paws-site, and open it with File, Open Folder.
- Open the agent with Cmd+I (Ctrl+I on Windows and Linux) and give it the first prompt below.
- Read the plan, answer its questions, then let it build. Terminal commands ask for your approval by default. Read each one before you click run.
- Open the page it starts in your browser. Check it on a phone-sized window.
- Ask for one change at a time. After each change that works, commit with Git so you have a point to return to.
Create a one-page website for a dog groomer called Paws & Co, using Next.js, TypeScript and Tailwind CSS.
Sections: a header with the name and a phone number, three services with prices, opening hours, and a contact form with name, email and message.
The form does not need to send anything yet; show a thank-you message after submit.
Mobile first. No gradients, no stock photo placeholders.
Before you write code: tell me which commands you will run and why. Then set up Git and make the first commit.Explain this codebase. Point me to the main entry points, key modules, and anything I should read before making changes.The second prompt is the one in Cursor's quickstart, and it is a good habit for any folder you did not write yourself. For anything that touches more than two or three files, press Shift+Tab to switch to Plan Mode first and use a prompt like this:
I want visitors to book a grooming slot from the contact section.
Plan this before writing code: which files change, what new files you add, how the date and time are validated, and what you will not touch.
Ask me anything that is unclear. Wait for my approval before building.After the agent finishes, ask it to run the checks the project has: the type checker, the linter, the tests and a build. Then read the diff yourself before you commit.
Run the type checker, the linter and the build, and show me the output. Then list every file you changed in this session and one line on why. Flag anything you are unsure about.What I built with Cursor in March 2025
In March 2025 I vibe coded 12 websites in one weekend with Cursor (post). I made a lead generation boilerplate and a migration boilerplate first, then reused them. A marketing funnel took 17 minutes. I used ChatGPT to plan and explain things and Cursor to write and change the code.
The start was messy. My code was full of typos, things broke often, and I asked ChatGPT how to use md on a Mac. Planning first and splitting the work into small steps is what fixed it.
Later that month I gave Cursor one prompt on a minimal landing page: "SEO the shit out of it." It added titles, meta descriptions, Open Graph tags, alt text and some internal links. It also added meta keyword tags, which are outdated (post). In the post I shrugged that off.
How do you give Cursor project rules?
Rules are instructions the agent reads every time, so you stop repeating yourself. Cursor reads two kinds (rules docs):
- AGENTS.md, a plain Markdown file in the project root or in subfolders. Other agents read it too, so it moves with the project.
- .cursor/rules, a folder of .mdc files with a short header that says when each rule applies. A plain .md file in that folder is ignored.
Start with AGENTS.md. Add a line when you notice you have typed the same instruction three times.
# Paws & Co site
- Next.js, TypeScript, Tailwind CSS. Do not add other frameworks.
- Components live in components/, one per file, PascalCase names.
- Run npm run lint and npm run build after every change and show the output.
- Do not add npm packages without asking first.
- Never edit .env files. Ask me for values instead.Prices and limits (checked 4 November 2026)
Cursor's plans changed completely in 2026. All prices are US dollars before tax. The full breakdown, with third-party model costs and spend limits, is on Cursor pricing.
| Plan | Monthly | Annual | What you get | Source |
|---|---|---|---|---|
| Free | $0 | $0 | Agent and Chat with limited usage. No numbers given | Pricing |
| Plus | $8 | $80 a year | Base usage pool. Cursor's own models only (Grok 4.7, 4.6, 4.5 and Composer 2.5). No cloud agents, no third-party models, no own API keys | Models and pricing |
| Premium | $30 | $300 a year | 4x Plus usage. Cloud agents, automations, Bugbot code review, third-party models at API list price | Models and pricing |
| Super | $100 | $1,000 a year | 15x Plus usage, priority access at peak times | Models and pricing |
| Ultra | $200 | $2,000 a year | 50x Plus usage, earliest access to new models | Models and pricing |
| Teams | $30, $100 or $200 per seat | Not listed | Usage matches the seat's tier, central billing, SSO, team-wide Privacy Mode | Models and pricing |
Three things in the fine print matter more than the price:
- The usage pool refreshes weekly, and it is shared with Grok. If you also use the Grok apps on the same subscription, they draw from the same pool (models and pricing).
- Claude, GPT and Gemini models are not included in any plan. On Premium and above you can use them at API list price, paid from prepaid credit. On Plus and Free you cannot use them at all (The X Subscription).
- If you had an old Pro, Pro+ or Ultra plan, Cursor says it keeps working as it was. Switching to The X Subscription is optional, and the switch is permanent (legacy plans).
Who owns Cursor?
SpaceX. Engadget reported on 16 June 2026 that SpaceX agreed to buy Cursor for $60 billion in stock (Engadget), and the deal closed on 14 August 2026 (GIGAZINE). The Next Web reported that Cursor's team now also works on Grok products (The Next Web).
For you, the visible change is the plan structure above. Cursor is now sold as part of The X Subscription, one subscription for Grok, Cursor and X, and the included models are Grok and Cursor's own Composer (The X Subscription). If your work depends on a specific Claude or GPT model, budget for API-priced usage on top of the plan.
Where does Cursor break?
- The agent edits files without asking: Cursor's security docs say agents can change workspace files without approval, except configuration files, and the changes are saved to disk immediately. Terminal commands need approval by default (agent security). Commit before every task so a bad change costs one git restore.
- Auto-approval has limits: Run Modes let trusted commands run without a prompt, up to an Auto-review classifier. Cursor calls them "best-effort guardrails rather than a hard security boundary" (agent security).
- Untrusted repositories: workspace trust is off by default, and turning on restricted mode breaks the AI features. For code you do not trust, Cursor's docs say to use a basic text editor (agent security).
- Extensions: Cursor uses the Open VSX registry, not Microsoft's marketplace. Most popular extensions are there, but not all of them (extensions help).
- Accepting without reading: the editor makes it easy to accept a long diff in one click. My meta keyword tags got in that way, and a missing auth check gets in the same way.
- Costs that depend on the model: the plan price covers Grok and Composer. Pick a third-party model and every request is billed at API list price on top.
What happens to your code if you stop using Cursor?
Nothing. Cursor edits ordinary files in a folder on your computer, so there is nothing to export. Your Git history, your package.json and your AGENTS.md all stay where they are. The only Cursor-specific parts are the .mdc files in .cursor/rules and your chat history.
To move to another tool:
- Copy the useful rules from .cursor/rules into AGENTS.md, which most coding agents read.
- Commit and push to GitHub or another Git host.
- Open the same folder in the new editor or agent and ask it to explain the project before you change anything.
What are the alternatives to Cursor?
| If you want | Look at | Read |
|---|---|---|
| Another AI editor with its own agent | Devin Desktop, formerly Windsurf | Cursor vs Windsurf |
| An editor with AI, but Microsoft's VS Code and marketplace | VS Code with GitHub Copilot | GitHub Copilot vs Claude Code |
| An agent you hand whole tasks to, in a terminal or editor extension | Claude Code, Codex, OpenCode | Cursor vs Claude Code, Claude Code alternatives |
| No install, the app hosted for you | Lovable, Replit, Base44, Bolt.new, v0 | Best AI app builders |
| The full list with prices | Editors, agents and builders side by side | Best vibe coding tools |
I use Cursor and Claude Code together. The two tools do not conflict as long as only one of them edits a file at a time, and a commit between them gives you a point to return to.
Frequently asked questions
Is my code used to train AI models in Cursor?
Not with Privacy Mode on. Turn it on in Cursor Settings, General. Cursor still sends prompts and code context to model providers such as OpenAI, Anthropic and Google to answer your requests. For teams, Privacy Mode is on by default.
Can I use my own OpenAI or Anthropic API key in Cursor?
Not on the Plus plan; Cursor's docs list own API keys among the things Plus does not include. On Premium and above, third-party models are available at API list price from prepaid credit.
Does Cursor work without the internet?
The editor opens offline, but the agent, Chat and Tab send requests to model providers, so the AI features need a connection.
Can I run Cursor and VS Code on the same computer?
Yes. They are separate applications, and you can open the same project in both. Cursor uses the same default keyboard shortcuts as VS Code.
Is there a Cursor command line tool?
Yes. Install it with curl https://cursor.com/install -fsS | bash on macOS, Linux and WSL, then start it with the command agent. It has the same Agent, Plan and Ask modes as the editor.
What happened to the $20 Cursor Pro plan?
New customers get the Free, Plus, Premium, Super and Ultra plans of The X Subscription. Existing Pro, Pro+ and Ultra subscribers keep their old plan, with its two monthly usage pools, until they switch or cancel.
Next guideCursor pricing →