Skip to content

Cursor AI: A Beginner Guide to the AI Code Editor

A developer at a desk looking at a large monitor showing a dark code editor with an AI chat panel beside blurred code


A developer at a desk looking at a large monitor showing a dark code editor with an AI chat panel beside blurred code

Cursor AI is an AI code editor built on Visual Studio Code that completes your code, answers questions about your project, and edits files when you ask.

Open it for the first time and almost nothing looks new. Anysphere built Cursor as a fork of Visual Studio Code, so it feels like home on day one.

What Cursor AI is

Cursor helps at three distances. Tab watches you type and predicts the next edit. Press Tab to take the suggestion or Esc to drop it. Inline Edit handles one selected chunk: highlight code, press Cmd+K on a Mac or Ctrl+K on Windows and Linux, describe the change, and accept the diff in place.

The Agent panel goes further. It searches your codebase and edits files across folders, then runs terminal commands to check its own work. The Agent opens with Cmd+I or Ctrl+I. Cursor’s Agent docs describe an assistant that finishes complex coding tasks alone, with no cap on tool calls.

The Claude Code guide covers a terminal-first agent with a similar brief, and the OpenAI Codex guide covers another agent that works across a whole codebase.


A close-up of a monitor showing a dark code editor with blurred colored code blocks and a keyboard in the foreground

The safe first hour with Cursor

Download and install Cursor like any desktop app. On first launch it offers to import your Visual Studio Code extensions and themes, plus your keybindings. Say yes. Your setup carries over in a minute.

Start with questions, not edits. Ask Cursor to explain the codebase. Then ask it to suggest a few small, safe improvements and pick one. Fix a label.

Review the diff, then run your project’s checks. Commit your code before you hand the Agent any task, because Git undoes the whole run. Never paste passwords, API keys, or private customer data into a prompt. The editor reads a .cursorignore file, so list secret files there.


Hands typing on a laptop showing a dark code editor with blurred colored code blocks, a notebook and coffee cup on the desk

Ask, Agent, Plan, and Debug: pick the mode first

One panel holds four modes, and the mode decides what the AI may do. The model dropdown picks which AI does the work. Rotate modes with Shift+Tab or the mode picker.

Mode What it does Use it when
Ask Reads your codebase and answers. Changes nothing. Understanding code before changing it.
Agent Edits files and runs commands end to end. Known changes you want done.
Plan Researches and writes a plan you approve first. Multi-file tasks or unclear approaches.
Debug Gathers runtime evidence before fixing a bug. Bugs you can reproduce but not explain.

Ask is the safest first step, because it cannot change a file. Plan earns its keep on bigger work: it asks clarifying questions, researches the project, and writes a reviewable plan. You approve the plan before the build starts.

For a quick fix, go straight to Agent. Ignore older guides that teach Manual mode or separate Chat and Composer panels.


A developer studying two monitors showing blurred green and red code comparison blocks, with a notebook and coffee on the desk

Rules are how Cursor remembers your project

Models forget everything between sessions, so Cursor gives you Rules that open every Agent session. Project Rules live in a .cursor/rules folder inside your repo as .mdc files. Git tracks them like any other file, so the team shares them.

One rule can pin down naming style or ban a library you dropped. Rules only work when you write them plainly and keep them short. A rule that says “follow the codebase style” tells the model nothing it can act on. Name the test command.

One trap: a plain .md file inside .cursor/rules gets ignored, because the rules system needs .mdc frontmatter.


A developer writing in a notebook beside a monitor showing a folder and file tree of document icons

What Cursor costs now

It now sells through one subscription shared with Grok and X. The free plan costs $0 and gives limited Agent and Chat requests. Plus costs $8 a month and covers Grok models only. Premium costs $30 a month with four times the usage of Plus, plus cloud agents and code reviews.

Super costs $100 a month with fifteen times the usage of Plus. Ultra costs $200 a month with fifty times the usage of Plus. Teams start at $30 per user a month. Enterprise pricing stays custom.

Annual billing: Plus runs $80 a year, Premium $300, Super $1,000, Ultra $2,000. Prices exclude tax. The usage pool refreshes weekly and covers its own models: Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5.

Third-party models cost extra. That surprises people. Premium and above let you use models from Claude, GPT, or Gemini at API list price, paid from prepaid credit.

Cursor’s pricing page states that third-party model usage is not included in the monthly prices, and the models and pricing docs confirm that the included pool covers its own models while third-party models bill at API list price. Set a spending cap.

A $20 Pro plan in older guides is a legacy plan that keeps working for existing subscribers. The GPT-4o guide covers one such family. The Claude Sonnet guide and the Gemini 3 guide cover two more.


A laptop showing a clean page with four blank plan cards, a wallet and a coffee cup on a wooden desk

The catch: your code leaves your machine

Every AI request travels through Cursor’s servers, where the final prompt gets assembled from the code and context the Agent gathered, and then on to the model provider. Privacy Mode changes what happens next. With Privacy Mode on, Cursor does not use your data for training and holds zero data retention agreements with its providers, so providers store nothing and train on nothing.

With Privacy Mode off, Cursor may store your prompts and snippets, plus editor actions, and use them to improve features and train models. Cursor’s data use overview spells out both settings and notes that prompts flagged by abuse detectors may still be stored for investigation. Turn Privacy Mode on for client code.

Review every change. The Agent sometimes edits the wrong part of a file or produces code that looks right and breaks in an edge case. The safe loop handles all of it: Ask first, take one small Agent task, review the diff, run your checks, commit.

The ChatGPT Agent guide covers a different agent with the same review habit, and the best AI tools compared puts Cursor beside other AI tools.


A person touching toggle switches on a settings screen, with a small padlock on the desk beside the keyboard

Frequently asked questions about Cursor AI

What is Cursor AI?

Cursor AI is an AI code editor built by Anysphere on a fork of Visual Studio Code. It adds Tab autocomplete and inline edits, plus an Agent panel for edits across your project.

Is Cursor AI free?

Cursor has a free plan with limited Agent and Chat requests. Paid plans run $8 a month for Plus to $200 for Ultra. Plus covers Grok models only.

Is Cursor AI safe for beginners?

Start in Ask mode, commit before Agent edits, review every diff, and never paste secrets. Use Privacy Mode for proprietary code.

Which AI models does Cursor use?

The included usage pool covers Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5. Premium and above add Claude, GPT, or Gemini models at API list price, paid from prepaid credit.

Does Cursor work with VS Code extensions?

Cursor imports your Visual Studio Code extensions and themes, plus your keybindings, on first launch. Most extensions carry over.

Can Cursor run terminal commands?

The Agent runs shell commands and watches the output, so it can run tests after an edit. Review each command before it runs on a shared machine.

Leave a Comment