Menu
Products

Cursor

4 min readbeginnerUpdated 28 Sept 2026
1 · In one line

Cursor is a coding tool with an AI agent that can read your project, edit files and run commands for you.

1 · What it is

Cursor describes itself as a coding agent. It helps you understand a codebase, build features, fix bugs and review changes. A codebase is all the files that make up one software project. Cursor helps in two sizes.

The small one is Tab, Cursor’s autocomplete. Like predictive text on a phone, it suggests what to type next. It looks at your recent edits, the code nearby and linter errors. Linters are tools that warn you about likely mistakes. The large one is the agent. You pick which AI model it uses for a task. Then it can search, edit files and run commands on its own.

Cloud agents do their work in the cloud, not on your computer. Each one runs in its own isolated virtual machine, a computer simulated in software.

Cursor builds its code search index on your own machine. The index works like the one at the back of a book. Cursor does not keep embeddings of your project for search. Embeddings are lists of numbers that capture meaning. But files the agent opens can still be sent to the model. So check the privacy settings before you use it on private code.

2 · Why it exists

Writing code with AI raises three practical problems.

Finding the right codeA real project has many files, so the tool has to search for the parts that matter before it can help.
Mistakes look fineAI-written code can look correct and still miss edge cases or add security problems.
Undoing a wrong turnAn agent that edits many files at once needs a simple way back when it goes in the wrong direction.
3 · How it works

Follow one bug fix through Cursor's agent.

How Cursor's agent handles one coding task You describe a task. The agent searches the codebase, with project rules in its prompt. In the highlighted tool loop it edits files and runs terminal commands, repeating as needed, and saves a checkpoint before big changes. You review the diff and keep it, redirect the agent, or restore the checkpoint. ONE TASK FOR CURSOR'S AGENT Your task Search Tool loop Diff You review Checkpoint fix the crash at score 100 grep code + rules in prompt edit files, run commands, repeat every changed line, shown live keep, redirect or restore snapshot of changed files, saved before big edits before big edit redirect and ask again
  1. 1 · askYou describe a task in the agent side panel, for example a bug to fix.
  2. 2 · searchThe agent searches your code for names, error messages and patterns to find the relevant files.
  3. 3 · actIt edits files and runs terminal commands, making as many tool calls as the task needs.
  4. 4 · saveBefore significant changes it saves a checkpoint, a snapshot of the files it touched.
  5. 5 · reviewYou watch the diff, which shows each change, and cancel, redirect or restore a checkpoint.

The agent does the typing; you still decide what gets kept.

4 · Where it's used
WhoWhat they askWhat it works with
Student“Why does my game crash when the score passes 100, and how do I fix it”The game's code files, found by searching for names and error messages
Web developer“Add a dark mode switch that follows our style rules”Rule files added to the agent's prompt
Backend team“Fix this bug in the cloud on its own branch”A copy of the project in a cloud virtual machine, with changes pushed back
5 · What it solves, and what it doesn't
solves
  • While you write, it proposes the next bit of code, including edits across several lines and in related files.
  • Its agent can search the codebase and the web, edit files and run terminal commands.
  • Rule files give the agent the same project instructions every time.
  • Cloud agents can work in parallel on separate branches and push changes back for handoff.
doesn't solve
  • It does not guarantee the code is right, so the changes still need human review and tests.
  • Checkpoints are not a replacement for Git, the usual tool for saving code history.
  • The model does not remember past sessions on its own; rules are how you give it lasting guidance.
  • With Privacy Mode off, Cursor may use code data to train its models, so the setting matters.
6 · Go deeper
AI agentsCode generationGitHub CopilotClaude CodeVibe coding

Sources used

This explainer is written in original language. The links below support its factual claims.

  1. docsCursor Docs, Cursor · read 28 Sept 2026
  2. docsAgent overview, Cursor · read 28 Sept 2026
  3. docsTab, Cursor · read 28 Sept 2026
  4. docsRules, Cursor · read 28 Sept 2026
  5. docsSearch, Cursor · read 28 Sept 2026
  6. docsCloud agents, Cursor · read 28 Sept 2026
  7. officialReviewing and Testing Code, Cursor · read 28 Sept 2026
  8. officialData Use and Privacy Overview, Cursor · read 28 Sept 2026
Suggest a correction