What is Cursor

Cursor is an AI code editor that describes itself as a coding agent for building ambitious software. It looks and feels much like VS Code, but AI is built into the core of the editor: you can ask it to explain a codebase, plan and build features, track down bugs and review changes, while keeping the extensions and habits you already have.

What separates it from a chat-style assistant is Agent. An agent in Cursor has three parts: instructions (the system prompt plus any rules you set), tools (file editing, codebase search, terminal execution and more) and a model (whichever one you pick for the task). Cursor tunes the instructions and tools separately for each frontier model it supports, so you choose a model, describe the job and let it handle the model-specific details.

Beyond the desktop editor, the product line includes Cursor CLI for the terminal, Cloud Agents that run in remote virtual machines, an iOS app and web interface, and Bugbot for automated pull request review. Together they cover the path from local coding to team workflows.

It helps to think of Cursor’s AI in three layers. Closest to your hands is Tab, which predicts your next edit as you type without breaking your flow. In the middle is Agent in the editor: give it a task and it reads code, edits files and runs commands locally. Furthest out are Cloud Agents, which work in their own virtual machines; you can send off several at once and come back to review their pull requests. In practice you mix all three: Tab for small edits, Agent for chunks of a feature, the cloud for long, independent jobs.

Key features

Agent. Opened in the side pane with Cmd+I, Agent handles complex tasks on its own. It searches files and folders, reads code (and images such as screenshots), applies edits, runs terminal commands, searches the web and can even drive a browser to take screenshots and test pages. The docs note there is no limit on how many tool calls it can make in one task.

Checkpoints. Before significant changes, Agent saves a snapshot of the affected files. If it goes the wrong way, click any checkpoint in the chat timeline to preview and restore. Checkpoints are stored locally and separately from Git, and are meant for undoing Agent changes.

Tab completion. Tab suggests grey inline text based on your recent edits, the surrounding code and linter errors. It can change several lines at once and add missing imports. Press Tab again after accepting and it jumps to where it predicts your next edit will be, including related changes in other files.

Rules. Models do not remember anything between completions, so rules supply persistent context. Cursor supports project rules (in .cursor/rules, versioned with the repo), user rules, team rules and the simpler AGENTS.md format.

Modes. Both the editor and the CLI offer Agent (all tools), Plan (clarify and design before writing code) and Ask (read-only exploration).

Cloud Agents. These run in isolated VMs with full development environments, so they can build, test and demo their changes without your machine being online. Start them from the desktop app, cursor.com/agents, the iOS app, Slack, GitHub, Linear or the API.

Bugbot. It reads PR diffs and leaves comments on bugs, security issues and code quality problems, with suggested fixes. GitHub and GitLab are supported.

How to use it

  1. Install. Download Cursor for macOS 12 or later (Apple Silicon and Intel) or Windows 10 or later. On Linux, use the official apt or yum repositories, or the AppImage.
  2. Sign in and open a folder.
  3. Let it learn the codebase. Open Agent with Cmd+I and ask it to explain the project and point out what to read first.
  4. Start small. Ask for a small change, review the diff, then accept. Once comfortable, use Plan mode for larger features.
  5. Capture conventions as rules. Put team standards in .cursor/rules or AGENTS.md so every session follows them.

Terminal users can install Cursor CLI on macOS, Linux or WSL with curl https://cursor.com/install -fsS | bash and start a session with agent. Use agent -p "instruction" for non-interactive runs in scripts and CI. Prefix a message with & to hand the task to a Cloud Agent and keep it running while you are away.

You do not have to sit idle while Agent works. Queue a message to be handled after the current task, or send a follow-up immediately to steer it at its next tool call. When something is unclear, Agent can ask you a question and keep reading files and making edits while it waits for the answer.

Use cases

  • Everyday coding: Tab handles repetitive edits while Agent takes on whole features.
  • Large refactors and migrations: break the work down in Plan mode, execute step by step, with checkpoints as a fallback.
  • Frontend work: Agent can drive a browser to verify visual changes.
  • Team workflows: mention @cursor in Slack or on a GitHub issue to start a task, and let Bugbot review every PR.
  • Parallel work: run several Cloud Agents at once, each opening its own pull request.

FAQ

How does Cursor relate to VS Code? The editing experience is very close to VS Code, so the learning curve is small for VS Code users. The difference is that Agent, Tab and the rest of the AI features are part of the editor’s core.

Can I choose the model? Yes. Cursor supports models from several providers and you can switch in the Agent input. Context lengths and pricing for each model are listed on the official Models & Pricing page.

What if Agent breaks something? Restore files to an earlier state with a checkpoint. Cursor recommends using checkpoints only to undo Agent changes and relying on Git for real version control.

Can I skip the editor and use only the command line? Yes. Cursor CLI supports the same Agent, Plan and Ask modes and also runs in CI pipelines.

Official website