What is Cursor

Definition:

Cursor is a software development environment created by Anysphere that combines a code editor with artificial intelligence features and coding agents. It allows users to write, analyze and modify files through natural language instructions, as well as run tools and review the results of changes.

Its editor is based on the Visual Studio Code codebase. The product also offers agents that work in remote environments, so its capabilities extend beyond autocomplete within a desktop application.

The Cursor editor and Visual Studio Code

Cursor uses the Visual Studio Code codebase and retains elements of its environment, such as the file explorer, integrated terminal, keyboard shortcuts and editor settings. It is a separate application, not an extension installed inside Visual Studio Code.

It is available for Windows, macOS and Linux. Settings, themes, shortcuts and extensions can be imported from a Visual Studio Code installation, although compatibility is not identical in every case.

Cursor uses the Open VSX extension registry. Many common extensions are available, but some may be missing or behave differently. The similarity between the interfaces does not mean that both products share every feature or the same update schedule.

Autocomplete, editing and agents

Cursor brings together different forms of assistance that vary in the scope of their intervention. Some suggest changes while code is being written; others carry out a task through several steps.

Its main options include the following:

• Tab: suggests code and changes based on surrounding content, recent edits and detected errors. It can complete several lines and suggest related changes elsewhere in the project.
• Instruction-based editing: allows a change to a snippet or file to be described without manually writing the entire implementation.
• Agent: can search the project for information, modify several files, run commands and use their results to continue the work.
• Plan Mode: examines the code and requirements, asks questions when clarification is needed and prepares a plan before implementation begins.

For example, adding a field to a form may require changes to the interface, validation and tests. Autocomplete helps write each part; an agent can approach the task as a set of related changes.

Project context and instructions

Cursor’s responses depend on the model used and the context it receives. That context can include files, selected snippets, search results, error messages and project instructions.

Rules allow instructions on coding conventions, folder structure or working procedures to be retained. Project rules can be stored in .cursor/rules, while AGENTS.md provides a simple alternative for documenting instructions intended for agents.

Not every rule has to apply to every intervention. Rules can be associated with particular files or included when they are relevant to the task.

These instructions form part of the agent’s context. Saving or changing them does not amount to retraining the language model, nor does it guarantee that the entire repository is included in every request.

Local agents, cloud agents and Projects

Agents can work on a local project or use a remote environment. In the first case, tools operate on the files and resources available on the computer. In the cloud, work runs on a separate machine and can continue even when the user’s computer is no longer connected.

The location where tools run should not be confused with where the model processes requests. Working with local files does not necessarily mean that all artificial intelligence inference takes place on the same computer.

Cursor also offers Projects, introduced in September 2026 as a beta feature. It organizes longer-running work around a coordinator that plans and delegates tasks, while other agents carry out the implementation.

A Project can bring together shared files, instructions and results from different agents. It can also receive work through events or schedules. Its beta status distinguishes it from a feature that is generally and stably available to all accounts.

Getting started and reviewing changes

Using the editor begins with installing it, signing in to an account and opening a working folder. An existing Visual Studio Code configuration can be imported from Cursor’s settings.

An initial task might involve requesting an explanation of the project structure or locating where a function is implemented. To make modifications, the agent uses the available tools and presents the changes for review.

File comparisons show which lines have been added, removed or modified. Commands and tests provide information about the program’s behavior, distinct from a visual review of the code.

Cursor provides restoration points, or checkpoints, to recover earlier file states during the agent’s work. These mechanisms are independent of Git: they do not replace a permanent version history or a team’s shared change record.

Relationship with other coding agents

Cursor combines work within an editor with the capabilities of a coding agent. It is therefore inaccurate to contrast it with tools such as Codex as though one were only an editor and the other could carry out tasks.

Both can analyze projects, modify files and run tools. Their differences lie in their interfaces, integrations, execution environments and ways of organizing work.

Cursor is not the name of a single language model either. The product integrates models and tools within a development environment; the selected model is one of its components, not the entire application.