Getting Started with Cursor

Cursor is an AI-first code editor and coding agent: you describe what to build or change in natural language, and Cursor helps you navigate the codebase, plan and apply edits, debug, and review diffs alongside familiar IDE workflows. The product pairs a text editor and tooling (search, terminal, VCS) with an agent that orchestrates instructions, tools (files, search, shell, and more, depending on setup), and the model you choose. See the high-level product framing in the Cursor documentation and the mental model in How the agent works.

This page summarizes concepts that are stable to teach in a workshop and calls out Cursor 3 changes that affect daily workflow.

What you install and where it runs

• Editor: a desktop app (Windows, macOS, Linux) you install with your platform’s quickstart flow, Linux APT install and related steps live under Get started.
• Optional CLI: Cursor ships a CLI for headless and automation flows (for example agent in CI). Use the current Cursor CLI installation instructions for your OS (including a Windows PowerShell one-liner in the official docs).
• Cloud agent (optional): a separate line of product for API-driven or remotely triggered work on repositories, documented under Cloud agent and related pages, useful to know it exists, distinct from the local editor when you are teaching “the IDE you open every day.”

Teams who standardize on Cursor should read Team pricing and usage for how seats, included usage credit, and on-demand spend work, handy if someone asks how billing differs from a flat “per seat” plan.

The agent in one paragraph

In Cursor’s own framing, the agent is built on three pieces: instructions (system prompt, Rules, and project context), tools (editing, codebase search, terminal execution, optional extras such as browser tooling where enabled), and the model you pick. Cursor aims to keep per-model tuning inside the product so you can switch frontier models as they ship without re-learning every low-level API yourself, see How Agent works.

Subagents (delegated specialists with their own context) and parallel patterns are covered in Subagents.

Cursor 3: what changed and why it matters

Cursor 3.0 is a meaningful workflow and platform update, not just a version bump. The parts called out in official documentation include:

1. Git worktrees and the end of the old “parallel agents” worktree story

Automatic worktree management was redesigned in Cursor 3.0 and replaces the previous “parallel agents” worktree feature as you knew it. The docs describe worktrees as now “fully agentic”: you can spin up or attach work to a separate Git checkout at more flexible points in a conversation’s lifecycle, not only at the very start. New chat commands and deeper CLI support accompany that change, see Worktrees and Using the agent in the CLI (worktrees).

Practical commands (illustrated in the docs):

• /worktree: start a task in an isolated checkout so experiments do not trample your main branch. Example from the product docs:

  /worktree fix the failing auth tests and update the login copy

• /best-of-n: run the same prompt with several models in parallel, each in its own worktree, so you can compare outcomes:

  /best-of-n sonnet,gpt,composer fix the flaky logout test

• CLI: agent --worktree runs the Cursor CLI agent in a managed worktree (paths such as ~/.cursor/worktrees and retention are described in the same documentation).

If you are teaching migration from an older mental model, the one-line distinction in How is this different from the previous parallel agents feature? is the official anchor: 3.0’s worktree system subsumes the older parallel-agent worktree approach with new commands and CLI integration.

2. Agents window: a unified, agent-first surface

The Agents window is described as an agent-first workspace that unifies how you work with agents across local, cloud, and remote SSH environments. Official copy positions it as combining what you liked about parallel agents with deeper editor-style control, read the page for the current feature list.

Exclusives called out in the same documentation include Multi-workspace (managing agents across projects), a diffs experience aimed at reviewing changes and PRs, and parallel agents in the cloud with integrations (phone, web, Slack, GitHub, Linear, exact availability evolves with the product, so point learners at the live docs page for the up-to-date matrix).

3. Ecosystem: Cloud agent API and CI

Cursor 3 lands in an ecosystem that already includes Cloud agent HTTP APIs (e.g. creating agents against a repository with PR targets) and GitHub Actions patterns that install the Cursor CLI and run agent -p. That is not “day-one beginner” material, but it answers “how does this scale past my laptop?”, see Cloud agent API and GitHub Actions.

When official docs change