Cursor Overview
Cursor is an AI-native code editor built as a fork of VS Code. Rather than bolting AI on as an extension, Cursor rebuilds the editor so that AI has deep access to your entire codebase from the start. The core differentiator is full codebase awareness — Cursor indexes your project semantically so every chat, edit, and agent task understands the whole codebase, not just the open file.
Why Not Just an Extension?
Extensions like GitHub Copilot add AI on top of an editor that wasn't designed for it. Context is limited to what fits in a prompt. Cursor starts from a different premise: build an editor where the AI is a first-class citizen that can read, search, and act across your whole project. This shows up in multi-file edits, codebase-aware chat answers, and agents that can run terminal commands and tests without being told where things are.
Cursor 3 — April 2026
Cursor 3 shipped in April 2026 with a redesigned interface that shifts the primary model from file editing to managing parallel coding agents. Key additions:
Agent-first workspace
The UI is now built around agents, not files. Open multiple agents in parallel and switch between them.
Cloud Agents
Agents run on Cursor's remote infrastructure — build, test, and demo features end-to-end without your machine staying on.
Multi-repo execution
A single agent task can span repositories, useful for monorepos or cross-service refactors.
Plugin marketplace
Third-party integrations (Linear, Figma, Jira, Postgres, Supabase) available directly in the agent context.
/multitask
Run async subagents in parallel within the editor to handle multiple independent tasks simultaneously.
Context usage breakdown
Real-time view of what is consuming context — rules, files, skills, subagents — so you can diagnose slow or confused responses.
Who It's For
Good fit
- • Professional developers working in large, established codebases
- • Teams doing frequent multi-file refactors
- • Engineers who want to stay in a familiar VS Code-like environment
- • Anyone who wants to run autonomous agents on their code without switching to a CLI tool
Less ideal
- • You already pay for GitHub Copilot and mostly want inline suggestions — Copilot is cheaper for that use case
- • You need agents that go beyond the IDE (file system, browser, system apps) — Claude Code covers that better
- • You prefer full BYOM control with no subscription — Cline/Continue.dev let you bring your own API keys
Plans & Pricing
| Plan | Price | What You Get |
|---|---|---|
| Hobby | Free | Limited Agent requests, limited Tab completions — enough to try the editor |
| Pro | $20 / mo | $20 monthly usage credits for frontier models, unlimited Tab (Auto mode), MCPs, Cloud Agents |
| Pro+ | $60 / mo | 3× usage credits vs Pro — for heavy daily users |
| Ultra | $200 / mo | 20× usage credits vs Pro, priority access to new features |
| Teams | $40 / seat / mo | Pro-equivalent AI + shared rules, centralized billing, SAML/OIDC SSO, RBAC, usage analytics |
| Enterprise | Custom | Pooled usage, invoice billing, SCIM, audit logs, granular model controls, priority support |
How usage credits work: Since June 2025, paid plans include a monthly credit pool equal to the plan price ($20 for Pro). Choosing Auto mode is unlimited — Cursor picks the best available model at no cost against your credits. Manually selecting a specific frontier model (Claude Sonnet, GPT-5, Gemini Pro) draws tokens from your credit balance at that model's per-token rate. Heavy users on manual model selection burn through credits faster.
Checklist: Do You Understand This?
- Cursor is a VS Code fork rebuilt with AI deeply integrated — not an extension
- Full codebase awareness means Cursor indexes your project so AI understands the whole thing
- Cursor 3 (Apr 2026) shifted from file-editing to an agent-first parallel workspace
- Auto mode is unlimited on paid plans — manual model selection draws from usage credits
- Cursor suits large-codebase, multi-file work; for simpler inline help, Copilot is cheaper