AI-first IDE forked from VS Code, by Anysphere. Tab completion is the headline (deep multi-file edit predictions), plus Composer for repo-wide changes, agent mode, and chat that understands the whole codebase via embeddings. Bring-your-own model (Claude, GPT-5, Gemini). Privacy mode keeps your code out of training.
Cursor is an AI-first IDE forked from VS Code. AI capabilities are implemented in the editor core rather than as extensions, allowing deeper integration with editor state than plugin-based approaches permit.
Composer/Agent mode is the primary agentic interface: it reads across multiple files, proposes coordinated multi-file edits, executes shell commands, and iterates on test failures within a conversational loop. Codebase indexing builds a semantic vector index of the repository at startup, making symbol and file references available to the agent without manual context insertion. Tab completion runs on Cursor's own low-latency model independently from the chat and agent models.
Model selection covers Claude Sonnet, GPT-4o, Gemini, and Cursor's own models; premium requests route to upstream Anthropic and OpenAI APIs. Privacy Mode disables code storage and excludes code from training; the product is SOC 2 Type II certified. Rules files and AGENTS.md encode project-level behavioral instructions for the agent. Pricing: Free (limited), Pro ($20/month), Business ($40/user/month).
Key Features
Composer/Agent mode: reads across multiple files, proposes multi-file edits, runs terminal commands, and iterates on test failures inside a single conversational loop without requiring manual context-pasting between files
Codebase semantic indexing: builds a vector index of the repository so prompts can reference symbols, function signatures, and file relationships by name — relevant context is retrieved automatically rather than pasted manually
Tab completion on Cursor's own model: low-latency inline completions trained separately from the chat model, fast enough for keystroke-level use without perceptible delay
Multi-model selection: switch between Claude Sonnet, GPT-4o, Gemini, and Cursor's own models per session — premium requests route to upstream APIs, standard requests use Cursor-hosted models
Privacy Mode: disables code storage and training use — required option for teams with source code confidentiality requirements; SOC 2 Type II certified
Rules and AGENTS.md: repo-level instructions that shape agent behavior — enforce coding standards, file naming conventions, or test requirements without prompting them manually each session
