⌥+⌃AltPlusCtrl

Cursor AI Inline Editing Shortcuts

Cursor's core differentiator from stock VS Code is inline AI editing — describing a change in natural language directly within the editor and having it applied in place — and this cluster of shortcuts covers triggering, accepting, navigating, and rejecting that AI-generated code.

ActionWindowsMacDescription
Open inline AI edit (Cmd/Ctrl+K)Ctrl+KCmd+KOpens an inline AI prompt directly at the cursor or over a selection, letting you describe an edit in natural language and have the AI generate or modify code in place, Cursor's signature feature distinguishing it from a stock VS Code installation.
Accept AI suggestion (Tab)TabTabCommits whatever autocomplete suggestion is currently on screen — potentially several lines or a chain of related edits across the file thanks to Cursor's 'Tab, tab, tab' prediction flow, well past what a traditional single-token autocomplete ever offers.
Reject AI suggestionEscEscDismisses the currently shown AI suggestion without applying it, returning focus to normal typing, the safety valve that keeps aggressive multi-line predictions from ever being applied against your will.
Generate new code block from promptCtrl+K (on empty line)Cmd+KUsing the same inline-edit shortcut on an empty line or new file prompts the AI to generate an entirely new code block from a natural-language description rather than editing existing code.
Accept all changes in a multi-file editCtrl+EnterCmd+EnterWhen an AI-proposed change spans several files at once, this applies every proposed edit across all of them in a single action rather than reviewing and accepting file by file, meant for larger, already-reviewed multi-file refactors.
Jump to next predicted edit locationCtrl+Shift+EnterCmd+Shift+EnterMoves the cursor directly to the next location Cursor's model predicts also needs editing as a consequence of the change just made, part of the chained multi-location prediction flow rather than a simple single-suggestion accept.
Cmd/Ctrl+K is the foundational AI shortcut, opening an inline prompt box either at your cursor position or anchored to a current selection. Used on a selection, it lets you describe a transformation ('refactor this to use async/await,' 'add error handling') and the AI generates a diff-style proposed edit for you to review before accepting. Used on an empty line or new file, the same shortcut instead prompts for generating entirely new code from a description, functioning as a natural-language-driven code generation entry point rather than an edit to existing content. Tab-based acceptance of AI suggestions extends considerably beyond traditional single-token autocomplete — Cursor's underlying model can predict multi-line completions or propose several related, coordinated edits across nearby code based on the pattern of what you're actively writing, and pressing Tab repeatedly can chain through several such predictions in sequence as each accepted suggestion surfaces context for the next one, a workflow some users refer to informally as Cursor's 'Tab, tab, tab' flow. Ctrl/Cmd+Shift+Enter specifically jumps the cursor to wherever the model thinks the next related edit belongs, which matters because those follow-on edits aren't always on the very next line — a rename in one function, for instance, often needs a matching update several functions or even files away, and this shortcut gets you there without manually scrolling or searching. Rejecting a suggestion (Esc) simply dismisses it without applying any change, returning you to normal typing — an important safety valve since AI suggestions, however capable, aren't guaranteed correct, and reviewing before accepting (rather than reflexively tabbing through every suggestion) remains an important habit, especially for more consequential multi-line or multi-location proposed edits. For larger changes that touch several files at once, Ctrl/Cmd+Enter accepts every proposed edit across the entire changeset in one action rather than stepping through file by file — genuinely useful once you've already reviewed the overall diff and trust its scope, but worth using more cautiously on unfamiliar code or anything without solid test coverage, since it trades granular review for speed. A practical workflow pattern many Cursor users develop: using inline edit (Cmd/Ctrl+K) for smaller, well-scoped, single-location changes where you already know roughly what you want, reserving the broader Chat panel for larger, more exploratory requests where you want to discuss the approach with the AI before committing to a specific implementation. Knowing which entry point matches the size and certainty of a given task is arguably a bigger productivity lever than memorizing the key combinations themselves. One more nuance worth understanding: the diff-review step before an inline edit is applied is not just a UI courtesy, it is the primary mechanism by which experienced Cursor users catch a model that has misread the intent of a request. A prompt like 'clean this up' is inherently ambiguous, and a five-second scan of the proposed diff before pressing Tab or Accept is consistently cheaper than discovering a subtly wrong change three commits later during a code review. Developers switching between several projects with very different coding conventions often notice that Cursor's suggestions adapt fairly quickly to a project's existing style once it has seen enough of the surrounding code, meaning the very first few inline edits in a freshly opened project tend to be less precisely styled than ones requested later in the same session, after the model has effectively absorbed more of that project's particular formatting and naming conventions from the files already open and edited.