5 tools · 35 shortcuts · each section verified against the vendor’s own documentation. Jump to: Grok (9) · NotebookLM (6) · Cline (9) · Bolt.new (5) · Replit AI (6)
Chat-style AI tools expose very few keyboard shortcuts on purpose: the interface is a prompt box, a response and a handful of panels, so the whole keymap of each product fits in a dozen rows. Rather than give each tool a page with six rows, this page collects the five assistants and coding agents that ship the smallest keymaps. Each section keeps its own source link and verification date, and the shortcuts are the same rows the search box and the practice game use. Tools with a fuller keymap — ChatGPT, Claude, Gemini, Perplexity, GitHub Copilot, Cursor, Windsurf — have their own pages.
Grok's web app has only a few keyboard shortcuts — send, newline, new conversation, and a panel that lists them — and the rest of what the table calls "modes" are settings chosen from the composer rather than keys. This page lists both so the mode names are searchable; the notes explain what each mode changes.
| Shortcut | Action |
|---|---|
| Enter | Send message |
| Shift + Enter | New line |
| Ctrl + K | New conversation |
| Ctrl + / | Keyboard shortcuts |
| Shortcut | Action |
|---|---|
| Default mode | Standard chat |
| Think mode | Extended reasoning |
| DeepSearch mode | Web research |
| Image generation | Generate images |
| Voice mode | Voice chat |
Sending is Enter; a line break inside the message is Shift + Enter, as in ChatGPT and Claude. Ctrl + K starts a new conversation from anywhere, and Ctrl + / opens the shortcut panel, which is the authoritative list for the version you are on and the place to check when a key stops working after an update. On macOS these use Cmd.
Default mode answers directly with the standard model. Think mode spends longer reasoning before answering and is the one for maths, code and multi-step questions where a wrong first guess is costly. DeepSearch mode runs a web research pass and cites sources, which suits current-events questions and anything where the answer should be checked against the web. Image generation produces images from a prompt and Voice mode switches to spoken conversation in the mobile apps.
Think mode is slower and worth it only when the question is hard; DeepSearch adds latency and network access and is the wrong choice for pure writing tasks. Modes are picked per message from the composer, so switching for one question and back is normal use rather than a settings change.
NotebookLM has no keyboard shortcuts of its own — the entries in this table are the six actions the interface offers, listed so the vocabulary is searchable. What follows explains what each action does and how they fit together, because the product's value is in the sequence: load sources, ask grounded questions, keep the answers.
| Shortcut | Action |
|---|---|
| Add source | Upload document |
| Ask question | Source-based Q&A |
| Create note | Save note |
| Audio Overview | Generate podcast |
| Generate FAQ | Auto FAQ |
| Summarize | Document summary |
Add source uploads PDFs, Google Docs, slides, web pages, YouTube transcripts or pasted text into a notebook, and everything else operates only on those sources — NotebookLM does not answer from general knowledge, which is the point. Up to several dozen sources fit in one notebook, and the source panel lets you toggle which ones a question should consider.
Ask question runs a query against the selected sources and returns an answer with numbered citations that open the exact passage, so a claim can be checked in one click. Summarize produces a document summary and Generate FAQ a set of questions and answers drawn from the sources, both useful as a first pass over unfamiliar material. Create note saves an answer or a summary into the notebook's notes, where it survives new questions and can be turned back into a source.
Audio Overview generates a two-host podcast-style conversation about the sources, several minutes long, which people use to absorb a report while commuting; newer versions let you steer the focus with instructions and join the conversation. It is a synthesis, not a reading, so check anything specific against the citations before repeating it.
Cline is an autonomous coding agent that runs as a VS Code extension, and its keyboard surface is the extension's panel plus the conventions inside it: approve, cancel, mention context, slash commands. The table writes the VS Code entry points as command-palette sequences because that is how Cline documents them. On macOS use Cmd for Ctrl.
| Shortcut | Action | Description |
|---|---|---|
| Ctrl + Shift + P → Cline | Open Cline | Launch Cline from command palette. |
| Ctrl + Shift + A | Cline panel | Toggle Cline sidebar panel. |
| Enter | Send message | Send the prompt. |
| Tab | Approve changes | Approve suggested code changes. |
| Esc | Cancel | Cancel current operation. |
| @ | Mention context | Reference files, folders, or URLs as context in your message. |
| / | Slash commands | Trigger commands like /newtask directly from the chat input. |
| Shift + Enter | New line | Insert a line break in the chat input without sending. |
| Ctrl + Shift + P → Cline: New Task | New task | Start a fresh Cline task from the command palette. |
Ctrl + Shift + P → Cline — the command palette with "Cline" typed — opens the extension, and Ctrl + Shift + P → Cline: New Task starts a fresh task; Ctrl + Shift + A focuses the Cline panel directly once it is set up. A task is a conversation with a goal, and Cline works through it in steps, proposing file edits and terminal commands as it goes.
Enter sends a message and Shift + Enter adds a line to it. When Cline proposes an edit or a command, Tab approves it and Esc cancels the current step — the loop that keeps a human in control of what touches the filesystem, and the reason to read each diff before pressing Tab. Auto-approve settings can relax this for safe categories such as reading files.
Typing @ opens the context picker: @file, @folder, @problems for the current diagnostics, @terminal for recent output and a URL for documentation. / opens slash commands such as /newtask and /smol for condensing the conversation when the context grows large. Cline's plan/act mode switch and model settings are buttons in the panel.
Bolt.new builds and runs a web app from a prompt in the browser, and its keyboard surface is the small set that the prompt box and the built-in code editor share: send, newline, undo, save, command palette. Everything else — previews, deploys, file operations — is a button in the current interface. The table lists what is bound; the notes explain what each key acts on.
| Shortcut | Action |
|---|---|
| Enter | Send prompt |
| Shift + Enter | New line |
| Ctrl + Z | Undo |
| Ctrl + S | Save |
| Ctrl + Shift + P | Command palette |
Enter sends the prompt and Shift + Enter adds a line, so multi-step instructions can be written as a list before sending — which produces better results than a single long sentence. The response edits the project files directly and restarts the preview; there is no separate apply step.
The editor pane is a Monaco-based editor, so Ctrl + Z undoes edits (including the ones Bolt made) and Ctrl + S saves the current file, triggering a rebuild of the preview. Editor actions such as formatting or changing the language mode sit behind Ctrl + Shift + P, the command palette. Edits you make by hand are seen by the next prompt, so fixing a small thing manually and then asking for the next feature is the normal loop.
Rolling back to a previous version, forking a project, opening the terminal and deploying are all mouse actions today. Because Bolt changes frequently, treat this table as a snapshot checked on the date below and report anything that no longer holds.
Replit's browser IDE has a compact keyboard layer, and its AI features — Ghostwriter completions and the Agent — reuse the conventions of the other AI editors: Tab accepts, Esc dismisses, a key opens chat. The table lists the six that matter; everything else in Replit is a button or the standard Monaco editor keys.
| Shortcut | Action |
|---|---|
| Ctrl + I | AI Chat |
| Tab | Accept suggestion |
| Esc | Dismiss |
| Ctrl + Enter | Run |
| Ctrl + S | Save |
| Ctrl + Shift + S | Fork |
Ctrl + I opens AI chat with the current file as context; the Agent, which builds and edits whole projects from a prompt, is a separate panel opened from the sidebar. Inline completions appear as ghost text: Tab accepts and Esc dismisses, the same as Copilot and Cursor, so a habit from any of them carries over. On macOS the Ctrl combinations are Cmd.
The Repl runs on Ctrl + Enter, executing whatever run command .replit configures, and shows output in the console; Ctrl + S saves, although Replit autosaves continuously and the key mostly reassures. Ctrl + Shift + S forks the Repl into your own account, which is the way to experiment on someone else's project or to snapshot your own before a risky change.
The editor is Monaco, so VS Code keys for multi-cursor, find and folding work inside a file. Deployments, secrets, the shell and the database are panels; Replit's command palette (Ctrl + K by default) reaches them by name when the sidebar is hidden.
Each of the five products documents between five and nine shortcuts, which is the whole keymap for a prompt-box interface. One page keeps them easy to compare; each section still carries its own source and check date.
Enter sends and Shift + Enter inserts a line break in every prompt box here. Beyond that they differ: Grok and NotebookLM use browser-style keys for new threads and panels, while Cline, Bolt.new and Replit AI sit inside an editor and reuse its command palette and terminal keys.
Open the Shortcut Speedrun game and pick the individual tool — the game keeps each product as its own platform even though the reference pages are combined here.