Cursor is the AI-native IDE. Power users want voice input for Composer, the chat sidebar, and inline edits. No Mac voice plugin? No problem. StarWhisper sits in the Windows tray and types into any focused field, including Cursor.
Cursor changed how developers write code. Voice changes how fast you talk to it.
Composer needs long, specific prompts to do its best work. Typing a 400 word task description slows the iteration cycle. Dictating one keeps you in the loop while Cursor produces the diff.
StarWhisper is an OS-level dictation layer, not a Cursor extension. That sounds like a limitation; it is actually the point. Cursor ships breaking changes often. A plugin breaks. A keyboard layer does not.
The Cursor workflow has more prose surfaces than most developers realize
Composer is at its best when you tell it the why, the constraints, and the files in scope. That is a paragraph or two. Dictating that paragraph keeps your hands free while you read the code you are about to refactor.
Asking Cursor's chat to explain a function, suggest a refactor, or debug a stack trace is a conversation, not a single shot. Voice keeps the back and forth fast, especially when each follow up is two or three sentences of context.
The inline edit popup at cmd+K is the fastest path from intent to diff. Click into the popup, hold the StarWhisper hotkey, dictate the change, release, hit Enter. The whole loop takes seconds for a one sentence instruction.
Cursor's integrated terminal is where many developers now run Claude Code, codex, or Aider. The prompt input for each agent is just stdin. StarWhisper dictates straight into the terminal pane, same as any other Windows text field.
Cursor accepts dictation into the editor surface as well. Position your cursor on the line above a function, dictate the explanation, edit a few words, save. Comments stop being the part of the codebase that nobody writes.
The Cursor source control panel, the integrated terminal git commit prompt, and the GitHub web PR form all accept dictation. A two paragraph commit message goes from 90 seconds of typing to about 15 seconds of speaking.
Cursor is structured around a few text input surfaces that you interact with all day: the chat sidebar, the cmd+K inline edit popup, the Composer pane, and the integrated terminal. Every one of those surfaces accepts prose, and every one of them rewards longer, more specific prose. StarWhisper is a Windows desktop app that adds dictation to any of them by listening for a hotkey and pasting transcribed text into the focused field.
The combination matters because Cursor moves the developer bottleneck. In a traditional editor, typing speed barely mattered, because the constraint was thinking, designing, and debugging. In Cursor, you spend more time describing what you want and less time hand-writing every line. The proportion of your day spent typing prose goes up. Dictation cuts that prose time roughly in half, since the average developer types around 60 words per minute and speaks around 130 to 160 words per minute.
There is no Cursor extension to install, no API key to manage, and no settings inside Cursor to configure. The integration happens at the Windows keyboard layer, which means it survives every Cursor update without any maintenance on your side. If a future Cursor release rearranges the sidebar or renames Composer, the same hotkey still dictates into whatever the new field is called.
As of mid 2026, Cursor does not include a first-party voice input feature on Windows. The Cursor team has discussed voice in community channels and on Twitter, and there are sporadic experiments on the Mac side that wire up the system Dictation key or third-party menubar Whisper apps. Nothing of that exists as a built-in Cursor feature on the Windows build.
The Windows path that actually works today is a desktop dictation layer. StarWhisper does exactly this: it captures audio when you hold a hotkey, runs it through OpenAI Whisper locally, and pastes the resulting text wherever your cursor is. Cursor sees the pasted text the same way it would see text from your keyboard. There is no plugin to install, no extension marketplace to navigate, and no Cursor API to integrate with.
This architecture is also why the same setup works in Claude Code, codex, Aider, Windsurf, VS Code, Zed, JetBrains IDEs, and any chat box in your browser. The layer that does the dictation is below the application layer, so every application gets it for free.
The most common Cursor workflow looks like this: you have a feature to build or a bug to fix, you open Composer or chat, you describe what you want, Cursor produces a diff, you review the diff, you ask for an adjustment, repeat. The typing parts of that loop are the describe step and each adjustment.
With voice, the describe step changes character. Instead of writing a terse one line prompt because the long one is annoying to type, you naturally talk through the context: which files are in scope, what the current behavior is, what the desired behavior is, which edge cases matter. The prompt gets longer. The quality of the diff Cursor produces gets better. The number of adjustment rounds goes down.
Many developers report that the second order effect, fewer adjustment rounds per task, is larger than the first order effect of faster typing. Specifying the problem clearly the first time is more efficient than three rounds of underspecification followed by clarification. Voice removes the typing tax that pushes developers toward underspecification in the first place.
The full setup takes about three minutes. Install StarWhisper from the download page or the Microsoft Store listing. The installer auto-detects whether you have an NVIDIA GPU and picks the matching pack: CPU, CUDA 11, or CUDA 12. First launch downloads the Whisper model files. After that the app lives in your system tray and listens for the hotkey.
For Cursor specifically, choose a hotkey that does not collide with anything Cursor maps. The right-side modifier keys, especially Right Ctrl or Right Alt, are good defaults because Cursor's own shortcuts almost always use the left-side modifiers. Mouse side buttons are popular for developers who already use a programmable mouse. Some developers wire up a USB foot pedal because the hands-on-keyboard ergonomics are nicer than any keyboard chord.
Once the hotkey is set, the interaction is: click into a Cursor text field (sidebar, popup, Composer, terminal), hold the hotkey, dictate the prompt, release. The text auto-pastes into the focused field. Press Enter or click the submit button in Cursor exactly as you would for typed input. There is no special voice command grammar to learn and nothing in Cursor's interface that changes.
The prompts you type into Cursor describe your codebase. If the codebase is proprietary, then the prose describing it is proprietary too. Sending that prose to a third-party cloud transcription service raises the same security review questions that come up around pasting code into a public LLM: where does the audio go, who has access, how long is it retained, can the vendor train on it.
StarWhisper runs Whisper locally on your CPU or GPU. Your audio never leaves the machine. There is no transcription cloud, no audio retention period, no vendor relationship to audit on this dimension. If your laptop is offline, dictation still works. That is structurally easier to defend in a security review than any "we delete after N days" cloud posture, because there is nothing to delete.
Cloud Mode, which sends audio to the OpenAI Whisper API for faster turnaround on weaker hardware, is opt-in and disabled by default. For dictation about proprietary code or sensitive prompts, leave it off. On a modern NVIDIA GPU the local model is fast enough that there is rarely a performance reason to enable it.
The same dictation layer works equally well across the Windows AI editor ecosystem. Switching between Cursor and VS Code with Copilot Chat, or experimenting with Windsurf, or trying Zed on Windows, does not change the dictation workflow. The hotkey works the same way in every editor because the dictation happens before the editor sees the input.
This matters because the AI editor market is moving fast. Cursor is dominant today; Windsurf has serious momentum; Zed is building toward similar territory; JetBrains has its own AI Assistant; Visual Studio added Copilot Chat. Betting on each editor shipping its own voice feature is a slow path. Using a dictation layer that works across all of them is a fast path. For a broader take on the developer use case across editors, see the voice typing for coding overview, which covers the cross-editor pattern in more depth.
Voice typing is for prose. Single line edits to a function body, quick syntax fixes, variable renames, and most refactor mechanics are still faster typed or done via Cursor's own refactor tools. The wins are real but they are in the prose surfaces: prompts, comments, commit messages, chat, PR descriptions, Slack threads explaining your design choice. If your Cursor workflow today is mostly hand-typing small edits and rarely using Composer or the chat sidebar, the upside is small.
If your Cursor workflow leans heavily on Composer for multi-file edits, on chat for design discussions, on cmd+K for inline transforms, and on terminal CLI agents for repo-wide tasks, voice typing pays back its setup time the first day. The richer the prose surface you live in, the more voice helps. For another related workflow, see how to use voice to text with ChatGPT, which covers the same pattern for browser-based chat tools.
The general developer overview covering Cursor, Claude Code, Copilot, and the rest of the stack.
How to dictate long prompts into Anthropic's CLI agent and other terminal-based coding tools.
Step by step setup walkthrough for the Cursor desktop app on Windows.
The browser based equivalent for ChatGPT, Claude, Gemini, and Perplexity prompts.