How do you use voice dictation in Cursor? You have two options. Cursor’s built-in Voice Mode (Cmd+Shift+Space) dictates prompts into the Agent. A system-wide dictation app such as Weesper Neon Flow types into every field instead: Agent, Cmd+K inline edit, terminal, comments and commit messages. If you use Weesper, move its default Shift+Tab hotkey, because Cursor uses Shift+Tab to switch Agent modes.
AI code editors have changed what developers write. In Cursor, much of that text is now natural-language instructions for the Agent, edit requests, bug reports and review notes. Speaking can be quicker than typing these longer prompts. This guide covers voice dictation in Cursor, how to dictate prompts to Cursor without shortcut conflicts, and when offline dictation in Cursor helps protect sensitive audio.
Why dictate prompts in Cursor at all?
Because good prompts are long, and long prompts are tiring to type. Speaking lets you give the Agent the context it needs instead of the shortest instruction you can be bothered to write.
Voice dictation in Cursor means speaking into a microphone and having your words appear as text in Cursor’s input fields, usually the Agent prompt, so the AI acts on them as if you had typed them.
Voice works best for natural-language tasks:
- Explaining a feature or a bug before asking the Agent to act
- Describing the constraints of a refactor: what must not change, which tests must pass
- Writing review comments, commit messages and pull request descriptions
- Thinking aloud in Plan mode before any code is written
It works badly for exact syntax. You will not want to dictate a regular expression or a nested type signature; type those, and let your voice handle the explanation around them. Our vibe coding guide covers the broader developer workflow; this article focuses on Cursor itself.
What does Cursor’s built-in Voice Mode do?
Cursor 2.0, released on 29 October 2025, added Voice Mode. The changelog describes it in one line: “Control Agent with your voice using built-in speech-to-text conversion.”
How to use it:
- Open the Agent side panel with Cmd+I or Cmd+L. Cursor’s documentation lists Mac shortcuts; on Windows, check Keyboard Shortcuts in Cursor’s settings.
- Press Cmd+Shift+Space to toggle Voice Mode, or click the microphone in the Agent input.
- Speak your instruction, then review the transcript before sending.
Cursor also lets you define custom submit keywords in its settings, so a spoken word such as “send” can trigger the Agent without touching the keyboard.
Voice Mode is already in Cursor and needs no extra software, which makes it the right first step. Its limits come from its design: it is built around the Agent prompt. It does not help when you are writing in the terminal, in a Cmd+K inline edit, in a GitHub pull request in your browser, or in Slack while you discuss the change. Cursor’s public changelog and data-use page also do not say where Voice Mode processes your audio, so check with Cursor if that matters to your team.
Built-in Voice Mode vs a system-wide dictation app
A system-wide dictation app works at the operating-system level: you hold a hotkey, speak, and the text is typed wherever your text cursor is, in any application. That changes what voice can do in a Cursor workflow.
| Cursor Voice Mode | System-wide app (e.g. Weesper Neon Flow) | |
|---|---|---|
| Agent prompt | ✅ | ✅ |
| Cmd+K inline edit | Not its focus | ✅ |
| Terminal prompt bar | Not its focus | ✅ |
| Code comments, commit messages | Not its focus | ✅ |
| Browser, Slack, email, docs | ❌ | ✅ |
| Spoken submit keyword | ✅ | ❌ (you press Enter) |
| Custom vocabulary | Not documented | ✅ Custom dictionary |
| Where audio is transcribed | Not documented | On your computer |
| Extra install | None | Desktop app (macOS or Windows) |
Use Voice Mode if you mostly talk to the Agent and want zero setup.
Use a system-wide app if you want one habit across Cursor, the terminal, your browser and chat tools, or if you need technical vocabulary and local transcription. Many developers end up using both: Voice Mode for quick Agent commands, a system-wide app for everything else. To try the system-wide option, download Weesper Neon Flow for macOS or Windows.
How to set up Weesper Neon Flow for Cursor
The setup takes about five minutes. The only Cursor-specific step is the hotkey.
1. Move the dictation hotkey off Shift+Tab
This is the step most people miss. Cursor uses Shift+Tab to rotate between Agent modes from the chat input, for example from Agent to Plan. Weesper Neon Flow uses Shift+Tab as its default dictation hotkey on both macOS and Windows.
Give each shortcut one job:
- Open Weesper Neon Flow Settings → Micro & Hotkey.
- Click the hotkey field and press a new combination. The Fn key also works as a trigger.
- Test it in Cursor’s Agent input: hold, speak, release.
Weesper uses hold-to-speak: recording runs while the key is held and stops when you release it. The hotkey configuration guide covers conflicts in detail.
2. Teach it your stack’s vocabulary
Speech models know English, not your codebase. Words like “Tailwind”, “Zustand”, “pnpm”, “tRPC” or your internal service names are where most errors come from.
Add them once in Settings → Dictionary, and add auto-correct rules for words that keep coming out wrong. The custom dictionary guide explains both, and our article on custom vocabulary for technical terms shows how to build a good list.
3. Pick a literal writing style for prompts
Weesper’s Output Style presets can rephrase your dictation. For Cursor prompts, a literal style is usually better: the AI model handles messy phrasing well, and you want your exact constraints preserved. Keep a more polished style for emails and documentation.
Prompt patterns that work by voice
Spoken prompts fail in a predictable way: they ramble. A simple structure fixes most of it. Say the parts in this order:
- Goal. “Add rate limiting to the login endpoint.”
- Context. “We already use Redis for sessions, the middleware lives in the auth folder.”
- Constraints. “Don’t change the public API, keep the existing tests passing, no new dependencies.”
- Output. “Show me the plan first, then wait.”
That last line pairs well with Plan mode: you switch with Shift+Tab, which is one more reason to keep that key free.
A few habits help:
- Say file names plainly. “auth middleware dot ts” is clear enough; the Agent resolves the path from your project.
- Dictate review feedback in full sentences. “This loop re-queries the database on every iteration, batch it” gives the Agent a reason, not only a command.
- Use Cmd+K for local edits. Select a function, press Cmd+K, then dictate: “convert this to async and add error handling.”
- Read before you send. One glance catches the odd misheard word before it reaches the model.
If voice coding interests you for health reasons, our guide to voice dictation for developers with RSI covers ergonomics and hands-free setups.
Is dictating into Cursor private?
Partly, and it helps to separate audio from text.
Audio is your voice recording. With an offline dictation app, it is transcribed on your own computer and never uploaded. With a cloud dictation service, the recording goes to that vendor’s servers before you see a single word. Voice recordings can also capture things you did not mean to share: a colleague’s name, a client on a call, the room around you.
Text is the prompt itself. Whatever tool you use to produce it, Cursor sends prompts through its own backend to the AI model; its data-use page states that requests go through Cursor’s backend even with your own API key. Privacy Mode controls what happens next: Cursor says that data is then not used for training and that it has zero-data-retention agreements with its model providers.
So the honest picture is this: offline dictation removes one data flow, your voice, and leaves the prompt under Cursor’s policies. For teams working on proprietary code, that is still worth having, because it avoids adding a second cloud vendor that receives recordings of your work. Our article on why offline dictation protects your privacy goes further.
Getting started
Start with what costs nothing. Turn on Cursor’s Voice Mode, dictate your next three Agent prompts, and notice how much more context you give when you speak.
When you find yourself wanting the same habit in the terminal, in Cmd+K, in your pull requests and in Slack, add a system-wide tool. Weesper Neon Flow runs Whisper locally on macOS and Windows, types into any app, and learns your stack through its custom dictionary. Just move its hotkey off Shift+Tab first.
Ready to talk to your AI editor? Try Weesper Neon Flow free for 15 days, no credit card required, and dictate your next Cursor prompt instead of typing it.