Voice coding in Cursor: what Cursor's built-in voice input actually does, where it falls short, and how to dictate anywhere in the editor

Update to Cursor 2.0 or later, open the chat/agent panel, and click the microphone icon next to the input box (or check Settings > Labs > Voice if it's not visible). Hold the mic or the Cmd+Shift+Space shortcut, speak, and release to transcribe into that box. For voice typing anywhere else in Cursor — the code editor, Cmd+K, comments, the terminal — you need a system-wide tool like FluidVox instead.

Cursor's own dictation lives only in its AI chat/agent input: click the mic icon, speak, release. It doesn't reach the code editor, Cmd+K, or the terminal. FluidVox works everywhere in Cursor because it types through the OS's Accessibility API rather than a feature built into one panel — hold Fn (Mac) or Ctrl+Shift+Space (Windows), speak, and text lands wherever your cursor is focused.

For the built-in option, click into Cursor's chat or agent prompt and use the microphone button that Cursor 2.0 added. For voice input into the actual code editor, inline comments, or Cmd+K — where Cursor's native feature doesn't reach — FluidVox adds a hold-to-talk hotkey that works in any focused text field across the whole app.

AI dictation combines speech-to-text with a language model that cleans up filler words, adds punctuation, and reformats your speech for the context you're writing in. FluidVox does this with 6 selectable styles — its Code style formats spoken descriptions into clean comments or structured prompts, which is exactly the kind of text you write most in Cursor.

Yes — Cursor's built-in voice input for the chat/agent box ships free with the editor itself, no separate charge. Coverage is narrow, though: it only fills that one input, and early users report it can't handle special characters, non-Latin scripts, or voice-based editing. FluidVox is a separate app with a 14-day free trial, then $2.99/month or $39 one-time for on-device dictation everywhere in Cursor.

Cursor's native voice input isn't a separate download — update Cursor itself to version 2.0 or newer and the microphone icon appears in the chat panel. If you want voice input in the code editor, Cmd+K, and terminal too, download FluidVox for macOS or Windows separately; it installs once and works across Cursor and every other app.

Common causes reported by users: microphone permission wasn't granted, the default Cmd+Shift+Space shortcut collides with an existing VS Code keybinding, or a custom submit keyword was used instead of the default (which frequently sends empty messages). It's also limited by design — no hotword activation, no voice editing, and no support for special characters like "@" or non-Latin scripts. FluidVox avoids the keybinding conflict by using Fn as its default trigger and supports 99 on-device languages.

Cursor 2.0, released in late October 2025, added a native microphone button to the chat/agent input: hold to speak, release to transcribe, and the text drops into that prompt box. It's push-to-talk only, has no hotword or wake word, can't edit or delete what you said, and doesn't extend to the code editor, Cmd+K, or terminal. Cursor's iOS app separately added voice input for launching cloud agents in version 3.9 (June 2026).

Anthropic's Claude.ai has its own voice mode for conversational chat, but that's a separate product from Cursor — when you select a Claude model inside Cursor, dictation into Cursor's interface is still handled by Cursor's own voice input (chat box only) or a system-wide tool like FluidVox, not by Claude itself.

Yes. Wispr Flow is a system-wide dictation app in the same category as FluidVox, and both work in Cursor's editor, chat, and Cmd+K since neither depends on Cursor's own voice feature. The practical differences come down to pricing, on-device vs. cloud processing, and platform support — see our honest comparison at FluidVox vs. Wispr Flow.

Yes — that's the main gap FluidVox fills. It injects text through macOS's Accessibility API or the Windows equivalent, so it works in any text field Cursor exposes: the editor for comments and docstrings, Cmd+K's inline prompt, the AI chat/agent panel, and the integrated terminal.

The Code style is built for this: it formats spoken descriptions into clean, well-structured text suited to prompts, comments, and commit messages, while preserving technical terms rather than "cleaning" them into prose. FluidVox's per-app style matching can apply Code automatically whenever Cursor is the focused app, so you don't have to switch styles manually.

Yes. FluidVox's personal dictionary lets you add project-specific terms, API names, and variable names so they're always transcribed correctly, and it auto-learns from your manual corrections over time (with an AI validation step before anything is added automatically).

FluidVox offers a 14-day free trial with full access and no credit card required. After that, the Local plan is $2.99/month or $39 one-time for fully on-device dictation, with Pro plans from $10/month adding cloud AI and the full Vox Agent.