How to Set Up Voice Input for Cursor

Voice Input Pad is a small mechanical keypad with a microphone, three labelled keys and one knob. It can provide a dedicated hold control for Cursor voice input, but the correct setup depends on the Cursor surface and version.

Does Cursor have built-in voice input?

Yes. The earlier claim that Cursor lacks voice input is outdated. Cursor introduced voice mode before 2026, and its current Agents Window documentation describes upgraded batch speech-to-text with a waveform, timer, cancel and confirm controls.

In that interface, press and hold Control-M to record, then release when finished. Because features can differ between the Agents Window, main editor chat, CLI and web surfaces, confirm the voice control in the exact place you use rather than assuming one release applies everywhere.

When does an external dictation tool still help?

System-wide dictation remains useful for editor text, terminal prompts or Cursor surfaces that do not expose the same built-in control. It inserts text at the focused field through its own hotkey. That path has different privacy, processing and auto-submission behavior from Cursor's integrated recorder.

Choose one route for a given key. Starting built-in recording and an external tool together can produce two transcripts or conflicting microphone access.

How do you trigger voice with hardware?

Make the keypad emit the documented Cursor chord or the selected external tool shortcut. Observe the key-down and key-up events, because the built-in Agents Window behavior is hold-to-speak. Test with a disposable agent conversation and use the visible waveform to confirm recording state.

Keep cancel available through the pointer or keyboard. Do not map confirm until several recordings have shown where the transcript appears and whether release submits, stages or merely stops capture in the installed version.

What breaks when focus is in the wrong pane?

A global dictation tool types into whatever accepts text, while Cursor's built-in command can choose a surface based on its active interface. In a tiled Agents Window, several conversations can be visible. Click the intended prompt and verify its cursor before pressing the hardware key.

The Voice Input Pad includes tick and cross labels that could later mirror confirm and cancel, but start with only the mic control. Record the Cursor version and surface beside the binding, and retest after interface updates. Keep notes on whether release merely ends recording or also confirms the transcript. Cursor's visible controls are the authority for the current state. A hardware label copied from an older version can encourage the wrong action after a voice-interface or shortcut update. Repeat the disposable conversation test before returning to a real project.